San Diego CNC Precision Guide

• 1
  Why precision is a stack, not a single number
• 2
  Five-axis setup and where the accuracy comes from
• 3
  Tolerance stacking and the datum question
• 4
  Surface finish, heat and the drift nobody measures
• 5
  Inspection, documentation and what to ask for
• 6
  Local supply and offshore production: the real trade
• 7
  When each setup route is the right call

A drawing that says ±0.005 mm is not a description of a machine. It is a description of an outcome, and that outcome comes from a chain: spindle and axis motion, workholding stiffness, cutter deflection, thermal drift, and the measurement method used to confirm the result. Break any one link and the number does not hold, no matter how new the machining center is.

This is the part most quotes skip. A supplier can own simultaneous 5-axis equipment and still miss a tight bore if the fixture lets the part move 0.02 mm under load, or if the datum on the drawing is not the datum used on the machine. The machine sets the ceiling. The setup and the inspection plan decide where you land beneath it.

In our plants we run 16 simultaneous 5-axis machining centers, 12 four-axis mills, 27 three-axis machines and 16 mill-turn centers, 127 high-precision CNC machines in total, across 7,600 m² in three wholly-owned plants. Capacity tells you what shapes are possible. It does not tell you what tolerance is repeatable. Those are separate questions and they deserve separate answers.

So read this San Diego CNC precision guide as a filter. It will not tell you which supplier to pick. It will tell you which claims are checkable and which ones are noise.

A three-axis machine moves the tool in X, Y and Z while the part stays put. Every new face means a new setup, and every setup adds a re-clamping error. A five-axis machine adds two rotary axes, so the tool can reach compound angles, deep pockets and contoured surfaces without the operator touching the part again. One setup, one datum, one error budget.

The gain is not just reach. It is the removal of repositioning error. If a part has four machined faces and three of them are critical to each other, a three-axis route introduces three chances to shift the part. A five-axis route machines them from one orientation. For a housing with a bored bore and a mating face, that difference often matters more than the raw machine tolerance.

Five-axis does not automatically mean tighter. Rotary axes bring their own error: table runout, rotary encoder resolution and the accuracy of the post-processor that converts CAM toolpaths into machine motion. A poorly posted five-axis program can cut a smooth-looking surface that is 0.03 mm off the nominal. The geometry looks right and the gauge disagrees.

Where five-axis earns its cost: impellers, medical instrument bodies, brackets with compound-angle bosses, and any part where the number of setups drives the error stack. Where it does not: a flat plate with two holes, which a three-axis mill will finish faster and cheaper.

Most arguments about precision are really arguments about datums. If the drawing calls out position relative to datum A, and the shop grips the part on a different surface, the inspection result will not match the intent even when the cut is perfect. This is the single most common source of a failed first article.

Stacking compounds it. A position tolerance of 0.05 mm on a hole that sits three features away from its datum absorbs the tolerance of every feature in between. Designers sometimes tighten the hole callout when the real culprit is the chain. Loosening an intermediate dimension often fixes the problem at lower cost than tightening the final one.

There is also a measurement floor to respect. If you specify ±0.005 mm, the shop needs a CMM or a coordinate measurement routine capable of roughly a quarter of that to prove it. A caliper or a pin gauge cannot confirm a number that small. On our side, ±0.005 mm (±0.0002 in) is achievable, and every shipment passes a raw material check, in-process monitoring and a final inspection, with reports issued on request.

Practical rule: specify the tolerance the function needs, name the datum the part actually locates on, and let the shop choose the process. A drawing that says what matters and stays quiet about how to get there will come back cheaper and more accurate.

Finish and tolerance are linked, and not in the direction people expect. A fine finish costs time, and the finishing pass is where small parts sometimes move. As the cutter takes the last 0.05 mm, residual stress in the stock releases and the part relaxes. A bore that gauged 0.004 mm oversize in the morning can read 0.008 mm by the afternoon.

On our machines the finish ranges run from Ra 0.2–0.8 μm for fine work, Ra 0.8–1.6 μm for high-quality functional surfaces and Ra 1.6–3.2 μm as-machined. Titanium, Inconel and PEEK behave differently in each band. Inconel work-hardens and pushes the tool, so a fine finish needs a sharper cutter and a lighter radial step, not just a slower feed.

Thermal drift is the quiet one. A spindle that has run for two hours is not the same machine it was at start-up. Shops that hold ±0.005 mm on a long run either warm up deliberately or measure and compensate. Ask how a shop handles this and you learn more than you learn from a tolerance claim.

Material choice belongs here too. We machine aluminium 6061, 7075 and 6082, stainless 303, 304, 316L and 17-4PH, steel 4130, 4140 and 4340, titanium TC4, brass C36000, and plastics from POM to PEEK. Each one moves differently after the cut. The tolerance you can hold is material-specific and geometry-specific, not a shop-wide constant.

A first article inspection report is worth more than a verbal assurance. It shows the measured value for each controlled dimension, the gauge used, and the ambient conditions if the shop tracks them. On a job with a ±0.005 mm callout, a report without gauge identification is not evidence.

Traceability matters in regulated work. We hold ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022, which cover quality systems, automotive, medical devices and information security respectively. For aerospace-adjacent parts, material certificates and lot traceability should travel with the shipment, not arrive later on request.

Confidentiality is part of the same conversation. Drawings for a new product are the design. Files uploaded through our quote page stay confidential and an NDA is available on request. If your program requires one before you release CAD, ask first; it costs nothing and avoids a rewrite of the release process later.

The practical ask list: measured values for critical dimensions, the gauge or machine used, material certificates, and a clear statement of which dimensions were not measured. The last one is the honest item and the one most suppliers omit.

A San Diego shop and a Dongguan shop can both hold ±0.005 mm. The difference is not capability, it is the shape of the service. Local sourcing shortens the feedback loop when a design is still changing weekly. Offshore production lowers unit cost once the design is frozen and the volume is real.

The trap is assuming you must choose one for the whole program. Prototype iterations often run better close to the design team, where a fixture change can be discussed across a table. Production runs often run better where capacity is deep and the machine is not shared with twelve other jobs.

Volume is the other variable. We run from a single prototype to 10,000+ part runs with no minimum order quantity, so a program can start with one piece and scale without changing supplier. Quotation and free DFM analysis come back within 12 hours, production can start within 24 hours, and parts typically ship in 3–5 days.

Ask any supplier, local or offshore, the same three questions: what tolerance is repeatable on this specific geometry, how will you measure it, and what happens when the first article misses. The answers tell you more than the location.