CNC Machining

The short answer CNC drilling removes the operator from the cycle. The machine positions each hole to within 0.01 mm, changes tools in under 2.5 seconds and runs drill, countersink and tap in one program. Measured throughput is 80 to 200 holes an hour against 20 to 40 by hand, and positioning scrap falls from 3 to 5 percent to under 0.5 percent. What actually changes when the operator leaves the cycle Manual drilling is a sequence of human decisions: mark the position, centre punch it, sight the drill, feed by feel, then deburr and gauge the hole. The operator is committed to the machine for the whole cycle, and every hole repeats that chain of judgement. CNC drilling replaces the chain with a program. The controller positions the table, ramps the spindle, executes peck cycles, switches tools and delivers coolant, and the operator is only involved when a part is loaded or unloaded. Read each row on its own terms. These are six different measurements, not six scores for the same scale. MeasureManual floor pressCNC machining centre Hole positionplus or minus 0.3 mm with a skilled operatorplus or minus 0.01 mm from servo axes Cycle time per hole8 to 15 seconds2 to 5 seconds Tool change30 to 60 seconds by handunder 2.5 seconds, chip to chip Operator attention100 percent of the cycle10 to 20 percent, load and unload Scrap from positioning3 to 5 percent of productionunder 0.5 percent First-part setup2 to 5 minutes15 to 45 minutes of program and fixture The last row is the honest one and it is why the process is not automatically faster. A manual press is ready in two to five minutes; a CNC machine needs a program and a fixture first. Everything above the last row is what the machine buys back once that setup is spread across a batch. Speed: where the three to five times comes from Published benchmarks put a manual machinist at 20 to 40 holes an hour once layout, centre punching, deburring and inspection are counted, and a CNC machining centre at 80 to 200 an hour depending on material and depth. That is the three to five times usually quoted, and it is a throughput figure for a whole operation rather than a spindle figure. The gap widens on parts with several hole sizes, because an automatic tool changer swaps drills in 1.3 to 2.5 seconds chip to chip where a manual change takes 30 to 60 seconds. Every bar is the same measurement, so the rows are directly comparable. The multi-spindle figure is several holes cut at once, not a faster spindle. The economics only close past a threshold, and the published one is worth knowing before the machine is bought. A CNC drill cell pays for itself above roughly 100 holes a day, or when tolerances tighten past 0.1 mm; below that, a manual press with a digital readout and a cross-slide vice is the better investment. On the machine side the entry cost runs from about USD 25,000 to well over USD 200,000 against USD 3,000 to 5,000 for a manual press, DRO and vice, and the break-even batch sits somewhere between 50 and 200 parts depending on complexity. Position and depth hold without an operator watching Position is where manual drilling loses most of its quality, not its speed. A servo-driven machine holds hole position near 0.01 mm and repeatability near 0.005 mm, and the same G-code that drills ten prototype parts drills ten thousand production parts identically. Manual drilling depends on the operator, achieves roughly 0.1 to 0.5 mm at best, and drifts as the shift goes on. Programmed depth control removes the second class of manual error, because the machine stops the drill at the commanded depth rather than at the moment the operator reacts. Two failure modes survive programming and are worth designing out. The first is drill wander: a drill tip that deflects at the start of the cut can put a hole 0.05 to 0.2 mm off position, and the usual causes are skipping the spot drill on a smooth or curved entry, excessive overhang above a length-to-diameter ratio of 5 to 1, a dull drill, or a weak fixture on a thin wall. The second is chip packing, which is the leading cause of drill breakage; peck drilling retracts the tool every one to three diameters of depth to clear the flutes, and deep holes above three diameters usually need coolant delivered through the tool at 70 to 150 bar rather than flood coolant from the outside. Combined operations: one setup replaces three or four The largest efficiency gain is not in the drilling itself but in what stops happening between operations. A manual route drills, then re-fixtures to countersink, then re-fixtures to tap, and every re-clamping introduces its own positional error. A single CNC program runs spot drill, drill, countersink and tap in one setting, so the datum never moves and the part is never re-clamped. Between each stage the part stays clamped. That is the part of the gain that survives even when the cycle time is the same. Depth-to-diameter ratio is the constraint that decides how much of this is possible in one pass. Below 3 to 1 the risk of a deep hole is low, from 3 to 1 up to 5 to 1 it is moderate, and above 5 to 1 it is genuinely high and calls for peck cycles and reduced speed regardless of material, because deep holes are limited by chip evacuation rather than by spindle capability. Where CNC drilling is the wrong answer One-off and repair work. A 15 to 45 minute setup against a 2 to 5 minute manual setup means the machine loses on single parts and short runs, however fast each hole is. Low daily volume with no hole variety. Below roughly 100 holes a day of the same size and material, the payback period on a drill cell becomes longer than most buyers will wait. Loose-tolerance holes. If the drawing allows plus or minus 0.5 mm on a clearance hole, paying for servo positioning buys nothing. Very deep small holes in hard material. Drills below 2 mm in stainless or titanium are fragile and break more often; a broken tool in a deep hole can scrap the part and the fixture. Unstable geometry. A part whose hole pattern is still moving is expensive to iterate in a program and a fixture. A supplier that quotes without asking. A drilling quote that arrives without questions about hole size, depth, material and thread callout is a price, not a process plan. How to specify a drilling job so the gain is real Send the STEP model and a toleranced drawing, the material and the stock form, a hole list with diameter, depth, thread callout and tolerance for each feature, the annual quantity, and the two or three holes whose position actually matters. Those five items let an engineer choose the peck cycle, the coolant strategy and the tool sequence before price is discussed. A plate with twenty-four holes of four sizes and a 10,000-a-year forecast is a CNC job; the same plate in a batch of five is a manual job with a fixture. See CNC machining for the process, sheet metal fabrication for the plate itself and surface finishing for deburring and coating afterwards. Send a hole list and get a cycle plan Scope and sources. Throughput, tolerance, cycle-time, setup and break-even figures come from a drilling machine selection guide (manual position plus or minus 0.3 mm against CNC plus or minus 0.01 mm, manual cycle 8 to 15 seconds against CNC 2 to 5 seconds, manual setup 2 to 5 minutes against CNC 15 to 45 minutes, operator attention 100 percent against 10 to 20 percent, machine cost USD 25,000 to 200,000 against USD 3,000 to 5,000, break-even 50 to 200 parts, payback above about 100 holes a day or tolerances under 0.1 mm, rapid traverse 40 to 60 metres a minute, chip-to-chip 1.3 to 2.5 seconds) and from a CNC drilling capability note (position 0.005 to 0.01 mm against 0.1 to 0.3 mm manual, 200 holes an hour for a 5 mm steel hole against 40 to 60 manual, batch size deviation 0.003 mm against 0.05 mm, first-pass qualification 99.8 percent against 85 percent, drilling and tapping about a third of all CNC operations). Failure modes come from a deep-drilling technology article (diameter consistency to H7, drilled finish Ra 1.6 to 3.2 micrometres, reamed finish Ra 0.4, air change under two seconds, scrap reduction 30 to 60 percent, per-part cost down 20 to 40 percent above 500 pieces, drill wander off-position by 0.05 to 0.2 mm, overhang above 5 to 1 without support, peck retract every 1 to 3 diameters, through-tool coolant at 70 to 150 bar above three diameters) and from a drawing-symbol reference (depth-to-diameter risk low below 3 to 1, moderate from 3 to 1 to 5 to 1 and high above 5 to 1). Figures are planning ranges from published sources and not a quotation; confirm them against your material, hole sizes and quantity.

The short answer A CNC drilling service produces through and blind holes, counterbores for socket screws, countersinks for flat-head screws, spotfaces for washers, reamed bores at H7, and tapped threads in UNC, UNF and metric sizes. Twist drills reach 5 to 8 times diameter, reaming lifts the fit to 0.005 to 0.02 mm, and gun drilling reaches 100 to 150 times diameter. Every hole starts as one of seven feature types Holes are the most common machined feature and the easiest to over-specify. Each type below exists for a reason, and each one adds a tool and an operation to the price. A through hole drilled in a single pass is the cheapest hole you can buy; a reamed fit is roughly one and a half to two times the cost of the same hole drilled, because it needs a second tool and a second pass. Choose the simplest feature that meets the function. A cosmetic counterbore that nobody sees is a cost with no benefit. FeatureHow it is madeTypical toleranceTypical use Through holeTwist drill in one passplus or minus 0.1 to 0.25 mmClearance and fluid passages Blind holeDrill stopped at a commanded depthdepth plus or minus 0.5 mmThreaded or dowel holes CounterboreDrill, then end mill or counterbore toolplus or minus 0.05 to 0.1 mm on the boreSocket head cap screws CountersinkCountersink cutter, matched to the screw angleangle within 1 degreeFlat head screws and deburring SpotfaceEnd mill or face cutter on a rough surfaceflat within 0.05 mmWasher seats on castings Reamed boreDrill undersize, then ream to sizeplus or minus 0.005 to 0.02 mmBearing bores and dowel fits Gun-drilled holeDedicated gun drilling machineplus or minus 0.01 to 0.05 mmOil galleries and cooling channels Two of these are worth a second look. A spotface exists only because a cast or forged surface around a hole is not flat enough to seat a fastener reliably, so leaving it out is a real risk on a rough surface and a waste of money on a machined one. A blind hole is more than a shallow through hole: the twist drill leaves a conical bottom, so the drawing has to allow for the drill point as well as the depth, and the chips have to evacuate upward through the flutes. Counterbores, countersinks and spotfaces: the three fastener seats These three features are the most confused in incoming drawings, and they are not interchangeable. A counterbore is a flat-bottomed cylindrical recess that hides a socket head cap screw so the head sits below the surface; its diameter has to clear the head and its depth has to be at least the head height. A countersink is a conical recess matched to the angle of a flat head screw, usually 82 or 90 degrees, and an angle that is one degree out will show as a screw head that sits proud on one side. A spotface is a shallow flat cut, not a recess, used where a washer or a nut has to seat on a surface that is otherwise rough or out of flat. All three need a second tool after the drilling pass, and the cost shows up as cycle time rather than as tooling. Where several of them appear on one part, the machining centre makes repeated tool changes, so consolidating a drawing to one screw type across a face is often the cheapest single edit a designer can make. Threaded features: standards, classes and depth rules Threads are cut either by tapping, which cuts a thread into a pre-drilled hole with a tap, or by thread milling, which cuts the thread with a rotating mill on a machining centre. Tapping is faster and cheaper for standard sizes; thread milling produces a better thread and can reach closer to the bottom of a blind hole, which is why it has grown in use on deep and precision threads. The standards that appear most often are UNC and UNF from the unified series, metric coarse and fine, NPT for pipe threads and BSPT for the equivalent British standard. The control size differs from the hole size. Leave at least one and a half times the nominal diameter of extra depth beyond the last full thread so the tap has room to run out. ThreadTap drillMinimum hole depthTypical use M3 x 0.52.5 mm6.0 mm for one diameter of engagementSensors and small brackets M4 x 0.73.3 mm8.0 mmEnclosures and panels M5 x 0.84.2 mm10.0 mmMounting hardware M6 x 1.05.0 mm12.0 mmGeneral engineering 1/4-20 UNCNo. 7, 5.1 mm12.7 mmUS fasteners M8 x 1.256.8 mm16.0 mmStructural joints Three rules cover almost every threaded callout. First, specify diameter, pitch, class of fit, thread depth and total hole depth separately, because a note that says only "M6 threaded hole, 12 deep" leaves both the pitch and the thread-versus-hole distinction undefined. Second, allow at least one and a half times the nominal diameter of extra hole depth beyond the last full thread, so the tap can bottom out without damaging the thread it has just cut. Third, use the standard classes: a 2B or 6H fit is the normal choice for a fastener, and a tighter 3B class is reserved for applications that actually need it. Precision bores: reaming, boring and gun drilling When a hole has to hold a shaft, a bearing or a dowel pin, drilling alone will not do it. A drill produces slight ovality and size variation that a fit class will not tolerate, so the bore is drilled undersize and finished with a reamer, which brings the diameter to 0.005 to 0.02 mm and the finish to around Ra 0.4 to 1.6 micrometres. Larger bores are finished with a single-point boring bar instead, which can correct position and straightness as well as size and reaches much deeper. Genuinely deep, straight holes are made on a gun drilling machine, which produces the oil galleries and cooling channels that a twist drill cannot follow over that length. The smallest hole each process can make is a practical limit rather than a theoretical one. A standard twist drill works down to roughly 0.5 mm, a reamer down to about 1 mm, a gun drill to about 2 mm, and small-hole EDM to about 0.1 mm. Below 2 mm in stainless steel or titanium, drills become fragile and break more often, so peck cycles and carbide micro-drills are the normal answer and wire EDM is sometimes the cheaper route once breakage is counted. Depth-to-diameter ratios decide the process The single number that decides whether a hole is routine, awkward or specialist is its depth divided by its diameter. A standard twist drill is comfortable to 5 to 8 times diameter; past that, chip evacuation rather than the spindle becomes the limit, and peck drilling with reduced speed is the usual answer. Reaming works from 10 to 15 times diameter, boring from 20 up to 50, and gun drilling from 100 to 150 times diameter on a dedicated machine. Deep holes on a machining centre without through-tool coolant are the routine way to break a drill. Depth ratio, not absolute depth, is what makes a hole difficult. The same 8 mm hole is routine at 40 mm deep and specialist at 1 metre. Where the feature list runs out True position drives cost more than diameter. A pattern with a tight positional callout needs precision fixturing and CMM verification, and that cost is set by the tolerance, not by the number of holes. Thread depth beyond about three diameters is a risk. Tap breakage rises with depth, small threads carry more risk than large ones, and a broken tap in a finished part is often a scrap decision. Very small holes in hard material are fragile. Below 2 mm in stainless or titanium, expect higher tool breakage and slower feeds than the diameter suggests. Sharp-bottomed blind holes are not a drill feature. The 118 degree point leaves a cone, so a flat-bottomed blind hole needs a second operation or a specific callout for the flat depth. Gun drilling is not universal equipment. Where a hole needs a 100 to 1 ratio, confirm that the shop owns the machine rather than subcontracting it, because the subcontract adds days and splits accountability. Cross-drilled intersections change the pressure story. Where two holes meet inside a part, the intersection is where a leak or a burr hides, and it needs an explicit inspection step rather than a general note. How to specify holes and threads on a drawing Send the STEP model and a drawing with every hole called out by diameter, depth, feature type, thread class and positional tolerance; list the material; mark the two or three holes whose position is functional; and state the annual quantity. That is enough for an engineer to choose between drilling and reaming, to size the peck cycles and to tell you which features should be milled and which cast or left as-drilled. See CNC machining for the process, surface finishing for what happens after the hole is cut and metal stamping where a pierced hole is the cheaper route at volume. Send a drawing for a hole and thread review Scope and sources. Feature definitions, tolerances, cost factors and depth limits come from a hole design guide (through and blind holes plus or minus 0.1 to 0.25 mm with blind depth holding 0.5 mm, counterbore plus or minus 0.05 to 0.1 mm at 1.3 times cost, countersink within 1 degree at 1.2 times cost, spotface flat within 0.05 mm, reaming plus or minus 0.005 to 0.02 mm at 1.5 to 2.0 times cost, twist drill 5 to 8 times diameter, reaming 10 to 15, boring up to 50, gun drilling 100 to 150, BTA 150, minimum diameters 0.5 mm drill, 1.0 mm ream, 2.0 mm gun drill, 0.1 mm EDM, 118 degree point leaving a cone about 0.3 times diameter deep) and from an CNC design guide (standard drilled holes plus or minus 0.003 inch, maximum depth 10 times diameter for standard drills, drill point sizes in fractional letter and number series, H7 fits at zero to plus 0.001 inch, a worked thread table for No. 4-40, No. 6-32, No. 8-32, No. 10-32 and 1/4-20 UNC, and a rule of at least 1.5 times thread diameter of extra depth beyond the last full thread). Thread milling growth and the share of machining operations taken by drilling come from a CNC drilling overview (drilling and tapping a substantial share of all CNC operations, blind holes, stepped holes and angled holes as the complex cases). Drawing callout practice and the fit-class decision tree come from a drawing symbol reference (drilling for loose-tolerance clearance holes, reaming or boring once an H7 fit is called, counterbore and countersink as separate tool operations, spotface on cast surfaces, depth-to-diameter risk rising above 3 to 1 and high above 5 to 1). Figures are planning ranges from published sources and not a quotation.

The short answer A professional CNC milling shop works to ISO 2768-m by default, which is 0.1 mm on a 6 to 30 mm feature and 0.3 mm from 30 to 120 mm. Precision work tightens to 0.025 to 0.05 mm, a controlled shop reaches 0.005 to 0.01 mm on a callout, and surface finish runs from Ra 3.2 to 0.4. The default is a standard, not a number When a drawing does not put a tolerance on a dimension, the shop still has to make a decision, and ISO 2768 is the decision. Part one of the standard covers linear and angular dimensions in four classes, fine, medium, coarse and very coarse, and part two covers straightness, flatness, perpendicularity and symmetry in classes H, K and L. The block that appears on most CNC drawings is ISO 2768-mK, medium linear tolerance with K geometric tolerance, and it represents the natural economic capability of a modern machining centre rather than a compromise. Tolerance grows with the nominal size. A single callout on the title block covers every undimensioned feature on the part. Nominal sizeFine, class fMedium, class mCoarse, class c 0.5 to 3 mmplus or minus 0.05 mmplus or minus 0.10 mmplus or minus 0.20 mm Over 3 to 6 mmplus or minus 0.05 mmplus or minus 0.10 mmplus or minus 0.30 mm Over 6 to 30 mmplus or minus 0.10 mmplus or minus 0.20 mmplus or minus 0.50 mm Over 30 to 120 mmplus or minus 0.15 mmplus or minus 0.30 mmplus or minus 0.80 mm Over 120 to 400 mmplus or minus 0.20 mmplus or minus 0.50 mmplus or minus 1.20 mm Over 400 to 1000 mmplus or minus 0.30 mmplus or minus 0.80 mmplus or minus 2.00 mm The size dependence is the point that surprises many buyers. A medium-class tolerance is 0.1 mm on a 10 mm feature and 0.5 mm on a 300 mm feature, because the same absolute error on a long dimension is a much smaller fraction of it. That is also why demanding a blanket plus or minus 0.05 mm across a 250 mm part is expensive: it forces multiple finish passes, thermal stabilisation and full CMM verification on dimensions that may not matter at all. Three tolerance tiers and what each costs Above the standard classes sits a set of named machining tiers, and it is worth separating them from the ISO block because they describe what a machine and a process can hold rather than what a drawing inherits. Standard machining holds roughly 0.1 to 0.2 mm on a general feature. Precision machining, on a calibrated machine with a dedicated finishing pass, holds 0.025 to 0.05 mm and is what a bearing bore or a sealing face usually needs. Tight tolerance machining reaches 0.005 to 0.01 mm and needs thermal stabilisation, in-process touch probing and full CMM verification. Ultra precision below that is a different industry, served by diamond turning or jig grinding in a climate-controlled room. Tolerance is a reverse indicator, so the axis starts at zero and the tightest band sits at the top of the chart. Cost does not rise linearly across those tiers. As a rule of thumb, specifying one class tighter than the part needs adds 30 to 50 percent to the unit price, because the shop has to slow the cut, add passes and add inspection. The same logic explains the title-block contradiction that costs buyers money: a drawing that states ISO 2768-mK but also carries a note demanding plus or minus 0.05 mm on every dimension forces the tighter reading, and estimators commonly add a 30 to 45 percent buffer to the price to protect against incoming inspection. One clear general-tolerance rule and specific callouts on the features that matter is cheaper and clearer than a blanket tight number. Geometric tolerances are a separate standard Dimensional tolerance says how big a feature may be; geometric tolerance says how straight, flat or square it must be. ISO 2768-2 covers this with classes H, K and L, and the numbers are small: flatness for a feature under 10 mm is 0.02, 0.05 and 0.10 mm across the three classes, rising to 0.10, 0.20 and 0.40 mm for a feature from 30 to 100 mm, and perpendicularity is 0.2, 0.4 and 0.6 mm up to 100 mm of the shorter side. On a machined part the geometric callout is often the one that decides whether the part assembles, because two faces can both be within their size tolerance and still not meet. One geometric convention is worth a note because it changes what a shop can hold without changing the function. A hole position written as plus or minus coordinates creates a square tolerance zone, while the same requirement written as a true position with a diameter creates a round one. Because the diagonal of the square is longer than its side, converting to a cylindrical zone increases the allowable area by about 57 percent, giving the machinist real margin without loosening the fit. That is a design edit that costs nothing and buys process capability. Surface finish is quoted in Ra, and it multiplies cost Finish and tolerance are separate requirements that are frequently confused. Surface finish is given as Ra, an average roughness in micrometres, and it is set by feed rate, tool condition and the number of passes rather than by the machine's positioning accuracy. Standard milling leaves Ra 3.2 micrometres, which suits most non-functional faces. A dedicated finishing pass reaches Ra 1.6; reaching Ra 0.8 usually needs grinding or high-speed milling with a specific tool; and Ra 0.4 needs a secondary grinding or lapping operation. Specify the finish per surface, not per part. A part can run Ra 3.2 everywhere and Ra 0.8 on one sealing face. The cost lesson here is about scope. A part might need Ra 0.8 on a sealing surface and nothing better than Ra 3.2 anywhere else, and the economical route is general machining for the bulk of the part with a local finishing pass confined to the sealing area. Specifying the finer finish across every face is one of the more avoidable cost mistakes in this category, because each step down the Ra ladder requires slower cutting, more passes or a different process entirely. What drives a shop above or below its brochure figure Machine positioning is the ceiling. A well-maintained three-axis vertical machining centre holds roughly 0.0127 to 0.0254 mm across its travel with repeatability near 0.005 mm; five-axis and high-speed machines do better on the same part, but only if the fixture is equally good. Aluminium moves three times as much as steel. Under the same cutting force an aluminium feature deflects about three times as far, and its thermal expansion coefficient, 23.1 parts per million per kelvin, is more than double that of carbon steel, so thin walls and long parts need control that a brochure figure ignores. Every re-clamp adds error. A part machined in three setups carries the tolerance of three fixtures, which is why five-axis work often holds a tighter feature-to-feature tolerance than the individual machine accuracy suggests. Inspection decides the honest number. A tolerance is only real if it is measured; ask for a first-article CMM report and confirm which features were checked, not just that the report exists. General tolerances do not cover fits or threads. ISO 2768 does not set a shaft fit, a thread class or a feature-specific control, so those still need an explicit callout. A blanket tight note is a cost, not a benefit. Tightening a non-functional dimension buys nothing and slows the cut. How to write a drawing that gets the tolerance you need Put one general-tolerance block in the title area, normally ISO 2768-mK, add explicit callouts only on the features whose function depends on them, mark which dimensions are functional and which are reference, and state the finish per surface rather than per part. Then send the STEP model, the drawing, the material and the annual quantity. That set lets an engineer tell you which features can be milled in one setup and which will need a controlled process. See CNC machining for the capability, surface finishing for the finishes below Ra 0.8 and aluminium die casting where a cast body plus a light machining pass is the cheaper route at volume. Send a drawing for a tolerance review Scope and sources. The ISO 2768 tables, the class-mK default and the cost of overspecification come from an ISO 2768 tolerance chart (linear classes f, m, c and v by nominal range, geometric classes H, K and L for flatness, perpendicularity and symmetry, and class m at plus or minus 0.1 mm for 0.5 to 3 mm rising to plus or minus 0.5 mm for 120 to 400 mm) and from a tolerance and GD&T cost analysis (classes f, m, c, v and H, K, L; class mK as the economic default; a 30 to 45 percent buffer added to unit price when a blanket note contradicts the title block; and a 57 percent larger allowable zone when coordinate tolerancing is replaced by a cylindrical true position). Machining tiers, machine accuracy and surface finish come from an aluminium machining tolerance guide (standard 0.1 to 0.2 mm, precision 0.025 to 0.05 mm, tight 0.005 to 0.01 mm, ultra 0.001 to 0.003 mm, aluminium deflects about three times steel and has a thermal expansion coefficient of 23.1 parts per million per kelvin against 11.5 for carbon steel, and ISO 2768-2 flatness values by class) and from a number of 2026 machining guides (as-machined Ra 3.2 micrometres, fine Ra 1.6, polished Ra 0.8, mirror Ra 0.4, and overspecifying one tolerance class adding 30 to 50 percent to cost). Machine positioning figures come from a milling machine type reference (three-axis vertical plus or minus 0.0005 to 0.001 inch with repeatability near 0.0002 inch, five-axis plus or minus 0.0002 to 0.0005 inch, and high-speed micro milling from plus or minus 0.00004 inch). Figures are planning ranges from published sources and not a quotation; every tolerance claim should be confirmed against a first-article report.

The short answer Judge a CNC milling supplier on evidence rather than on price: the registered entity and its own machine list, the tightest tolerance it holds every day with a first-article CMM report behind it, the inspection plan, and how it answers DFM questions. Ask for two past lot sizes and run a trial order before a mass-production release. Start with the entity, not the price list The first question is not what the shop charges but who you are dealing with. A factory owns and operates its machining floor, gives you direct access to the engineer who will programme your part, prices material plus machining, and can usually accept a single prototype. A trading company relays your questions through a salesperson, subcontracts the cutting, and adds a margin on top of a factory cost. Neither is automatically wrong, but only one of them can answer a technical question without going somewhere else, and you should know which you have before you release a drawing. The practical test is a live one. Ask for a video walkthrough of the machining floor, ask for the registered business entity name and its licence scope, and ask who writes the CAM programme. A factory can answer all three immediately; a broker stalls on the first and cannot answer the third. A quote that arrives instantly on complex geometry with no engineering question attached is another signal worth noting, because real manufacturability feedback takes a reading of the model. Machine list, axis coverage and the travel envelope We do CNC tells you almost nothing. What matters is whether the machine park can hold your tolerance, reach your geometry and scale to your volume without subcontracting. Ask how many machines they own, their brands and controllers, how many are three-axis, four-axis and five-axis, and whether the five-axis work is simultaneous or only 3 plus 2 indexing, because those are not the same capability. The travel envelope matters as much as the count: a shop with a 4,000 mm machine solves large parts and a shop with a 500 mm machine cannot, whatever its tolerance claim. Work down the gates in order. A supplier that cannot clear the first one is not a manufacturing partner however good the price looks. Axis coverage decides how many setups your part needs, and setups are where tolerance is lost. A three-axis machine handles prismatic brackets and housings at the lowest cost. A four-axis machine with a rotary table cuts features on four faces in one setting. A 3 plus 2 machine indexes and locks two rotary axes for angled holes and tilted faces. A five-axis simultaneous machine keeps one datum across every face and reaches free-form surfaces, deep cavities and undercuts without re-fixturing. Match the configuration to the part rather than to the brochure, and confirm that the volume you are planning can be absorbed in-house if it triples. Tolerance claims and quality paperwork Tolerance is the number that separates one quote from another and the one most often quoted without evidence. A shop stating 0.005 mm across a production run is making a different claim from a shop that hit 0.005 mm once on a sample, and the way to tell them apart is to ask for the routine daily holding tolerance per feature and for a first-article CMM report on a comparable part. Ask which equipment verifies it: a coordinate measuring machine, a surface roughness tester, a hardness tester, an optical comparator. A supplier who cannot explain how the tolerance is achieved and which instrument proves it is offering an aspiration rather than a capability. Print this as a side-by-side scorecard. The supplier with the weakest answers on tolerance and inspection will cost the most regardless of price. The same evidence standard applies to surface finish and material. Finish is a process number, so ask what Ra the shop holds as machined and whether reaching Ra 0.8 needs grinding. Material is a traceability question, so ask for mill test reports on the exact grade quoted, because a substituted alloy that is not disclosed changes strength and corrosion resistance and will not show up until the part fails in service. Where your own contract names a quality-system standard, ask for the certificate number and scope and verify it with the issuing body rather than with the supplier. Paperwork is the third thing a supplier sells, alongside metal removal and process control, and it is the easiest to verify. Ask for the inspection plan rather than an inspection promise: raw-material verification, in-process checks and a final dimensional review are three separate activities, and a shop that only performs a final check has no way to catch a drift before the whole lot is cut. Ask for a first-article inspection on the first part of a new programme, a mill test report per material batch, and material declarations such as RoHS and REACH where your market requires them. For precision work, ask for CMM data that covers the geometric tolerances and not just the key dimensions, because flatness and true position are usually what decide assembly. Defect history belongs in the same conversation. Ask what scrap rate the shop considers normal and how it handles a non-conformance, then judge the answer by whether it includes a corrective action rather than a discount. A supplier that reports a production problem before delivery is worth more than one that reports it after, and the behaviour you see during quoting is the behaviour you will see during production. DFM feedback and communication are the cheapest signals Design-for-manufacture feedback is the single cheapest way to judge a supplier, because it costs you nothing and reveals whether anyone engineering your part has actually read the model. Good feedback identifies inaccessible internal corners, deep pockets that will deflect a long tool, thin walls that will distort, difficult thread locations, tolerances that are tighter than the function needs and datum structures that will not survive inspection. Published estimates put the saving from good DFM at 10 to 25 percent of part cost, and a technical answer that arrives within one to two working days during quoting is a reasonable communication baseline. Slow or absent answers at the quoting stage rarely improve once the order is placed. Order in stages: sample, trial lot, then mass production A sample and a pilot lot are the last gate, and they are worth the time because they test things a quotation cannot. The sample verifies machining quality and finish; the pilot lot verifies that the second part comes out like the first, which is the whole point of process control. Include representative critical features in the trial rather than an artificially simple part, and check the inspection report against the drawing yourself. Build the schedule with the real stages in mind, because lead time includes technical review, material procurement, programming and fixtures, machining, finishing, inspection and freight rather than spindle time alone. Confirm production time and shipping time separately. Quotation and DFM feedback usually arrive inside 12 to 48 hours and are not part of these figures. Confirm quantity bands before you commit to a volume. A shop that runs one prototype and ten thousand-piece lots is set up for both; a shop whose minimum order is tied to a sub-supplier's minimum is not. Ask for two past lot sizes as evidence rather than for a policy statement, and match the production slot to your real demand so that a first order is not also a warehouse decision. Red flags that predict a bad outcome No DFM feedback before the quote. A supplier that never asks about tolerances, undercuts, thread depth or material alternatives will cut exactly what the drawing says, even where it cannot be made well that way. A price far below the market. Very low quotes usually exclude finishing, inspection or packaging, or assume a substituted material; the difference surfaces later as rework or a rejected lot. Vague on tolerance and inspection. If the shop cannot say how it holds the tolerance or which instrument verifies it, treat the number as aspirational. No material certificates. Mill test reports are standard practice for a metal part; a supplier that cannot produce them may be using unverified stock. Slow technical responses. Answers that take more than two working days before an order rarely get faster after one. A refused factory walkthrough or an instant quote on complex geometry. Both suggest the engineering step is missing from the process. What to send, and how to compare Prepare a small RFQ package before contacting anyone: the STEP model and a toleranced drawing, the material and finish callout, the tightest tolerance on the drawing, the annual quantity, the largest outside dimension, and a note on which features are functional. Those six items decide which shops can quote at all, and they make two quotations comparable. Then compare fully loaded delivered prices, including finishing, inspection, packaging, freight and terms, rather than unit prices. See CNC machining for the capability set, become a partner for the supplier-side view and surface finishing for the secondary operations that a quotation should include. Send a STEP file and read the DFM notes Scope and sources. The evaluation checkpoints, red flags and stage gates come from a CNC supplier evaluation checklist (entity, machine capability, axis coverage, tolerance and repeatability, materials and mill certificates, finishing, inspection plan, CMM on GD&T, first-article inspection, DFM support, minimum order, lead time, capacity, communication, IP, sample gate, consistency, shipping and after-sales; a factory starting at one piece against a trader tied to sub-supplier minimums; three-axis, four-axis, 3 plus 2 and five-axis simultaneous as different capabilities) and from a China supplier selection guide (similar-part experience, tolerance capability, material sourcing and mill certificates, in-house against outsourced finishing, quality systems and first-article policy, inspection equipment, communication, capacity, packaging and after-sales; red flags including no DFM feedback, a quote 30 percent or more below market, no material certificates, technical responses slower than 48 hours, and a refused factory walkthrough). Lead-time stages and delay causes come from a China cost and quality guide (drawing review, material procurement, process preparation, first article, batch machining, finishing, final inspection and transport as separate stages) and lead-time bands from a volume economics guide (prototype 5 to 10 business days, small batch 10 to 15, medium volume 15 to 25 and high volume 25 to 45, with dedicated fixturing from USD 500 to 2,000 justified at 100 pieces and up) and from a lead-time breakdown (1 to 3 pieces 2 to 7 days, 4 to 10 pieces 4 to 10 days, 11 to 50 pieces 6 to 14 days, 51 to 250 pieces 10 to 15 business days, a quotation within 24 hours of a complete RFQ package, and a standard 3 to 5 business day prototype window with a 48-hour express route for simple geometry). DFM saving and quote turnaround figures also draw on a factory audit guide (quote plus DFM notes in 12 hours, parts shipping in 3 to 5 days, and a machine mix that explains what the shop can and cannot do). Figures are planning ranges from published sources and not a quotation; no claim about any supplier's certification is made here, and a quality-system document should be requested and verified per supplier and per project.

The short answer CNC milling holds the workpiece and moves a rotating cutter along a toolpath generated from the CAD model, removing material in thin layers. Three axes cut prismatic faces, pockets and slots; a fourth adds indexing, helical and cylindrical work; five simultaneous axes cut free-form contours and undercuts in one setting. The four steps from model to finished face Milling is subtractive, so the part starts as a block or a plate and the machine removes everything that is not the part. The work begins in software: CAM takes the CAD model, chooses the tool for each region, and generates toolpaths that tell the machine where to cut, how fast to move and how deep to go. On the machine, the workpiece is fixed to a table and a datum is established so that every feature is measured from the same reference. The spindle then drives the cutting tool, usually an end mill, a ball mill, a drill or a tap, at speeds that depend on the material and the tool, from roughly 1,500 revolutions per minute in steel to 12,000 and beyond in aluminium, and the tool removes material in layers until the surface matches the model. The fixture is not a detail. A part that moves under the cutter loses the tolerance the machine was bought for. Two numbers describe what the machine can do and they move together. Positioning accuracy is how close the machine brings the tool to the commanded point, and repeatability is how consistently it returns to the same place. On a well-maintained three-axis vertical machining centre those are typically 0.0127 to 0.0254 mm and about 0.005 mm respectively. The part in the fixture usually lands wider than that, because tool deflection, thermal growth, workholding stiffness and material batch all add to the machine's own error. These are machine capability figures on an axis where lower is tighter. The achievable part tolerance is set by the fixture and the tooling as much as by the machine. What each axis configuration unlocks Axis count is the single design decision that most changes cost and tolerance, and the four common configurations are not a ladder of quality but a set of different capabilities. A three-axis machine moves the tool in X, Y and Z and handles prismatic parts, brackets, housings and faces at the lowest cost and the fastest setup. A four-axis machine adds a rotary table, so features on four faces or around a cylinder can be cut without turning the part by hand, and a high-precision fourth axis indexes to roughly 5 to 15 arc-seconds. A 3 plus 2 machine indexes two rotary axes and then locks them, which is the economical way to drill angled holes and machine tilted faces. A five-axis simultaneous machine tilts and rotates while cutting, so it holds one datum across every face and can use shorter, stiffer tools. The practical consequence is fewer setups, and setups are where cumulative error lives. A part that a three-axis machine needs three fixtures to finish carries the tolerance of all three; the same part in a single five-axis setting carries one. That is why five-axis work often holds a tighter feature-to-feature tolerance than the individual machine accuracy would suggest, and why it is the normal route for free-form surfaces, impellers and turbine blades. The shape families milling produces, and their limits Most milled parts are made of a small number of shape families, and each family has a configuration that produces it cheapest and a physical limit that decides whether it is practical at all. Free-form contours and undercuts are the two families that genuinely require five axes; the rest can usually be done for less. Shape familyConfigurationPractical limit Flat faces and square pockets3-axis, X Y ZInternal corner radius must be at least the tool radius Slots and keyways3-axisDepth under about 10 times the slot width in aluminium Drilled and tapped holes3-axis, or 5-axis for angled entriesDepth under about 10 times diameter on a standard drill Angled holes and tilted faces3 plus 2 index and lockReplaces two or three separate setups Helical, cam and cylindrical work4-axis rotaryIndexing accuracy of roughly 5 to 15 arc-seconds Free-form contours, impellers5-axis simultaneousOne datum across every face Undercuts and compound angles5-axis or a special cutterT-slot, lollipop and dovetail tools reach in The pattern is that every shape family has a narrower, cheaper configuration than the one that can technically make it. A pocket with a generous internal radius is a three-axis job; the same pocket with a sharp corner becomes a two-operation job because a round cutter physically cannot leave a square internal corner, and clearing it needs a smaller tool or a wire EDM pass. A part with six faces of features is a five-axis job only if the volume justifies it; at low volume, three setups on a three-axis machine can be cheaper even though they are slower. Design limits worth writing into the model Minimum wall thickness. Around 0.5 mm is achievable in metal, but 0.8 mm and up is the reliable zone, because thin walls vibrate and deflect under the cutting force and stainless deforms more than aluminium. Internal corner radius. Keep the radius at a third of the pocket depth or larger, and never smaller than the tool radius. Corners below about 0.25 mm need a specialist tool or a second process. Pocket depth to width. An end mill is stable to roughly 4 to 1 in aluminium, 3 to 1 in mild steel and 2 to 1 in stainless; past that the tool deflects, the finish suffers and the tolerance drifts. Hole depth to diameter. A standard twist drill is comfortable to about 10 times diameter, after which peck cycles or a gun drill take over. Floor radius. A flat-bottomed end mill leaves a small nub at its centre, so allow a floor radius of about 0.25 mm or call out that the floor needs a secondary clean-up. Undercuts are a five-axis or special-tool feature. If no straight-down view of the model shows every feature, the part contains an undercut and needs a configuration that can reach it. Where milling stops being the right process Milling removes material from a solid block, so the geometry has to be reachable, the material has to be cuttable and the volume has to justify the cycle time. Deep, narrow pockets need long, slender tools that deflect, which is why a design with a 10 to 1 pocket is often cheaper as a cast or a two-piece assembly. Sharp internal corners are impossible with a round cutter and belong in wire EDM. Hard materials such as titanium and high-temperature alloys cut slowly and wear tools quickly, so a part with a simple profile may be cheaper as a forging or a casting with a light milling pass. And any geometry that is still changing is expensive to mill at volume, because every revision is new programming and sometimes a new fixture. The honest comparison is always between a milled part, a formed part and a cast part at your real quantity, not between milling and nothing. How to send a part for milling Send the STEP model and a toleranced drawing, the material and stock form, the annual quantity, the finish per surface, and a note on which features and datums are functional. Those five items let an engineer choose the axis configuration, the number of setups, the tool sizes and the fixture before a price is set. See CNC machining for the process and its limits, metal bending where a formed shape beats a cut one, and surface finishing for the anodising, coating or polishing that follows the cut. Send a model and get a setup plan Scope and sources. Axis capability, machine accuracy, design limits and feeds come from a CNC milling machine reference (three-axis vertical plus or minus 0.0005 to 0.001 inch with repeatability near 0.0002 inch, four-axis rotary indexing to 5 to 15 arc-seconds, five-axis simultaneous plus or minus 0.0002 to 0.0005 inch, high-speed micro milling from plus or minus 0.00004 inch, and worked feeds of 8,000 to 12,000 revolutions per minute at 40 to 80 inches per minute in 6061 aluminium against 1,500 to 3,000 revolutions per minute at 10 to 25 inches per minute in 4140 steel). Shape families and their configuration come from a supplier capability guide (three-axis for prismatic parts, four-axis for features on four faces, 3 plus 2 for angled holes and tilted faces, and five-axis simultaneous for free-form surfaces, impellers and deep cavities with one datum across faces). Design limits come from a CNC design guide (pocket depth to width 4 to 1 in aluminium, 3 to 1 in mild steel and 2 to 1 in stainless, a worked slot example, minimum diameter 0.020 inch with 0.060 inch preferred, maximum depth 10 times diameter for standard drills, floor radius of at least 0.010 inch, and undercuts requiring five axes, a special cutter or an extra setup) and from a tolerance and feature guide (as-machined finish Ra 3.2 micrometres with a finishing pass at Ra 1.6, minimum wall 0.5 mm in metal, internal corner radius a third of cavity depth, and hole depth under 10 times diameter to avoid deflection). Figures are planning ranges from published sources and not a quotation; confirm them against your material, geometry and quantity.

The short answer Five industries carry most precision CNC turning demand: automotive at 27 percent of precision machining revenue, aerospace and defence at 24 percent, medical devices at 18 percent, semiconductors and electronics at 16 percent, and industrial machinery at 15 percent. All five buy a round form that must run true at speed. Where precision turning demand actually sits Turning is the second largest process in precision machining after milling, and the split is documented rather than anecdotal. Global precision machining was worth about USD 129 billion in 2025, of which CNC turning took an estimated 28.1 percent, or USD 36.4 billion, against 37.9 percent for milling, 15.1 percent for grinding and honing, and 11.9 percent for EDM. Turning keeps that position because rotational parts cannot be made another way at the same cost: a lathe removes material in one continuous cut around an axis, where a mill has to step around it. All five bars are the same measure, so they are directly comparable. The note on each row names the turned parts the market typically buys. End marketShare of precision machining revenue, 2025What it buys from a lathe Automotive26.8 percent, about USD 34.7 billionCrankshafts, pistons, transmission shafts, brake caliper pistons, EV motor shafts Aerospace and defence23.8 percent, about USD 30.8 billionHydraulic fittings, landing gear pins, engine shafts, high-pressure sleeves Medical devices18.4 percent, about USD 23.8 billionBone screws, dental implants, surgical instrument bodies, catheter parts Semiconductor and electronics16.0 percent, about USD 20.7 billionConnector pins, sensor bodies, RF housings, wafer-handling hardware Industrial machinery15.0 percent, about USD 19.4 billionValve stems, guide bushings, rollers, hydraulic spools, gear blanks Two more markets sit just outside the top five and are worth naming, because they buy turning in a different way. Oil and gas buys reached-out geometry more than cylinders for motion: in that sector around 64 percent of valve and pipe fittings are machined on CNC lathes, holding dimensional accuracy below 10 microns. Robotics and automation buy turned parts for balance rather than for fit, because an actuator shaft that runs out of round vibrates at speed even when every diameter is in tolerance. Automotive and aerospace: volume first, runout second Automotive is the largest single consumer of turned parts, and it consumes them for two separate reasons. High-volume drive components such as pistons and shafts are turned because the process is fast and repeatable at volume: turning accounts for roughly 58 percent of cylindrical component production in automotive manufacturing, and around 30 percent of all installed CNC machines worldwide are in automotive plants. The electric vehicle transition is adding a second, more precise family of turned parts, because motor shafts and battery terminal connectors carry tolerances that a legacy drivetrain part did not need, and EV work increased CNC machining hours by about 36 percent at the plants that build them. Aerospace buys turned parts for the opposite reason. Volume is low and the driver is geometric integrity: a hydraulic fitting that leaks, or a gear pin that is not concentric, is a functional failure rather than a cosmetic one. Aerospace is also where the material stays hard - titanium and nickel alloys account for roughly 47 percent of aerospace component production - so the turning shop is being asked to hold tight form in a material that pushes back. That combination, a tight geometric callout in a difficult alloy, is what separates a turning shop that can quote aerospace work from one that cannot. Medical, oil and gas, semiconductor: the precision end The sectors that rank third to fifth in size ask for the tightest work per part. Medical turning is dominated by slender, small-diameter geometry: around 40 percent of dental implants are produced on high-precision Swiss-type lathes, and Swiss machines hold roughly 0.005 mm on long slender features and reach Ra 0.2 to 0.4 micrometres in a single pass, which is why bone screws, implant bodies and dental abutments are made that way rather than by milling. Semiconductor and electronics work is smaller still and is judged on cleanliness and surface finish as much as on size, because a particle or a burr in a vacuum component is a yield problem rather than a fit problem. Read the axis as tighter to the left. These are industry practice bands from published turning capability tables, not the limit of the machine. Oil and gas is the sector where the environment, not the drawing, sets the requirement. Downhole tools, wellhead equipment and valve components run at pressure, at temperature and in fluids that corrode, so the bought materials are corrosion-resistant alloys and hardened steels, and the acceptance criteria are usually hardness records and non-destructive testing rather than a single dimensional callout. Around 12 percent of large-format CNC boring mill capacity in the energy sector is consumed this way. Six industries, and what each one buys from a lathe The useful pattern for a buyer is that the industry name matters less than the feature on the drawing. Every one of these sectors orders the same four families from a turning shop: external diameters and faces, internal bores, threads on a round body, and grooves or tapers. What changes is which of those families is functional, and therefore which tolerance actually has to be met. Read the last column first if you are choosing a supplier. Each industry audits something different, and that is what a capability claim has to answer. A useful way to test whether a part belongs on a lathe is to ask whether it is built around one axis. If the majority of its features are concentric to a single centreline, turning will usually be faster, more concentric and cheaper than milling it. If the features are spread across several faces and none of them is dominant, it is a milling part. Parts with both, such as a valve body with round ports and flat mounting faces, are the normal reason a turn-mill machine or a two-operation route exists. What these industries audit before they buy Each sector names a different piece of evidence, and a turning supplier that cannot produce it will be excluded before price is discussed. None of the following is a substitute for checking the specific paperwork your own contract requires. Automotive asks for process capability and a production part approval file on the critical features, which means the shop must be able to run the part, measure it, and show the spread rather than a single good sample. Aerospace and defence asks for heat-lot traceability from the mill certificate through to the finished part, plus first-article inspection reports on the critical features. Medical asks for the material grade, its condition and the surface finish result, because biocompatibility and cleanability are functions of the surface, not only of the geometry. Oil and gas asks for hardness and non-destructive test records, because the failure mode being prevented is a crack, not an out-of-tolerance diameter. Semiconductor and electronics asks for cleanliness and a measured Ra figure, because a burr or a particle matters more than a few microns of size. If your own contract names a quality system standard, ask for the certificate number and its scope and verify it with the issuing body rather than with the supplier. What matters on the floor is whether the shop can show measured results on your features, batch after batch. Where turning is the wrong answer Turning is a poor fit in four situations, and recognising them early saves a redesign. The first is prismatic geometry: if the part is essentially a plate with pockets, a mill is the correct machine and a lathe cannot reach the features at all. The second is off-axis features in quantity: a cross hole is a live-tooling operation on a lathe, but six cross holes on three faces usually make the part a mill-turn job or a two-operation route, and the price reflects the second setup. The third is very large diameters: conventional chuck turning handles parts up to roughly 250 mm, while a Swiss-type lathe tops out near 32 mm of bar, so the two machine classes are not interchangeable. The fourth is huge volume of a simple part, where cold forming or stamping beats chip removal on unit cost once the tooling is amortised. How to brief a turning job Send the STEP model and a toleranced drawing, the material grade and its condition, the quantity with an annual forecast, the surface finish required per surface, and a note on which diameters and which faces are functional. Those five items let an engineer decide the machine class, the bar size, the number of operations and the gauging plan before price is discussed. A shop that answers by naming a machine class and a setup count has understood the part; one that answers with a number only has priced it. See CNC machining for the process itself, surface finishing for what happens after the cut, and become our partners for how a supplier enters our manufacturing network. Send a drawing and get a turning plan Scope and sources. Market shares and segment values come from a worldwide precision machining market study (CNC turning 28.10 percent of process value at USD 36.36 billion in 2025 against milling at 37.94 percent, grinding and honing at 15.08 percent and EDM at 11.85 percent; automotive 26.81 percent of end-user revenue at USD 34.7 billion, aerospace and defence 23.81 percent at USD 30.8 billion, medical devices 18.38 percent at USD 23.8 billion, semiconductor and electronics 15.99 percent at USD 20.7 billion, industrial machinery 15.01 percent at USD 19.4 billion; total precision machining market about USD 129 billion). Share-of-production and machine-installation figures come from a CNC machine tool market report (lathes 32 percent of installed machines with over 1.1 million units, turning 58 percent of cylindrical component production in automotive, 64 percent of oil and gas valve and pipe fittings machined on CNC lathes to under 10 microns, EV machining hours up 36 percent, titanium and nickel alloys 47 percent of aerospace component production, 21 percent of CNC demand from aerospace and defence). Sector consumption and end-use shares come from a CNC industry statistics compilation (automotive 30 percent of installed CNC machines, medical device machining growing 7.5 percent a year, over 60 percent of hydraulic components produced on turning centres, 40 percent of dental implants on Swiss-type lathes, energy sector 12 percent of large-format boring mill capacity) and from an industry ranking by market size. Tolerance bands per industry come from a CNC turning precision reference (aerospace 0.002 to 0.005 mm, medical 0.003 to 0.010 mm, electronics 0.005 to 0.010 mm, automotive 0.010 to 0.020 mm) and Swiss machine capability from a Swiss turning comparison (0.005 mm on long slender features, Ra 0.2 to 0.4 micrometres, bar capacity up to 32 mm, spindle 6,000 to 12,000 revolutions per minute). Figures are planning ranges from published sources, not a quotation and not a statement about any named facility; confirm them against your own drawing, material and quantity.

The short answer Custom CNC turning holds ±0.01 to 0.05 mm to standard, tightens to 0.002 to 0.005 mm on a precision lathe, and reaches 0.001 to 0.002 mm on a Swiss-type machine. Roundness runs 0.5 to 3 micrometres and surface finish reaches Ra 0.4 micrometres. Geometry, not the machine alone, sets the limit. Three tiers, and what separates them Turning capability is usually quoted as a single number, which is misleading, because the same lathe that holds 0.05 mm on a long slender shaft will hold 0.005 mm on a short rigid collar in the same program. The published tiers are still a fair starting point. General industrial turning sits at 0.01 to 0.05 mm, precision turning at 0.002 to 0.005 mm, and Swiss-type turning at 0.0025 to 0.005 mm with repeatability in the 0.002 mm band on a high-end machine. At the extreme end, a purpose-built machine can reach 0.002 mm and a diamond-turned optical part tighter still, but that is a specialised process rather than a turning shop's normal output. The dimensional figure is rarely the one that fails an assembly. Read the roundness and finish columns as well. TierDimensional toleranceRoundnessFinish RaTypical use Standard latheplus or minus 0.010 to 0.050 mm2 to 3 micrometres0.8 to 1.6 micrometresBrackets, spacers, general shafts Precision latheplus or minus 0.002 to 0.005 mm0.5 to 1.0 micrometre0.4 to 0.8 micrometreBearing seats, hydraulic spools Swiss-type latheplus or minus 0.0025 to 0.005 mm0.5 to 1.0 micrometre0.2 to 0.4 micrometresBone screws, pins, connectors The reason the two ends of the scale behave differently is the guide bushing. On a conventional lathe the bar hangs out of the chuck, so the unsupported length grows as the tool moves along it, and a long part deflects under cutting force. A Swiss-type lathe feeds the bar through a guide bushing a millimetre or two behind the cutting edge, so the supported span never changes and the effective overhang is constant whatever the part length. That is what lets it hold 0.005 mm on slender work and produce finishes down to Ra 0.2 to 0.4 micrometres in a single pass. The geometric callouts nobody quotes, and everybody needs Size and form are different things, and a part can be perfectly on diameter and still useless. A bearing seat that is 0.003 mm oversize but two micrometres out of round will transmit vibration; a spool that is round in every cross-section but bowed by five micrometres will not seal. The four callouts that carry that function are roundness, cylindricity, concentricity and runout, and they are quoted separately from the diameter for a good reason. Every band is a deviation in the same unit, so the rows can be read against each other. Lower is tighter across the whole chart. CalloutWhat it controlsTypical achievable figure RoundnessWhether every cross-section is a true circle2.5 micrometres standard, 0.5 to 1.0 micrometre controlled CylindricityRoundness and straightness combined over the whole surface5.0 micrometres standard, 1.0 to 2.0 micrometres controlled ConcentricityWhether inner and outer diameters share one axisAbout 0.01 mm, or 0.005 mm with one-setting machining Total runoutThe total indicator reading as the part rotates0.010 mm standard, 0.003 to 0.005 mm controlled One worked example shows why these matter more than the diameter in hydraulics. On a dynamic seal seat the roundness figure should not exceed 0.002 mm, and a deviation above 0.005 mm is enough to make the seal snake and wear unevenly; on a piston rod 500 to 2,000 mm long, cylindricity has to stay under 0.01 mm over the whole length. Studies cited in that sector put the penalty bluntly: a cylindricity deviation beyond 0.003 mm on a dynamic seal cuts seal life by 30 to 50 percent. Surface finish is a tolerance in its own right Ra is the roughness average of the turned surface, and on a rotating part it decides how the part behaves at the interface rather than how it fits. As-machined turning typically lands between Ra 0.8 and 1.6 micrometres. A programmed finishing pass with an appropriate nose radius and a wiper insert takes the same part to 0.4 to 0.8 micrometres, and a Swiss-type machine with a rigid setup reaches 0.2 to 0.4 micrometres without grinding. Going finer than that normally means grinding or lapping rather than turning. The counter-intuitive part is that smoother is not always better. Hydraulic piston rods specify Ra 0.2 to 0.4 micrometres after grinding and plating, and a surface below Ra 0.1 micrometres can hold too little oil to keep the seal lubricated, which increases friction rather than reducing it. Above Ra 0.8 micrometres the surface abrades the seal. At 350 bar, an Ra above 0.8 micrometres on a rod has been reported to cut seal life from a planned 2,000 operating hours to under 800. Corrosion resistance behaves differently again: anodising and plating amplify surface texture rather than hiding it, so a rough turned surface finishes rough. What actually limits the precision you receive Four factors decide whether a shop delivers the tier you asked for, and only the first one is bought rather than managed. Machine condition sets the ceiling. The other three decide how much of that ceiling you collect on a production batch. Machine condition. A lathe is only as good as its current geometry. A machine with 0.002 mm positioning accuracy and closed-loop tool wear compensation is the entry condition for micron work; a worn spindle bearing is not fixed by slowing the feed. Tooling and insert grade. A sharp edge with the right nose radius and coating cuts consistently; a dull insert pushes the material instead of shearing it and the diameter drifts through the batch. Thermal stability. A one degree change in shop temperature moves a steel part by roughly 11 micrometres per metre of length, so a temperature-controlled cell is a real requirement above the precision tier rather than a marketing line. The gauging loop. Precision comes from measuring and compensating, not from hoping. Machines with thermal compensation and automatic tool offsets hold about 0.003 mm through a whole shift without an operator rescuing the dimension. Material belongs on the same list. Aluminium and brass are geometrically stable and cut cleanly; austenitic stainless work-hardens and moves, and titanium generates heat, work-hardens quickly and springs back. A tolerance that is routine in 6061 may need a changed cutting strategy in Ti-6Al-4V, and a shop quoting the same figure for both is quoting a catalog rather than a process. Capability versus assured capability A single good part proves nothing about a batch. The statistic that settles it is process capability: a Cpk above 1.33 means the natural spread of the process sits inside the tolerance band and corresponds statistically to roughly 66 non-conforming parts per million, while high-pressure hydraulic work commonly requires Cpk above 1.67. This is the number to ask for when the part rotates, because it is measured on your feature, on your material, at your batch size, rather than on the machine specification sheet. Annual quantity also shapes what is achievable. Conventional turning setup runs roughly USD 200 to 600, while a Swiss-type machine runs USD 400 to 1,200 plus guide bushing preparation, for a machine hour rate of about USD 90 to 180 against USD 60 to 120 for a conventional lathe. Swiss-type work becomes the right answer above roughly 3,000 parts a year, and a conventional lathe with a tailstock or steady rest is the sensible route below that, even when the tolerance could be met either way. Where tight turning tolerances are the wrong thing to buy Three situations call for a discussion before an order. First, when a tight callout is on a feature nobody measures functionally: every extra decimal place adds inspection time and scrap risk. Second, when the tight dimension spans a long unsupported length: ask instead whether a stepped diameter, a shoulder, or a change of process would remove the need. Third, when the part is plastic. POM, nylon and PEEK move with moisture and temperature, so holding turning tolerances tighter than the functional need on a plastic part buys a measurement that will not survive the first week in service. How to specify a turning tolerance Send the STEP model and a drawing that carries the general tolerance block plus explicit callouts on the functional features, the material grade and condition, the quantity and annual forecast, the finish per surface, and the specific diameters whose roundness or runout actually matters. Five items, and an engineer can then tell you honestly whether the figure is routine on their machines, achievable with a changed strategy, or impossible. See CNC machining for the process, surface finishing for what follows turning, and SOMI 3D printing when the feature you need cannot be cut from solid at all. Send a drawing and get a tolerance review Scope and sources. Tier figures, geometric callouts and the industry tolerance table come from a CNC turning precision reference (standard 0.01 to 0.05 mm, high precision 0.002 to 0.005 mm, Swiss tighter still, roundness within 0.001 mm, Ra 0.4 to 1.6 micrometres, repeatability within 0.002 mm, and 0.002 mm positioning accuracy for medical work) and from a precision turning capability table (geometric tolerances 0.005 mm standard against 0.001 to 0.002 mm controlled, Ra 0.8 against 0.2 to 0.4 micrometres, roundness 2.5 against 0.5 to 1.0 micrometres, cylindricity 5.0 against 1.0 to 2.0 micrometres, total runout 0.010 against 0.003 to 0.005 mm). Geometric tolerance ranges also draw on a CNC turning accuracy specification (roundness within 0.005 mm, cylindricity within 0.01 mm, straightness within 0.005 mm, concentricity within 0.01 mm, perpendicularity within 0.005 mm, Ra 0.8 to 1.6 micrometres, tolerances per ISO 2768 or as designed). Hydraulic sealing figures and process capability come from a hydraulic turning reference (roundness under 0.002 mm for dynamic seal seats, cylindricity under 0.01 mm over 500 to 2,000 mm, Ra 0.2 to 0.4 micrometres after grinding, seal life cut 30 to 50 percent by a cylindricity deviation above 0.003 mm, Ra above 0.8 micrometres at 350 bar cutting rod seal life from 2,000 hours to under 800, Cpk above 1.33 equal to about 66 parts per million, Cpk above 1.67 for high-pressure hydraulic work, plus or minus 0.003 mm held through a shift with thermal compensation). Swiss-type capability, cycle structure and setup economics come from a Swiss turning comparison (Swiss plus or minus 0.005 mm against conventional plus or minus 0.02 mm, Ra 0.2 to 0.4 against 0.8 to 1.6 micrometres, spindle 6,000 to 12,000 against 3,000 to 6,000 revolutions per minute, setup USD 400 to 1,200 against USD 200 to 600, machine hour USD 90 to 180 against USD 60 to 120, break-even around 3,000 parts). Figures are planning ranges from published sources and not a quotation; confirm them against your material, geometry and quantity.

The short answer The difference is motion. CNC turning spins the workpiece against a stationary single-point tool, so it makes round, axially symmetric parts fast and concentrically. CNC milling holds the part still and rotates a multi-point cutter, so it makes flat faces, pockets and slots. Round part: turn it. Everything else: mill it. One mechanical fact explains the rest Almost every practical difference between the two processes follows from a single fact: on a lathe the part turns and the tool is still, and on a mill the tool turns and the part is still. That decides which cutter can be used, how the cut is loaded, what shape comes out naturally, and which tolerance the process is good at holding. A turning tool is a single point in continuous contact with the material, and a milling tool is a multi-edge body entering and leaving the cut on every revolution. Read down the columns rather than across. Turning and milling are not two settings of the same machine; they are different cutting mechanics. AspectCNC turningCNC milling What movesThe workpiece, held in a chuck or colletThe cutting tool, on a spindle Cutter typeSingle point, on a turret or gang slideMulti-point, end mills and drills Cutting actionContinuous contact around the barInterrupted, one tooth at a time Natural shapeCylindrical, conical, threaded, groovedPrismatic, flat, pocketed, contoured Tolerance strengthDiameter, roundness, runout, concentricityPosition, planar flatness, feature-to-feature Surface left behindContinuous finish around the diameterFine finish, with cutter marks on curves Typical partsShafts, pins, bushings, fittings, fastenersBrackets, housings, plates, moulds, covers Axis count2 linear axes, plus C and live tooling3, 4 or 5 axes, plus rotary tables Process selection is not a matter of taste. Turning occupies about 28 percent of global precision machining value and milling about 38 percent, and those shares track the fact that most engineered parts are prismatic while most rotating assemblies are not. Choosing the wrong side of that line is the most expensive routing mistake a buyer can make, because it cannot be recovered by asking for a tighter tolerance. The cutting action decides the failure mode Continuous cutting means the turning tool stays engaged, so heat and load are steady, tool life is predictable and the resulting surface has no step-over marks along the cut. It also means the part is loaded steadily and deflects steadily, which is why a slender turned part bends rather than chatters. Milling is the opposite: each tooth enters and leaves the cut, so the load pulses, and the process is more tolerant of interrupted surfaces and harder materials but leaves a cutter mark wherever the tool steps over. That is why the two processes are good at different tolerances rather than at different qualities. Turning holds diameter, roundness, cylindricity and runout because those are the dimensions that follow the rotation of the work. Milling holds position and flatness, because those are the dimensions that follow the programmed path of the tool. A part can be a perfect turned diameter and a terrible milled pocket, or the reverse, on the same machine shop floor. Where each process is the wrong answer Three mistakes appear repeatedly in quotations. The first is milling a part that is fundamentally round: a lathe cuts it faster and more concentrically, and the mill cannot match the continuous finish on a long cylindrical span. The second is turning a part whose function lives on flat faces: a lathe cannot reach a pocket at all, so the feature either moves to live tooling, which is slower and often shorter in reach, or the part needs a second operation. The third is ignoring how much of the part is off-axis. A round body with one cross hole is a turning part with a live-tooling step. A round body with eight cross holes on three faces is a mill-turn part or a two-operation route, and pricing it as plain turning guarantees a schedule problem later. Feature on the drawingNatural processWhy External and internal diametersTurningA single point cuts a true circle by construction Threads on a round bodyTurningThe thread is generated by the same rotation Grooves, tapers, chamfersTurningAll concentric with the axis of rotation Flat faces, pockets, slotsMillingThese require the tool to travel in a plane Holes away from the centrelineMilling, or live toolingOff-axis features need a second axis of motion Keyways and splinesMilling, or broachingBroaching is the volume answer for blind splines Free-form 3D surfacesMilling, 5 axisThe tool vector has to change continuously Undercuts and compound anglesMilling, or a special toolNo straight-in approach exists from one direction Turned first, milled second: the hybrid route Most parts that look like a routing dilemma are actually both. The normal workflow is to turn the round profile to size, then move the part to a mill for flats, holes, slots or a keyway. Every transfer between machines adds setup time, another workholding step and a fresh chance to lose the datum, so the cumulative tolerance grows with the number of operations rather than with any single machine's accuracy. Every bar is the same measure, so the rows are directly comparable. Turning is the smaller share but it is not the second choice for anything round. A mill-turn machine removes the transfer rather than speeding it up. It turns the part, then either indexes the C axis or stops the spindle and uses live tooling to mill flats, drill cross holes and cut slots without unclamping. The consequence is that concentricity survives the transition from cylindrical features to off-axis ones, because nothing was ever released from the chuck. That is why the hybrid route is the standard answer for valve bodies, drive shafts with off-centre holes, and pump or motor housings with round bores and flat mounting faces. Work top to bottom and stop at the first line that describes the part. The later stages cost more per part, not less. Where the two processes are compared fairly The comparison that matters is not turning against milling in general, but turning against milling on your part. For round work, turning is normally faster, more concentric and cheaper, and it pulls further ahead as volume rises because bar feeders allow near-continuous cutting and unattended running. A conventional lathe runs at roughly 3,000 to 6,000 revolutions per minute and a Swiss-type machine at 6,000 to 12,000, with typical cycle times of 25 to 60 seconds and 8 to 20 seconds respectively for comparable small parts. For prismatic work the comparison reverses, and milling becomes the only process that can produce the geometry at all. The economics follow the same split. Turning setup runs roughly USD 200 to 600 and milling setup is dominated by fixture and programming time, which grows sharply with axis count and with the number of faces to be reached. That is why the practical question is rarely which process is better and almost always how many setups the drawing forces, because setups are where cost, lead time and accumulated error all come from. How to route a part before you quote it Send the STEP model and a toleranced drawing, the material and stock form, the quantity with an annual forecast, the finish per surface, and a note naming which features and which datums are functional. Those five items let an engineer decide the machine class, the number of setups and whether the part should be turned, milled or both, before a price is attached. A supplier that answers a routing question by naming machines and setups has read the drawing; one that answers with a number and no questions has not. See CNC machining for both processes, sheet metal fabrication when the part is a formed profile rather than a cut one, and metal stamping when volume makes chip removal the wrong economics. Send a model and get a routing review Scope and sources. Process mechanics, tool types, cutting action and the routing rules come from a milling versus turning comparison (workpiece stationary against high-speed rotation, single-point inserts against multi-point cutters, continuous against interrupted cutting, 2 to 4 axes with live tooling against 3, 4 or 5, and the mill-turn route for parts that are neither purely round nor purely prismatic) and from a process selection guide (general dimensions per ISO 2768-1 with critical features toleranced per ASME Y14.5, and the rule that a round part with flats, slots or cross holes belongs on a mill-turn machine). Cost drivers and geometry rules come from a buyer's comparison (machine time, tool changes and multi-axis complexity as milling cost drivers against material diameter, turning speed, depth of cut and secondary operations as turning cost drivers, with concentricity and roundness the turning strength and positional accuracy across faces the milling strength). Market shares come from a worldwide precision machining market study (CNC milling 37.94 percent and CNC turning 28.10 percent of process value in 2025). Cycle time, spindle speed, setup cost and break-even figures come from a turning comparison (25 to 60 seconds against 8 to 20 seconds for comparable small parts, 3,000 to 6,000 against 6,000 to 12,000 revolutions per minute, USD 200 to 600 against USD 400 to 1,200 setup, USD 60 to 120 against USD 90 to 180 per machine hour, break-even around 3,000 parts a year). Figures are planning ranges from published sources and not a quotation; confirm them against your own geometry, material and volume.

The short answer Plan on 3 to 7 working days for a prototype, 7 to 10 days for a small batch and 14 to 30 business days for production, measured from an approved drawing. Add 2 to 4 days for a specialty alloy and 2 to 5 days if the part needs finishing, then 2 to 7 days by air. A lead time is a queue, not a machining speed Buyers usually compare lead times as if they measured how fast a supplier cuts metal. They mostly do not. A typical small machined part spends 8 to 20 seconds on a Swiss-type lathe or 25 to 60 seconds on a conventional one, so the cutting time for a 500-part order is measured in hours. The days come from everything around the cut: the design review, the material release, the first article, the finishing queue and the inspection paperwork. That is why two suppliers quoting the same number can be describing completely different schedules, and why the useful question is which stages the number covers. Read the note, not just the bar. Each row is a midpoint of a published range, and the range is what you will actually get. Order typePublished rangeWhat sits inside the range Prototype, 1 to 5 parts3 to 7 working days, or 24 to 72 hours for a simple part in stock materialProgramming, one setup, first article and packing Small batch, 10 to 1007 to 10 business daysBoth added on top of any finishing or specialty material Production, 100 to 10,00014 to 30 business days, with complex 5-axis work toward the upper endMultiple setups, in-process inspection and phased packing Expedited service24 to 48 hours is quoted for simple parts in stock materialPriority scheduling at a rush premium of roughly 20 to 50 percent The table also explains the most common disappointment in sourcing. A quote of seven days usually means seven days to production readiness, not seven days to your dock. Freight, export packing, customs paperwork and the final delivery leg are frequently excluded, and a number that ends at factory collection is a different commitment from order to door. What the four stages actually consume Working through the stages one at a time is the only reliable way to predict a date, because each one has a different failure mode. Stage two and stage four are the ones buyers forget to schedule. Both are queues you join rather than work you buy. The design review is short if the drawing is complete and long if it is not. A manufacturability check takes roughly 24 to 48 hours, and it exists to find missing dimensions, unclear datum references and finish callouts that need clarification. When the file goes back to you for a revision, the schedule stops until the revised requirement is approved, and that pause is entirely avoidable with a complete release package. Material release is fast for stock grades and slow for everything else: aluminium 6061, 304 stainless and brass are normally on the shelf, while titanium, nickel alloys and PEEK may need procurement that adds 2 to 4 days, and a mill certificate requirement can extend it further. Machining begins with a first article, which is the point where the operation is proven rather than merely quoted. Finishing runs to its own queue: anodising, plating and coating add 2 to 5 days in normal planning and longer in peak periods, because the finisher is a separate process step with its own schedule rather than a station on the machine. What adds days, and what does not The useful distinction is between delays that come from the part and delays that come from the paperwork. A more complex part genuinely costs more days; a better-organised file costs none. Everything on this chart is a number of days added to the machining schedule. Nothing here is cutting time. Geometry, not quantity, is the main driver. A simple three-axis part in stock aluminium is the fastest thing a shop can make. Multiple setups, five-axis toolpaths, thin walls and deep pockets each add programming, fixturing and proving time, and five-axis work pushes an order toward the top of the four-week band. Material availability is binary. A stock grade adds nothing; a specialty alloy adds 2 to 4 days of procurement before the machine is even scheduled. Finishing is a separate queue. Two to five days in normal conditions, more at peak, and every additional finish step is another queue rather than another pass. Inspection depth is a choice you make. A standard dimensional check is included. A formal first article inspection report or an additional metrology package adds one to two days, and it is worth deciding at quotation whether your contract needs it. Clarification loops are the avoidable ones. An incomplete drawing, an unnamed material grade or an unconfirmed finish turns a two-day review into a week, and the schedule pauses while it is resolved. The one fixed date in the Chinese calendar Chinese New Year is the single seasonal event that reliably moves a schedule, and it is worth planning around rather than discovering. The public holiday runs roughly seven to nine days, but the disruption is wider: many factories scale down three to four weeks beforehand, close for ten to fifteen days, and take further weeks to return to full output. The date moves each year, falling between late January and late February. Orders that must land in the first quarter are best released with a schedule that separates machining, finishing, inspection and shipping, so that a queue in one of them does not consume the whole buffer. Compressing a lead time honestly Four actions actually shorten a schedule, and none of them shortens inspection or finishing. Release final drawings rather than work-in-progress files, so the design review is approved once. Name the material grade on the drawing, so procurement can confirm stock without a clarification loop. Confirm the surface finish at quotation, so finishing time enters the plan from the start. State the required date and the shipping mode together, so the factory can sequence your order against a real commitment rather than against a guess. What does not work is asking for a shorter lead time without removing anything. Expedited service is real and is quoted at roughly 20 to 50 percent premium for simple parts in stock material, but it buys priority, not capacity. It cannot compress an alloy procurement queue, a finishing backlog or a first-article approval. Where a short lead time is the wrong thing to buy Three situations call for a slower, better-planned order. First, when the part is a first article on a new design: a rushed first article usually becomes a rushed revision, and the total elapsed time is longer than one careful pass. Second, when the tolerance is tight and the material is difficult: titanium and nickel alloys need stable cutting conditions, and compressing that stage trades scrap risk for days. Third, when the order is the first from a new supplier: a trial lot at normal lead time is the cheapest way to find out whether the supplier can repeat, and it is a poor place to buy speed. How to brief for a predictable date Send the STEP model and a toleranced drawing, the material grade with its condition, the quantity with an annual forecast, the finish per surface, the inspection documentation your contract requires, and the date the parts are genuinely needed. Those items let a supplier sequence design review, material, machining, finishing and shipping against a real date instead of quoting a generic range. See CNC machining for what the process can hold, surface finishing for the queue that most often moves the date, and become our partners for how suppliers are onboarded into our network. Send a drawing and get a dated plan Scope and sources. Stage-by-stage lead time ranges, the effect of specialty alloys and finishing, and the Chinese New Year window come from a China CNC machining lead time study (prototype in standard material 3 to 7 days with 2 to 5 days added for finishing, 2 to 4 days added for a specialty alloy, standard production 2 to 4 weeks with complex five-axis work toward the four-week end, DFM review 24 to 48 hours, finishing 2 to 5 days, express 2 to 4 days, air freight 3 to 7 days, sea freight 3 to 5 weeks, and a Chinese New Year holiday of roughly 7 to 9 days with a wider 10 to 15 day shutdown and a three to four week ramp-down). Order-type ranges also draw on a China sourcing process description (design and DFM 24 to 48 hours, prototyping 3 to 5 days for 1 to 5 parts, low-volume production 2 to 3 weeks for 100 to 5,000 parts, with in-process checks every 10 to 15 parts and CMM final inspection) and on a lead time guide (prototyping 1 to 50 parts in 3 to 7 business days, low-volume 50 to 1,000 parts in 1 to 3 weeks, expedited service as fast as 24 to 48 hours for simple parts in stock material, post-processing adding 2 to 4 days, and a rush premium in the region of 20 to 50 percent). Machining and setup time used for context come from a turning comparison (25 to 60 seconds per small part on a conventional lathe against 8 to 20 seconds on a Swiss-type machine). Figures are planning ranges from published sources and not a quotation or a delivery commitment; confirm them against your drawing, material, finishing specification and the date you actually need.

The short answer Aluminium 6061 and 7075, stainless 303, 304 and 316, carbon and alloy steel, titanium Ti-6Al-4V, brass C360 and copper C110 cover most metal work; POM, nylon, PEEK and ABS cover most plastic work. Material and machinability together drive 35 to 60 percent of part cost. Machinability, not price per kilo, drives the part cost Two materials can cost the same per kilogram and produce part prices three times apart, because the part price is set by how long the cutter stays in the material, how often the tool is replaced, and how much scrap the process generates. That is what a machinability index measures, and it is why material choice and machining cost cannot be separated. The reference point in almost every published index is free-cutting brass C360, rated 100, which is the easiest common metal to cut: chips break cleanly, tool life is long and the surface comes off smooth. Aluminium 6061 sits near 90, aluminium 7075 near 70 because its higher strength slows the cut, stainless 304 around 45 and 316L around 40 because austenitic stainless work-hardens as it is cut, and titanium Ti-6Al-4V around 22 because it generates heat, wears tools and springs back. On the plastic side, POM is roughly comparable with brass at about 95, nylon around 80, and PEEK about 55. Broad indices put anything below 40 out of 100 in the band that carries an extra 75 to 150 percent of machining cost over a free-cutting grade. The index is normalised to brass, so metals and plastics can share one scale. It measures how willingly the material cuts, not how strong it is. Two practical consequences follow. First, material selection alone accounts for up to 35 to 60 percent of total manufacturing cost and cycle time, which is why the material decision is a cost decision rather than a specification formality. Second, a lower-grade but adequate material is often the largest single saving available on a part, larger than any negotiation on machine rate. Six families, and the grade inside each that is usually right Published capability data is usually organised around a small number of families, and within each family there is a default grade that suits most work. Starting from the family and then refining to a grade is faster and more reliable than trying to compare twenty materials at once. Read the last column before you commit. Every family has a constraint that shows up in production rather than in a datasheet. FamilyTypical gradesTensile strengthMachinabilityWhere it fits Aluminium6061-T6, 7075-T6310 MPa, 572 MPaExcellent, goodEnclosures, brackets, frames, prototypes Stainless steel303, 304, 316, 17-4 PH515 to 1,170 MPaFair to moderateFood, medical, marine, chemical duty Carbon and alloy steel1045, 4140, 4340570 to 650 MPaGood to fairShafts, gears, jigs, structural parts TitaniumGrade 2, Ti-6Al-4V344 MPa, 950 MPaPoorAerospace, medical, corrosive service Copper alloysC360 brass, C110 copper385 MPa, 220 MPaExcellent, fairFittings, connectors, bus bars, heat sinks Engineering plasticsPOM, nylon, PEEK, ABS70 to 100 MPaGood to excellentInsulators, gears, bushings, wear parts The metals, grade by grade Aluminium 6061-T6 is the default for most machined metal work: 310 MPa tensile, excellent machinability, good corrosion resistance when anodised and the cheapest route to a precise part. It is soft, so thin walls deflect and deep threads strip, and both are design constraints rather than machining problems. Aluminium 7075-T6 is nearly twice as strong at 572 MPa and is the aerospace standard for high-load parts, but it costs about 30 to 40 percent more, welds poorly, is less corrosion resistant and anodises to a duller, streakier finish. The rule is to default to 6061 and move to 7075 only when the mechanical load demands it. Stainless steel is a family of trade-offs rather than one material. Type 303 is the free-machining grade and is the right choice for high-volume fittings and fasteners, Type 304 is the general corrosion-resistance grade for food equipment and welded assemblies at 515 MPa, Type 316 raises chemical and marine resistance at 579 MPa and roughly one extra cost tier, and 17-4 PH is precipitation hardening and reaches 1,170 MPa, which puts it in aerospace and oil and gas work. All of them work-harden, so the cutting strategy must take a real depth of cut rather than rubbing the surface, and all of them need coated carbide tooling. Carbon and alloy steels are the value option while corrosion is controlled by coating. Grade 1045 is a straightforward medium-carbon shaft and gear steel, and 4140 and 4340 add alloying for toughness and wear resistance at the cost of slower cutting and heavier tooling. Titanium is the opposite end of the value spectrum: Ti-6Al-4V combines 950 MPa tensile with a density of 4.43 grams per cubic centimetre and excellent corrosion resistance, and in exchange it conducts heat poorly, work-hardens quickly and demands low cutting speeds, which is why titanium parts cost several times an equivalent aluminium part. Copper alloys split by function: C360 brass is the reference free-cutting material and is used for fittings, connectors and decorative work, while C110 copper is specified for electrical and thermal conductivity and machines gummily, raising a burr that has to be controlled. The plastics question, and when PEEK is justified Engineering plastics machine cleanly and are the right answer when the part must insulate, run without lubrication, resist chemicals or stay light. POM is the most machinable plastic, with high stiffness, low friction and excellent dimensional stability, which makes it the default for gears, bushings and fixtures. Nylon 6/6 is tougher and more impact resistant than POM but absorbs moisture and swells, so tight tolerances on a nylon part are a risk rather than a specification. ABS is an economical general-purpose material for housings and prototypes with no load duty. PEEK is the material most often over-specified. It is genuinely necessary when the part sees sustained heat above roughly 150 degrees Celsius, aggressive chemicals, or a medical or semiconductor environment that requires the combination, and it holds up to about 250 degrees Celsius continuously. Outside those cases it costs roughly twenty to fifty times the material cost of a sane alternative, and it is harder to machine than POM because it needs sharp tooling and controlled temperatures. Plastics also move more than metals with moisture and heat, so the correct approach is to tolerance only the features whose function needs it. The four questions that decide the material The published selection guides converge on the same sequence of questions, and answering them in order avoids most specification errors. The first yes normally settles the family, and the grade then follows from cost and availability. Work top to bottom. The later questions only matter if the earlier ones have not already decided the family. Will it see weather, salt, chemicals or food contact? If yes, the family is 316 stainless, titanium or PEEK. If no, aluminium or steel is still on the table. Is weight critical, or is load the driver? For weight, 7075 aluminium is the inexpensive answer and titanium the best strength-to-weight answer. If neither matters, any metal fits. Must it conduct electricity, or must it insulate? Conductivity points to copper, brass or aluminium and often overrides everything else for electrical parts; insulation points to a plastic. What is the cheapest grade that passes the first three? That is usually 6061 aluminium, 1045 steel plated, or POM, and it is the answer unless a functional requirement rules it out. Over-specifying is the most common material mistake Three failure patterns show up repeatedly. The first is specifying titanium or PEEK for a part whose actual duty is indoor and light, which multiplies cost for a property nothing in the application uses. The second is combining a high-grade material with an ultra-tight tolerance because both feel like quality: every extra decimal place of tolerance adds inspection time and scrap risk, and the two cost premiums compound rather than overlap. The third is ignoring geometry limits that the material imposes: thin walls in aluminium deflect, deep threads strip, long slender parts deflect under cutting force above a length-to-diameter ratio of roughly 4 to 1, and a material that is hard to cut does not become easier because the drawing calls for it. A practical rule is to write the function next to the material on the drawing. A note such as 316L because the part sees washdown chemicals is worth more to the supplier than a bare grade callout, because it lets the engineer confirm the real requirement rather than guess at it. What to send with a material callout Send the STEP model and a toleranced drawing, the material grade and its condition, the quantity with an annual forecast, the finish per surface, and a note naming the environment and the functional features. Those five items let an engineer confirm whether the specified grade is sufficient, whether a cheaper grade would do the same job, and which cutting strategy the material requires before a price is fixed. See CNC machining for how each family behaves in the cut, surface finishing for what can be applied after machining, and injection moulding when the same part in plastic at volume belongs in a mould rather than on a machine. Send a drawing and get a material review Scope and sources. Machinability ratings, tensile strengths, cost tiers and the share of part cost driven by material come from a CNC material selection guide (material choice influencing 35 to 60 percent of total cost and cycle time, aluminium 6061-T6 at 310 MPa and 9 out of 10 machinability, 7075-T6 at 572 MPa and 8 out of 10, stainless 303 at 620 MPa and 8 out of 10, 304 at 515 MPa and 6 out of 10, 316 at 579 MPa and 5 out of 10, 17-4 PH at 1,170 MPa, titanium Grade 2 at 344 MPa and 3 out of 10, Ti-6Al-4V at 950 MPa and 2 out of 10, brass C360 at 385 MPa and 10 out of 10, copper C110 at 220 MPa, PEEK at 100 MPa, POM at 70 MPa and nylon at 80 MPa, with grades below 40 out of 100 carrying a 75 to 150 percent machining premium, aluminium cutting 30 to 50 percent faster than steel, and each extra decimal place of tolerance adding 20 to 40 percent to cost). A second machinability scale expressed against brass as the 100 reference, plus grade-level detail, comes from a CNC milling material guide (aluminium 6061-T6 at 90 percent, 7075-T6 at 70 percent, stainless 304 at 45 percent, 316L at 40 percent, Ti-6Al-4V at 22 percent, brass C360 as the 100 percent reference, POM at 95 percent and PEEK at 55 percent, with 7075 costing about 30 to 40 percent more than 6061 and PEEK holding continuous service to about 250 degrees Celsius). Selection sequence, cost tiers and the PEEK rule of thumb come from a material selection matrix (a five-question sequence led by environment, then weight, then conductivity, then cost, with PEEK justified only above roughly 150 degrees Celsius or in aggressive or clean-room service, where it otherwise costs about twenty to fifty times a sane alternative). Tensile, yield, hardness and density values used in the grade notes also draw on a precision machined parts material table (6061-T6 at 310 MPa and 2.70 grams per cubic centimetre, 7075-T6 at 570 MPa, 1045 at 570 to 650 MPa, 304 at 520 to 620 MPa, 316 at 515 to 620 MPa, brass C360 at 400 to 450 MPa, bronze C932 at 275 to 345 MPa, copper C110 at 210 to 240 MPa and Ti-6Al-4V at 900 to 1,000 MPa and 4.43 grams per cubic centimetre, with a length-to-diameter limit of about 4 to 1). Figures are planning ranges and typical published values, not a quotation and not a certification of any mill lot; confirm the grade, condition and paperwork against your own specification.

The short answer Treat an OEM award as a gated programme, not a search. Sign the NDA, place a paid sample order of 5 to 20 pieces, clear first article inspection and a written quality agreement, then run a pilot lot at 10 to 30 percent of annual volume. Budget 4 to 8 weeks from first contact to a qualified supplier. An OEM award is a programme, not a search The question most buyers ask is which supplier to pick. That is the wrong first question for an OEM project, because the risk sits after the award rather than in the shortlisting. A supplier that passes every capability check can still ship the wrong hardness, the wrong finish or a bore that is 0.15 mm out of tolerance in month seven, and the usual cause is not a bad factory but a good factory that was never briefed on what "good" means in writing. Choosing a supplier is a decision; onboarding one is a process, and the process is what determines whether the second order matches the first. The published supplier-qualification guidance puts the whole exercise on a timeline. From first contact to qualified status is normally four to eight weeks for general commercial work, stretched to eight to sixteen weeks for regulated programmes where the documentation and audit load is heavier. That window is not the supplier being slow; it is the work of proving that a process can repeat, and it is worth mapping before the first purchase order is raised. Each gate has a written output. The value is not the ceremony; it is that the factory's habits are set early and scale afterwards. Gate one: the NDA, the drawing review and a paid sample The first thirty days are documentation, engineering alignment and a small physical proof. Begin with a mutual NDA, because a machine shop receives the intellectual property that makes the part worth making; the standard practice is a signed agreement before any design file is opened, secure file transfer rather than email attachments, and role-limited access so that only engineers who need the model can open it. Ask at this stage how the supplier stores tooling and whether your drawings are used for any other customer's part, because that answer is easier to give before the contract than after it. Send the full release package at once rather than in pieces. That means a 2D drawing with GD&T, the 3D model as STEP or IGES, the material grade with its temper and condition, the surface finish specification as an Ra value or a plating standard, the annual volume and release pattern, the packaging and labelling requirements, and your inspection expectations. Piecemeal drawings create piecemeal quotes, and a supplier that returns a price with no questions at all has guessed rather than understood. Good design-for-manufacture feedback names the tolerance-versus-process reality, the feature accessibility, the material availability and the deburr sequence. If a grade carries a three to five week lead time in the local market, you want to hear it now, not at the first article. The sample order is the first real proof and should be paid, small and deliberate. Five to twenty pieces is the working range. A paid order puts it in the production queue behind a real commercial commitment and tells you how the supplier behaves when the part is not free. What you are buying at this gate is evidence that the supplier can read your drawing, not evidence that it can run a thousand an hour. Gate two: first article inspection and the quality agreement Days thirty-one to sixty are where the project is actually won or lost. A first article inspection is a formal, documented verification that the process can produce parts conforming to every dimension, tolerance and specification on the drawing. The package normally contents a balloon drawing on which every characteristic is numbered, a dimensional inspection report covering those numbered characteristics, and the material test records. It is run on one to five parts from the first production run, and for any regulated or production programme it is not optional; a supplier that cannot produce a ballooned report on request is telling you something about how it works. Alongside the first article sits the quality agreement, and this is the document that most first-time buyers skip. It should state the acceptable quality level, the inspection routine, the measurement equipment and its calibration status, the sampling plan, and above all the deviation process: who is told, in what form, and within what time, when a characteristic drifts. Writing it at gate two is cheap. Writing it after a bad lot has reached your customer is a negotiation you have already lost, because by then the question is not what the process should do but who pays for the recall. GateWritten outputWho signs itIf it is skipped NDA and IPMutual NDA, secure transfer, role-limited accessLegal, both sidesDrawings reused or leaked Sample order5 to 20 paid pieces, inspected on receiptBuyer engineeringThe process was never proven First articleBalloon drawing, dimensional report, material certsQuality, both sidesDimensions go unchecked into series Quality agreementAQL, sampling plan, deviation process, calibration recordsQuality managersNo formal acceptance criteria exist Pilot lot10 to 30 percent of annual volume, full flowProgramme managerScale-up surprise in week twelve ReleaseChange control, delivery scorecard, escalation routeBoth programme leadsA silent engineering change ships The measurement side belongs in this gate too. Ask to see the calibration status of the instruments used on your features, and ask how the supplier verifies a claimed tolerance. A shop that promises a tight figure without the metrology to prove it is selling a number rather than a capability, and the difference only becomes visible when the second lot arrives. Gate three: the pilot lot and release to production Days sixty-one to ninety are a pilot production run, and the target is a meaningful fraction of real volume rather than a cosmetic extra. Running ten to thirty percent of annual usage exercises the fixtures, the tooling, the inspection routine and the packing exactly as they will run in series, and it does so while your leverage is still high and your account is still small enough to get attention. Suppliers who skip this gate are the ones that surprise buyers in month seven, because nothing about a five-piece sample tests what happens when a fixture wears, a tool dulls, or a machine is shared across six jobs. After the pilot lot the programme moves to release, and the two things that keep it stable are change control and a delivery scorecard. Any change to a drawing, a process, a material source or a machine must be notified and re-approved rather than absorbed silently. Re-qualification, when it comes, is normally triggered by one of five events: a quality escape that reached your customer, a change of manufacturing location or key equipment, a lapse in the supplier quality documents, delivery performance dropping below about ninety percent on time for two consecutive months, or a gap of twelve months or more with no orders. Naming those triggers in the agreement is how a buyer keeps a long relationship honest without re-auditing every quarter. The bars are the same measure, so the rows compare directly. Complex or regulated programmes sit at the long end, not because the shop is slow but because the proof takes longer. What has to be agreed in writing before you scale Six items settle most of the disputes that OEM projects generate, and all six are cheap to write down before volume. The drawing revision you are buying. A controlled revision number on the drawing, the model and the purchase order, so a mismatch is caught at the quote rather than in the goods-in area. The acceptance level. A stated AQL or key-characteristic list, so "good" has a definition both sides can measure against. The deviation route. Who is informed, in what form, within what time, and who owns the disposition of the non-conforming parts. The inspection package. Whether you receive a standard dimensional check, a full report, material test records, or a formal first article file, and whether that is included or quoted separately. Capacity and scheduling. A realistic monthly ceiling and the notice required to change it, so a demand spike does not become a delivery failure. The escalation path. A named owner on each side and a response time, so a problem has a route that does not depend on remembering who to email. Read the last column if you are short of time. Every failure mode listed there is cheaper to prevent at the gate than to repair in production. Where this route is the wrong one The gated programme is designed for a part you intend to buy repeatedly, and it is the wrong tool in four situations. It is over-engineered for a single one-off prototype, where the right control is to inspect the part yourself on receipt rather than to write a quality agreement. It is unnecessary when the part is a catalogue item bought through distribution, where the supplier's process is fixed and your input is a part number. It is misleading when the only reason to qualify a second source is price, because a second source that has not been through the same gates carries a different risk profile regardless of its rate. And it is wasted when the buyer cannot commit to a volume forecast, because the gates are there to protect a repeat relationship and there is nothing to protect. Two more limits are worth stating plainly. First, nothing here substitutes for your own engineering judgement on the drawing; a supplier can only manufacture to a specification, and an ambiguous one will be resolved by somebody's interpretation. Second, the weeks quoted are planning ranges drawn from published sourcing practice rather than a schedule for any particular shop, so confirm them against the material, the finishing and the volume on your part before you commit a launch date to anyone. How to brief an OEM CNC project Send a controlled drawing revision with GD&T, the STEP model, the material grade and condition, the annual volume and release pattern, the finish per surface, the inspection package your contract requires, and the named person who owns the programme on your side. Those seven items let a supplier plan the gates, the fixtures, the inspection routine and the capacity before a price is fixed, and they let you compare two suppliers on the same basis instead of on two different guesses. See CNC machining for what the process itself can hold, become our partners for how suppliers enter our manufacturing network, and surface finishing for the queue that most often moves a delivery date. Send a drawing and get an OEM landing plan Scope and sources. The gated onboarding model, the ninety-day split, first-article contents and the re-qualification triggers come from a new-supplier onboarding guide (days 1 to 30 for the NDA, DFM and a paid sample of 5 to 20 pieces; days 31 to 60 for first article inspection with dimensional reports, material test records and a signed quality agreement; days 61 to 90 for a pilot run at 10 to 30 percent of annual volume, then release) and from a supplier qualification guide (4 to 8 weeks from first contact to qualified status, 8 to 16 weeks for regulated programmes, first article inspection on 1 to 5 parts from the first production run, two qualified suppliers per critical process as a common baseline, and the five re-qualification triggers). The document package, design-for-manufacture feedback and IP handling come from the same onboarding guide plus a China OEM sourcing guide (prototyping 5 to 12 days, low-volume production 2 to 4 weeks, high-volume 4 to 8 weeks, NDAs before design review with secure transfer and siloed engineering networks, and the note that some grades carry 3 to 5 week lead times). Payment and milestone practice, first-article approval timing and IP protection come from a contract machining partner page (production typically 30 percent at purchase order with the balance at shipment, tooling milestone-based at 40 / 30 / 30 around the first article, standard production 3 to 6 weeks after first-article approval, and drawings held under a mutual NDA) and from a drawing-to-delivery partner page (NDAs signed up front, encrypted transfer, restricted facility access, DFM and engineering alignment over 1 to 2 weeks, prototyping and first article inspection over another 2 to 4 weeks, and estimated-annual-usage-based agreements in place of hard minimums). Figures are planning ranges from published sources and not a quotation or a delivery commitment; confirm them against your own drawing, material, finishing specification and volume.

The short answer Precision CNC machining is defined by three measurable things: tolerances of 0.025 mm or tighter against 0.1 to 0.2 mm for general work, surface finish down to Ra 0.4 micrometres, and repeatability proven by Cpk 1.33 or above. It cuts without tooling and suits almost any metal. Precision is three measurable numbers, not one "Precision machining" is a phrase that suppliers use loosely and buyers accept loosely, which is why two shops can both claim it and deliver parts three cost tiers apart. In practice it resolves into three parameters that can each be measured. The first is tolerance capability: how tight a dimensional callout the process can hold on a sustained basis, not on one lucky part. The second is surface finish, expressed as an Ra figure, because on a sealing face or a sliding surface the finish carries the function rather than the size. The third is repeatability, expressed statistically, because a capability that appears on the first article and disappears by the thousandth part is not a capability at all. A useful threshold sits at plus or minus 0.025 mm, which is one thousandth of an inch. General machining typically operates in the 0.1 to 0.2 mm band. A shop that consistently holds 0.025 mm or tighter, and can show the finish and the capability statistic to go with it, is doing precision work by any reasonable definition. Precision is not, however, a property of the machine alone; it is the combined output of machine accuracy, tooling quality, programming, workholding, thermal control and measurement, and removing any one of those degrades the result. Every band is the same measurement in millimetres, so the rows compare directly. A narrower band further left is the more capable process. The tolerance a CNC shop can hold Achievable tolerance is not a fixed number. It moves with the material, the process and the setup, and published capability tables show the same machine holding different figures depending on what it is cutting and how many times the part is re-clamped. Process or setupStandard tolerancePrecision toleranceVerified with metrology 3-axis milling, aluminium0.05 mm0.02 mm0.01 mm 5-axis milling, steel0.05 mm0.02 mm0.01 mm Turning, brass or steel0.025 mm0.01 mm0.005 mm Swiss-type turning, slender0.01 mm0.005 mm0.001 mm Precision grinding0.005 mm0.002 mm0.001 mm Two things about that table matter more than the numbers themselves. Tighter tolerance is exponentially, not linearly, more expensive: moving from 0.05 mm to 0.01 mm might add thirty percent to cycle time because a finishing pass and a settled machine are required, while moving from 0.01 mm to 0.001 mm might double it, because tool wear, thermal growth, fixturing force and spindle runout all have to be controlled at once. And a tolerance is verified rather than claimed. A shop that promises a micron-level figure without the coordinate measuring capability to prove it is quoting a catalogue rather than a process, so the honest question is not what the machine can do but what the shop can measure. Repeatability: the advantage that shows up at volume The advantage that separates CNC from manual or semi-manual work is not the best part a skilled operator can produce; it is that once the program is proven, every part follows it. Manual machining depends on an operator reading a dial, and the spread between operators and between shifts shows up as assembly variation. CNC machines with closed-loop feedback monitor and correct tool position continuously, so the thousandth part is made to the same program as the first. Repeatability is proved with a capability statistic rather than with a single sample. A Cpk of 1.33 or above means the natural spread of the process sits inside the tolerance band with margin, which corresponds statistically to roughly sixty-six non-conforming parts per million; demanding work such as high-pressure hydraulic components commonly asks for 1.67. This is the number to request when a part must assemble without hand-fitting, because it is measured on your feature, in your material, at your batch size, rather than on a machine specification sheet. Read the last column first. A parameter you cannot see proved is a parameter you are taking on trust. Material, geometry and speed advantages CNC machining accommodates virtually every engineering material, which matters because it means a prototype, a bridge batch and a production part can all be made in the same grade, so the material properties you validated are the properties you ship. Aluminium, stainless, carbon and alloy steel, titanium, brass and copper; the engineering plastics POM, nylon, PEEK, PTFE, polycarbonate and ABS; and composites and specialty grades all cut on the same equipment with different tools and parameters. The second freedom is geometric. Multi-axis machines reach undercuts, deep cavities, compound angles, thin walls and features that can only be approached from several directions, and they do so directly from a CAD model with no mould, die or pattern. The absence of tooling is what makes CNC economical at low volume, because a process that needs a mould has to spread that mould across thousands of parts before it competes. It is also what makes design iteration cheap: a geometry change is a program change, not a tooling change. The speed advantage is real and often understated. Eliminating repeated clamping and allowing unattended running has been reported to cut delivery cycles by more than half against traditional methods, and the finishing operations that manual work needs are largely built into the program. Optimised toolpaths also raise material utilisation, which reduces the raw stock that gets turned into chips rather than into part. The trade-off is that a tight tolerance is not free, and buying one you do not need is one of the most common ways to pay more for nothing. Applying a precision callout to a cosmetic feature multiplies inspection time and scrap risk while changing nothing that functions. The disciplined approach is to tolerance the features that carry load, sealing, fit or alignment, leave the rest at the general tolerance block, and let the supplier tell you which callouts are driving the price. Machine accuracy sets the ceiling. Tooling, programming and thermal control decide how much of that ceiling appears on a production batch. The four things precision actually depends on When a part arrives out of tolerance, the cause is almost always one of four things rather than a mystery. Machine condition. A lathe or mill is only as good as its current geometry. A worn spindle bearing or a machine that has not been calibrated is not rescued by slowing the feed, and a shop with tight positioning accuracy and closed-loop tool-wear compensation has a different starting point from one without. Tooling and insert grade. A sharp edge with the right nose radius and coating shears the material; a dull insert pushes it, and a diameter that drifts through a batch is usually a tool-life story rather than a machine story. Programming and workholding. Toolpath strategy, feed and speed, step-over and the rigidity of the fixture all set how much the part moves under cutting force. Thin walls deflect and slender parts bow, and no tolerance callout removes that physics. Thermal stability and gauging. A one degree change in shop temperature moves a steel part by about eleven micrometres per metre, so a controlled cell matters above the precision tier. Precision comes from measuring and compensating rather than from hoping, which is why the gauging loop belongs in the same list. Where a precision claim is not worth buying Three situations call for a conversation before an order rather than a tighter number. The first is a tight callout on a feature that no one measures in service, because it adds cost and risk without adding function. The second is a tight dimension spanning a long unsupported length, where the better answer is usually a design change such as a stepped diameter or a shoulder rather than a heroic machining strategy. The third is a plastic part, because POM, nylon and PEEK move with moisture and temperature, so holding a metal-grade tolerance on a plastic part buys a measurement that will not survive the first week in service. Ask instead for the capability statistic on your critical features, the inspection package that proves it, and a plain answer about which callouts are driving the price. A supplier that can explain your part back to you in those terms is demonstrating the expertise that the tolerance number is only standing in for. How to specify for precision without overpaying Send the STEP model and a toleranced drawing with a general tolerance block, explicit callouts only on the functional features, the material grade and condition, the quantity with an annual forecast, the finish per surface, and a note naming which dimensions and which datums actually carry the function. Those items let an engineer tell you honestly whether a callout is routine on their machines, achievable with a changed strategy, or impossible, and they let you buy the precision the part needs rather than the precision a catalogue offers. See CNC machining for the processes themselves, surface finishing for what follows the cut, and SOMI 3D printing when the geometry cannot be cut from solid at all. Send a drawing and get a capability review Scope and sources. The definition of precision as three measurable parameters, the tolerance and finish thresholds and the capability figures come from a precision CNC machining capability guide (precision defined as tolerances of plus or minus 0.025 mm or tighter against 0.1 to 0.2 mm for general machining, surface finish Ra 1.6 micrometres as-machined and Ra 0.4 micrometres finished, repeatability demonstrated by Cpk of 1.33 or higher, and the six elements of machine accuracy, tooling, programming, fixturing, temperature control and measurement) and from a engineering buyer's comparison (tolerances of 0.005 to 0.05 mm, surface finish Ra 0.1 to 3.2 micrometres, high repeatability, short lead time and the prototype-to-production transition without redesign). Process-by-process tolerance tiers come from a CNC capability reference (3-axis and 5-axis milling 0.05 mm standard, 0.02 mm precision, 0.01 mm verified; turning 0.025 mm, 0.01 mm, 0.005 mm; Swiss turning 0.01 mm, 0.005 mm, 0.001 mm; grinding 0.005 mm, 0.002 mm, 0.001 mm, with tighter tolerance adding roughly 30 percent to cycle time from one tier to the next and doubling it at the tightest tier). Repeatability, geometric freedom and the tooling-free route come from a CNC machining overview (tolerances to 0.025 mm or better maintained across a run, closed-loop feedback correcting tool position, parts produced directly from CAD with no tooling investment, and material coverage from aluminium and titanium through to PEEK and PTFE). The Cpk thresholds and the productivity and material-utilisation claims come from a benefits analysis (axis repeatability within plus or minus 0.005 mm with contour accuracy within 0.01 mm, Cpk above 1.5 with high-end work above 1.67, and industrial CNC typically 10 to 30 percent cheaper than casting below 5,000 units) and from a precision milling review (micron-level tolerances of plus or minus 0.01 mm, Cpk of 1.67 or above, delivery cycles reduced by more than 50 percent, and material utilisation improved by over 20 percent through optimised toolpaths). Thermal growth of roughly eleven micrometres per metre per degree Celsius is a standard steel expansion coefficient used here as an order-of-magnitude illustration. Figures are planning ranges from published sources and not a quotation; confirm them against your own drawing, material and volume.

The short answer A CNC quote is six numbers, not one. Machine time carries 30 to 60 percent, material 15 to 40 percent, setup and programming 10 to 25 percent, tooling 5 to 15 percent, finishing 5 to 25 percent and inspection 5 to 10 percent. Quantity then changes the unit price more than any negotiation. A quote is six numbers, not one Buyers often treat a machined part price as a property of the part, when it is actually the sum of six cost components whose shares move with the material, the geometry and the order quantity. Published quoting guides converge on the same structure: machine time as the largest single element at roughly thirty to sixty percent of the total, material at fifteen to forty percent, setup and programming at ten to twenty-five percent, tooling at five to fifteen percent, surface finishing at five to twenty-five percent, and inspection with documentation at five to ten percent. Overhead and profit sit on top of those, typically in a ten to twenty percent band that widens for specialised work. The practical value of knowing the structure is that it tells you which lever moves the price. A negotiation on the machine rate moves one component. A design change that removes a setup or relaxes a tolerance can move three at once, which is why the largest savings in machining are found in the drawing rather than in the purchase-order discussion. A proper quote breaks material, machining and finishing out separately; a single lump sum hides which component is heavy, and comparing two lump sums tells you almost nothing about where the money sits. The ranges are wide on purpose. The split moves with the part, so the useful move is to ask which components dominate on yours. Machine time is the largest single lever Machine time is calculated as the shop's hourly rate multiplied by the cycle time, and because it is the biggest component it is also the first place to look. Cycle time depends on how much material is removed, on how much surface area has to be finished, on how many tool changes the program forces, and above all on how many times the part is re-clamped. A five-axis cycle that takes thirty minutes can cost the same as a three-axis cycle that takes sixty, because the five-axis rate is higher but it eliminates setups. Shop rates vary widely by region and by machine class, and the regional gap is larger than most buyers expect. In the same published 2026 comparison, a three-axis machine runs about thirty-five to fifty US dollars an hour in China and about eighty to one hundred and fifty in the United States, while a five-axis machine runs about sixty to ninety in China against one hundred and fifty to two hundred and fifty in the United States, with Germany and Switzerland above that for micron-level work. A five-axis machine also bills at roughly one and a half to two and a half times the three-axis rate for the machine itself, so a five-axis route has to remove enough setups to justify the premium. Tool changes and positioning moves are the hidden part of the cycle. A common estimation rule is to sum the pure cutting time for every operation and multiply by one and a half to two and a half to allow for tool changes, rapid moves and handling, so a part with twenty minutes of cutting typically bills thirty-five to fifty minutes of machine time. Anything that reduces non-cutting time, such as a fixture that holds the part for two faces, pays back immediately. Material: you pay for the billet, not the part Material cost is driven by the grade, the billet size and the buy-to-fly ratio, which is the ratio of stock bought to part shipped. The grade effect is large and predictable in direction: aluminium 6061 usually sits at a few dollars per kilogram, stainless 304 in a similar range, and titanium Ti-6Al-4V can exceed eighty dollars per kilogram with Inconel 718 above ninety. Because the same part in titanium can cost twenty to fifty times the aluminium price before a single cut is made, the material grade is often the largest single decision a buyer makes. The billett effect is subtler and more controllable. Machined parts consume far more raw material than their finished weight: a simple turned part might use forty to sixty percent of the stock, a part machined from bar or billet fifteen to thirty-five percent, and a complex five-axis part as little as five to fifteen percent, so a one hundred gram aluminium bracket can start as four to six hundred grams of stock. Specifying a billet size that sits close to the part envelope is one of the cheapest savings available: a twenty-five millimetre part cut from a thirty millimetre billet wastes seventeen percent of the material, and trimming the billet to twenty-six millimetres removes almost all of that waste. Setup, tooling and finishing Setup and programming are fixed costs per part design, which means they are painful on a one-off and nearly free at volume. A single prototype might carry eighty to one hundred and fifty dollars of setup, while the same setup spread across a hundred parts adds one or two dollars each. This is the component that explains why the first unit is expensive and the thousandth is not. Every additional setup also adds fifteen to thirty minutes of non-productive time plus its own programming and fixturing, and reducing a part from three setups to two has been reported to cut ten to twenty percent of total cost. Tooling is small on aluminium and can be decisive on titanium or Inconel, where tool life can fall to a tenth of the aluminium figure and ceramic or cubic-boron-nitride inserts cost several times a standard carbide insert. Custom form cutters and special inserts add a one-time charge that belongs in its own line, typically fifty to two hundred dollars per tool. Finishing is the component that most often surprises buyers, because it is quoted per operation and the operations stack: anodising is the inexpensive end, hard anodising and electroless nickel sit in the middle, and a PTFE-impregnated or multi-layer coating can move the part price by a large fraction on its own. Inspection is similar in shape: a standard dimensional check is usually included, while a full coordinate-measuring report, material test records and a formal first-article package add a project-level charge. Each band is a published cost multiple rather than a physical quantity, so the rows share one scale. Read it as the price of precision. How quantity changes the unit cost The single largest variable in the unit price is the order quantity, because setup and tooling are fixed and cycle time is not. A published illustration of a medium-complexity aluminium bracket shows the shape of the curve: about one hundred and eighty to two hundred and fifty dollars for a single piece where setup dominates, forty-five to seventy dollars at ten pieces as the setup splits, eighteen to twenty-eight dollars at a hundred pieces where cycle time takes over, ten to fifteen dollars at a thousand pieces with optimised fixturing and feeds, and six to ten dollars at ten thousand pieces where a dedicated cell and tooling investment finally pay off. The bars are one measure in US dollars, so they compare directly. The steepest part of the curve is at the left, where setup is being divided. The consequence for a buyer is that a small quantity is not a slightly more expensive version of a large one; it is a different economic animal. If the design is still moving, buying a hundred pieces at twenty-three dollars each to test the market is usually better than buying ten at fifty-seven, because the extra cost of the first ninety is small and the information is worth more. If the design is frozen and the volume is real, moving from a thousand to ten thousand is the difference between a general-purpose fixture and a dedicated cell, and that is a supplier conversation rather than a design decision. The five levers you can actually pull Most of a machined part's cost is set before a quote is issued, and the levers below are the ones that move it without touching performance. Pick the right material, not the strongest one. Aluminium 6061 machines about thirty percent faster than 7075 and costs less per kilogram, so defaulting to the strongest grade when a mid-grade would do compounds the difference at volume. Tolerance only what functions. Every additional decimal place of precision multiplies cost; applying a precision callout to a non-functional feature buys inspection time and scrap risk and nothing else. Design for fewer setups. Orient the critical features so they can be reached from one direction, and keep a flat reference surface for stable clamping. Each setup removed takes non-productive time, error and cost out at once. Use standard tooling. Internal corner radii that match stock end-mill sizes avoid custom cutters entirely, while deep narrow cavities force long-reach tools, lighter passes and more tool wear; a four-to-one depth ratio can cost two to three times a two-to-one cavity. Buy in the right quantity. Combining parts into one order amortises the setup across the batch and is usually the largest single saving available without changing the part at all. Where the cheapest quote is the wrong one Three situations should make a buyer slow down rather than sign. The first is a quote that comes in far below the others: a variance of up to about thirty percent between quotes is normal, but a quote sixty percent or more below the market usually indicates something has been assumed, omitted or cut, and the omitted item tends to be inspection or finishing. The second is a lump-sum quote with no breakdown, because it leaves you unable to see whether the saving came from the machine rate or from a tolerance nobody is going to hold. The third is a request for a tolerance the process cannot deliver at the quoted price, where the honest answer is a different process, a different quantity or a different material rather than a lower number. Two limits belong on the same page. The percentages here are planning ranges from published quoting guides and not a rate card for any particular shop, and they will move with the material market, the finishing supply chain and the exchange rate. And the cost structure is not a substitute for a real quote: the useful move is to send a complete drawing and ask which components dominate, because that answer is specific to your part in a way no range can be. Send the STEP model and a toleranced drawing, the material grade and condition, the quantity with an annual forecast so the setup can be amortised realistically, the finish per surface, the inspection package your contract requires, and a note naming which features and which tolerances are functional. Those items let a supplier separate material, machining, finishing and inspection, and let you see which lever to pull. A supplier that answers with a breakdown and a note on what is driving the price is giving you a tool rather than a number. See CNC machining for what the process can hold, surface finishing for the cost that most often surprises buyers, and die casting when the volume is high enough that a mould beats cutting from solid. Send a drawing and get a costed breakdown Scope and sources. The six-component cost structure and the share ranges come from a 2026 CNC cost guide (material typically 15 to 40 percent, machine time 30 to 60 percent, setup and programming 10 to 25 percent, tooling 5 to 15 percent, surface finishing 5 to 25 percent and inspection 5 to 10 percent, with 2026 hourly rates of about 35 to 50 US dollars for three-axis and 60 to 90 for five-axis in China against 80 to 150 and 150 to 250 in the United States, aluminium 6061 at roughly 4 to 7 US dollars per kilogram, stainless 304 at 6 to 10, titanium Ti-6Al-4V above 80 and Inconel 718 above 90, and the bracket cost curve from about 180 to 250 dollars at one piece down to 6 to 10 dollars at ten thousand) and from a cost estimation guide (machine time 45 to 65 percent, material 15 to 30 percent, setup 10 to 20 percent and tooling 5 to 15 percent, hourly rates of 60 to 90 dollars for a three-axis vertical machining centre and 100 to 150 for a five-axis machine, the rule of multiplying pure cutting time by 1.5 to 2.5 for tool changes and positioning, and material utilisation of 40 to 60 percent for a simple turned part against 15 to 35 percent from bar and 5 to 15 percent for a complex five-axis part). Design rules and their cost impact come from a CNC cost optimisation guide (each additional setup adding 15 to 30 minutes of non-productive time, cutting a part from three setups to two saving 10 to 20 percent of total cost, custom tools adding 50 to 200 dollars each, a four-to-one deep cavity costing two to three times a two-to-one cavity, and the tolerance cost multipliers of 1.0 at 0.1 mm, 1.1 to 1.2 at 0.05 mm, 1.3 to 1.5 at 0.025 mm, 1.5 to 2.0 at 0.01 mm and 2.0 to 3.0 at 0.005 mm). The finishing and inspection shares and the material-weight effects come from a cost calculation guide (material 20 to 40 percent, machine time 30 to 50 percent, programming and setup 5 to 20 percent, tooling 5 to 15 percent and surface treatment 10 to 30 percent, with a five-axis rate 1.5 to 2.5 times a three-axis rate) and from a quote breakdown study (material 15 to 30 percent, machining time 30 to 45 percent, fixturing and programming 5 to 12 percent, tooling 3 to 8 percent, fixturing 2 to 6 percent, surface treatment 5 to 12 percent, inspection 3 to 8 percent, packaging 2 to 5 percent and profit 8 to 20 percent). Figures are planning ranges from published sources and not a quotation; confirm them against your own drawing, material, finishing and quantity.

The short answer A precision CNC floor is built from six groups of machines: three-, four- and five-axis mills, turning centres and Swiss-type lathes, grinding machines, wire and sinker EDM, and inspection equipment such as CMMs. Between them they cover tolerances from 0.05 mm down to 0.002 mm on the right part. What a machine list should tell you When a buyer asks what equipment a supplier has, the useful answer is not a list of model numbers. Machines are tools with capability envelopes, and the same model number can sit in a well-maintained shop that holds a tight tolerance and in a neglected one that does not. What matters is which classes are present, what tolerance each class can hold on a sustained basis, what size and shape envelope the shop can reach, and what equipment exists to prove the result. A list of a hundred machines says nothing about any of those things, while a short list matched to the tolerances and shapes your parts need says everything. It also helps to know why the classes exist rather than treating them as interchangeable. A three-axis mill is the workhorse for prismatic parts. A four-axis machine indexes the part around one rotary axis so that features on four sides can be cut without a second fixture. A five-axis machine adds a second rotary axis so the tool can approach from any angle, which is what makes contoured surfaces and multi-face parts possible in one setup. Turning machines do the opposite of milling: the workpiece spins and a single-point tool cuts, which is the natural route for anything built around an axis. Grinding and EDM exist for what cutting tools cannot reach or cannot hold. Inspection equipment exists to prove that any of it worked. Read the tolerance column against the class. A machine class that is present but not used for your tolerance adds nothing. Milling machines: three, four and five axes The three-axis mill moves the tool in X, Y and Z and covers the majority of prismatic work: plates, brackets, housings, pockets, flats, holes and slots. It is the least expensive class per hour and the easiest to program, and for a part whose features are all reachable from one direction it is the correct answer rather than a compromise. Its limitation is that additional faces require repositioning, and each re-clamp introduces a datum shift in the region of 0.02 to 0.05 mm that accumulates across the part. The four-axis machine adds a rotary table, so a part can be indexed to four sides in a single fixture. That improves positioning accuracy between faces and cuts the handling time, and it suits cams, gear blanks and cylindrical housings with features around the circumference. It does not solve undercuts or compound angles, which need a second rotary axis. The five-axis machine adds that second rotary axis, and its value is not simply a tighter number. It is that a part which would need five three-axis setups can be completed in one or two, and every eliminated setup removes both a handling step and a tolerance stack-up. Published comparisons put a well-fixtured five-axis mill at 0.01 mm where a long-reach three-axis setup holds 0.02 mm, and describe thermal warm-up protocols on the rotary axes reducing first-article variation substantially. The counterweight is cost: a five-axis machine bills at roughly one and a half to two and a half times the three-axis rate, so on a part that only has features on one face it buys nothing but a higher rate. Turning machines and Swiss-type lathes A CNC turning centre rotates the workpiece against a stationary single-point tool, which makes it the natural and cheapest route for any body of revolution. It produces shafts, pins, bushings, sleeves, threaded parts and fittings with excellent concentricity, and the strength of the process is precisely the set of dimensions that follow the rotation: diameter, roundness, cylindricity and runout. A conventional turning centre normally handles parts up to a few hundred millimetres in diameter, and with live tooling it can also mill a flat, drill a cross hole or cut a keyway without releasing the part from the chuck. A Swiss-type lathe is a fundamentally different machine rather than a smaller one. The bar stock feeds through a guide bushing set a millimetre or two behind the cutting edge, so the unsupported span never grows and the effective overhang stays constant whatever the part length. That is what lets it hold tight roundness on slender work at length-to-diameter ratios that a chucking lathe cannot manage, and it is why bone screws, micro-valves, contact pins and similar small precise parts are made that way. The class tops out at a small bar diameter, so it is not a substitute for a conventional lathe on larger work; the two are complementary. Grinding, EDM and secondary processes Some features lie outside what a cutting tool can hold, and the machines that reach them are the finishing classes. Grinding uses an abrasive wheel to remove material in very small increments, achieving sub-micron roundness and finishes below Ra 0.2 micrometres on hardened steel. It is slow and expensive per cubic millimetre, so it is almost always a finishing operation rather than a roughing one, and it is reserved for bearing journals, seal faces, gauge blocks and similar features where the drawing calls for a tolerance or a finish that milling cannot deliver. Electrical discharge machining erodes conductive material with controlled sparks through a dielectric fluid, so hardness is irrelevant: a wire EDM machine cuts a hardened tool steel as readily as an annealed one. Wire EDM cuts two-dimensional profiles through a full plate thickness with no cutting force at all, which suits thin walls and sharp internal corners, while sinker EDM burns cavities and undercuts with a shaped electrode. Both are slow, so they belong on features that cannot be produced any other way rather than on general work. Deburring, polishing, heat treatment and coating are secondary operations that usually run outside the machining cell, and each is a separate queue rather than a station on the machine. Every band is the same measurement, so the classes compare directly. The tighter bands sit further left. Inspection equipment: how the numbers are proved The equipment that decides whether a capability claim is real is the metrology, not the machining. A coordinate measuring machine performs full three-dimensional dimensional verification of complex geometry and is the instrument that settles a first article; a two-dimensional vision system handles fast inspection of small precision parts in batch; a surface roughness and hardness tester confirms the finish and the material condition that the function depends on. A final check before packaging, and a traceable inspection record against your drawing, are what turn a claimed tolerance into a demonstrated one. This is the part of the equipment list a buyer should read most carefully. A shop that promises a tight figure without a CMM or equivalent metrology is selling a number rather than a capability, and the gap only becomes visible when the second lot arrives and nobody can show what changed. The right question is not which machines are on the floor but which measurements are made on your features, with what instrument, and against what calibration record. What to ask instead of "what machines do you have" Six questions extract more than any asset list, and each one has a useful answer and a weak one. Which class would you route my part to, and why? A strong answer names a machine class, a setup count and the reason; a weak one names a model number. What tolerance do you hold on that class across a batch, not on a sample? Watch for a capability figure with a batch size behind it rather than a best-case number. What is your size and shape envelope? Ask for the machine travels in the axes that matter, because a part outside the envelope is a different quote or a different supplier. How do you prove the tolerance? The answer should name the instrument, the feature, and the report you will receive. Is the same equipment used for the prototype and the series? If the process changes between the sample you approve and the run you receive, the approval means less than it appears. What is the calibration and maintenance routine? A worn spindle is not fixed by slowing the feed, and the routine is what keeps a machine inside its specification. Work left to right. The inspection stage is what converts a machine list into a repeatable capability. Where a long equipment list misleads a buyer A long asset list can be actively misleading in four ways. First, more machines do not mean more relevant capability; a floor of a hundred three-axis mills cannot hold a five-axis tolerance or reach a five-axis shape. Second, a machine class that is present but rarely used for your tolerance adds nothing, because capability is a function of routine rather than of ownership. Third, the envelope matters as much as the class: a shop that advertises five-axis work but whose travels are half your part size will quote it as a two-operation job, which changes both price and tolerance stack-up. Fourth, and most often missed, the equipment list describes the machining side and stays silent on the measurement and finishing sides, which is where most delivery surprises originate. The limits of this answer are worth stating plainly. The tolerance figures above are published capability ranges for each machine class and not a specification for any particular shop, and the same class can differ by a factor of two between a well-maintained machine and a tired one. The right way to use an equipment list is as a first filter for class and envelope, followed by a conversation about which specific machines would run your specific features and how the result would be measured. Send the STEP model and a toleranced drawing, the material grade and condition, the quantity with an annual forecast, the finish per surface, the size and weight limits your assembly imposes, and a note naming which features are functional. Those items let a supplier route the part to a machine class and tell you whether it needs one operation or three, which is a more useful answer than a floor tour. See CNC machining for how the classes compare in practice, surface finishing for the operations that follow the machine, and become our partners for how a supplier's equipment is reviewed before it enters our network. Send a drawing and get a machine routing Scope and sources. Machine classes, their functions and their capability envelopes come from a CNC machining guide (three-axis milling 0.02 to 0.05 mm for prismatic work, four-axis 0.02 mm for cams and four-face parts, five-axis 0.01 mm with good fixturing for impellers and implants, conventional turning 0.025 mm and Swiss-type turning 0.01 mm standard down to 0.001 mm verified, precision grinding 0.005 mm standard and 0.001 mm verified, and Swiss-type machines as the only sensible process for slender parts above a 10:1 length-to-diameter ratio) and from a machine-type comparison (3-axis 0.01 mm, 4-axis 0.01 mm, 5-axis 0.005 mm, turning 0.005 mm, turn-mill 0.005 mm, wire EDM 0.005 mm and surface grinding 0.002 mm, with the warning that a five-axis machine is not automatically more accurate than a three-axis one and that the accuracy gain comes from eliminating re-clamping). Datum shift per repositioning, five-axis setup consolidation and warm-up effects come from a CNC machine type guide (each re-fixturing adding 0.02 to 0.05 mm of datum shift, a long-reach three-axis setup at 0.02 mm against an optimal five-axis setup at 0.005 mm, ground tolerances of 0.002 mm with roundness under 0.001 mm, and wire EDM accuracy around 0.002 mm with a finish of Ra 0.4 to 0.8 micrometres). Equipment classes, envelope sizes, inspection practice and the CAD-to-packing workflow come from a CNC equipment overview (3-, 4- and 5-axis milling with large-format travel up to 1,500 by 1,200 mm and turning up to 300 mm diameter, a machining tolerance of plus or minus 0.01 mm, wire and sinker EDM for corners and fine detail, and CMM plus two-dimensional vision, surface and hardness testing before packing) and from a precision machining equipment guide (three- and four-axis positioning accuracy of plus or minus 0.0001 to 0.0005 inch, five-axis positioning accuracy of plus or minus 0.0001 inch linear with plus or minus 0.3 arc-second rotary, spindle speeds of 15,000 to 40,000 revolutions per minute, tool capacities of 40 to 120 positions, and wire EDM cutting materials above 65 HRC with zero cutting force). Figures are published capability ranges for each machine class and not a specification for any particular shop; confirm them against your own features, material and volume.