Judging From the Toshiba Machine Tool Incident: Why Soviet Five-Axis Accuracy Fell Short
This page is for engineers and buyers who want the real mechanical reasons behind the 1987 Toshiba machine tool incident, not the spy-story version. It walks through the accuracy chain that separates a five-axis propeller machining center from a look-alike, then gives you a symptom-to-fix table you can apply to your own parts.
Symptom, likely cause, and what to do about it
Use the left column to match what you see on the machine or the part. The middle column is the mechanical reason. The right column is the first move.
| Symptom | Likely cause | First action |
|---|---|---|
| Blade pitch drifts 0.05 mm over a 2 m cut | Rotary axis growth from spindle and drive heat | Let the machine idle 90 min, then re-zero the rotary table |
| Chord error grows on the last of seven passes | Servo following error on the A/C axes | Cut feed to 60% and check axis tuning on a ballbar |
| Surface shows chatter at the blade root | Weak fixturing on a thin, unsupported wall | Add a tailstock or steady rest, then reduce radial depth |
| Part is round on the table but oval in the fixture | Clamping distortion of a thin-walled blank | Switch to soft jaws or vacuum, measure before and after |
| Five-axis position repeats, but the contour is wrong | Post-processor or RTCP setup error | Verify the tool center point and re-post the toolpath |
| Good first article, drift after 40 parts | Thermal growth in the ballscrew and bed | Warm up per the machine log, then re-probe the datum |
| Tool marks appear only on overhang passes | Tool deflection at long reach | Shorten the gauge length or take two lighter passes |
The takeaway
Five-axis accuracy is a chain, not a specification. Fix the thermal state, the rotary axes, and the fixture before you blame the toolpath, and most of the Toshiba-class problems disappear at any scale.
What the Toshiba machine tool incident actually exposed
In 1987 Toshiba Machine sold several nine-axis, five-link machining centers to the Soviet Union through a Norwegian trading company. The machines were built to mill the seven-blade, high-skew propellers used on strategic submarines. The story usually ends there, as a Cold War export-control case. The engineering point is more interesting: the Soviet Union already had large five-axis machines that could swing the same propeller blank. What it could not do was hold the accuracy on the finished blade.
The gap showed up in three places. The control had to keep the tool tip on a continuously changing normal vector while the table rotated. The rotary axes had to hold angular position within a few arc-seconds under a heavy, unbalanced load. And the bed had to stay dimensionally stable while a 30 kW spindle pushed a 100 mm cutter through a nickel-aluminum-bronze casting for hours.
None of those is a single-part problem. They are a system problem. A machine that is stiff enough at the spindle but sloppy at the rotary table will cut a good roughing pass and a poor finishing pass. That is exactly the pattern a submarine propeller would show: acceptable geometry, unacceptable blade-to-blade variation.
This is still the way five-axis accuracy fails today, on parts far smaller than a propeller. When a shop reports that the first article passes and the tenth does not, the cause is almost never the cutting tool. It is the accuracy chain: control, rotary axes, thermal state, and fixturing, in that order.
- 1Control and RTCPThe controller must keep the tool tip fixed while A and C rotate, within microns.
- 2Rotary axis stiffnessAngular error at the table turns into chord error at the part edge, multiplied by radius.
- 3Thermal stabilityA bed that grows 20 μm over four hours will not repeat on a long finishing pass.
- 4Fixture rigidityThin blades and thin walls deflect under clamping force before the cutter even touches them.
Why five-axis accuracy is a chain, not a spec sheet
A tolerance of ±0.005 mm on a drawing is a claim about the finished part. On a five-axis machine, that number is the product of several independent errors that stack. Position error at the tool tip equals the linear axis error, plus the angular error of each rotary axis times its distance to the cutting point, plus thermal drift, plus deflection under cutting load.
Angular error is the one buyers underestimate. A rotary table that holds 5 arc-seconds sounds fine. At a 400 mm radius from the table center, 5 arc-seconds is about 0.010 mm of chord error at the part edge. Double the radius and you double that error. This is why a machine that holds tolerance on a 100 mm part can miss it badly on a 600 mm part with the same setup.
Thermal behavior follows the same rule. Spindles, drive motors, and ballscrews all heat under load. A machine that is cold at 7 a.m. and warm at noon will produce parts that drift across the run. Shops that fight this either warm the machine for a fixed period before cutting or probe the datum between parts and adjust the offset.
The Toshiba machine tool incident is often framed as a hardware story. It is more accurate to call it a metrology and control story. The Soviets could build a heavy machine. Measuring and compensating it in real time, at that scale, was the harder problem.
- 1Stack the errorsAdd linear, angular, thermal, and load errors before you promise a tolerance.
- 2Watch the radiusAngular error times distance to the cut is the number that bites.
- 3Warm up firstA fixed warm-up routine beats chasing offsets all day.
- 4Probe between partsRe-datum on long runs instead of trusting the first zero.
What to check before you buy a five-axis job
If you are sourcing a contoured part, ask the shop three questions. What is the angular repeatability of the rotary axes? How is the tool center point calibrated, and how often? What is the documented warm-up routine? A shop that can answer those has thought about the chain. A shop that answers only in spindle speed and table size has not.
Machine size matters, but not the way most buyers assume. A 4,000 mm travel machine is not automatically more accurate than a compact one. Long travels add stack-up. The useful question is whether the part fits inside the machine's calibrated volume, where the manufacturer has mapped and compensated the errors.
For the propeller-class geometry in the Toshiba machine tool incident, the part is large, thin, and continuously curved. That combination demands simultaneous five-axis motion, not 3+2 positioning. A 3+2 machine can reach the surface, but it stops and indexes between orientations, which leaves witness marks and adds cycle time on a curved blade.
On smaller work, the same logic applies at a different scale. Impellers, turbine blades, medical implants, and injection-mold cores all share the pattern: curved surfaces that need the tool axis to tilt continuously. If the geometry is prismatic, three-axis milling is cheaper and just as accurate.
- 1Ask for angular repeatabilityArc-seconds, not just degrees, on both rotary axes.
- 2Ask about RTCP calibrationHow the tool tip is kept fixed during rotation, and how often it is checked.
- 3Match the volumeKeep the part inside the calibrated envelope where errors are mapped.
- 4Choose 3+2 only when the part allows itIndexed positioning is fine for prismatic features, not curved blades.
Step by step: isolating a five-axis accuracy problem
Run these in order. Each step tells you whether to keep looking or stop and fix.
- 1Log the machine's thermal stateRecord spindle and bed temperature at the start of the shift. Cut a test part cold, then again after a 90-minute warm-up. If the two parts differ by more than 0.01 mm, thermal drift is your first problem, not the toolpath.
- 2Check the rotary axes with a ballbarA ballbar sweep around the rotary axes shows backlash, squareness, and servo mismatch in one run. Look for a step in the polar plot. A step means mechanical backlash; a smooth oval means servo tuning.
- 3Verify the tool center pointMount a known-length tool and rotate the A and C axes through their range with the tip touching a dial indicator. Any movement over 0.005 mm means the RTCP setup or the post-processor is wrong.
- 4Cut a test part at the real radiusDo not prove out on a small block. Cut at the same distance from the rotary center as the production part. Angular error scales with radius, so a small test part hides the problem you are trying to find.
- 5Measure before and after clampingProbe the blank on the table, then again after the fixture is tightened. A shift over 0.02 mm means clamping distortion. Move to soft jaws, vacuum, or a lower clamp force.
- 6Re-probe the datum between partsOn runs over 20 parts, re-probe and update the work offset every few pieces. This absorbs the slow drift that a warm-up alone will not catch.
- 7Shorten the tool gauge lengthIf chatter appears only on overhang passes, reduce the tool stick-out by 20% and split the pass. Long-reach tools deflect, and no amount of control tuning fixes a bending tool.
Questions engineers ask next
Was the Toshiba machine tool incident really about machine accuracy, or was it about export law?
Both, but the technical half is what still matters to a shop. The machines were capable of a class of work the Soviets could not otherwise do, and the legal case followed from that capability.
The engineering lesson is that buying a capable machine does not transfer the process knowledge. Calibration, thermal control, and metrology have to be built up locally.
Can a three-axis machine hold the same tolerance as a five-axis machine?
Yes, on parts that do not need the tool axis to tilt. For prismatic parts with features reachable from a few fixed directions, three-axis milling is often more rigid and easier to control.
The tolerance a machine holds depends on the error stack, not the axis count. Adding rotary axes adds error sources, so use them only when the geometry requires it.
How much does thermal drift actually move a part?
On a typical vertical machining center, a cold-to-warm shift of 0.01 mm to 0.03 mm over a few hours is normal. On large five-axis machines with long travels, the number can be larger.
A documented warm-up routine and periodic re-probing are the practical controls. There is no way to design the drift out entirely.
What is the practical tolerance limit for five-axis work?
For well-controlled simultaneous five-axis work, ±0.005 mm is achievable on parts within the calibrated volume. Surface finish in the Ra 0.8–1.6 μm range is routine with the right cutter and step-over.
Tighter than that, inspection becomes the bottleneck. You need a CMM and a temperature-controlled room to prove it, not just a good machine.
When should a shop use 3+2 instead of simultaneous five-axis?
Use 3+2 when the features are planar or can be reached from a handful of orientations. It is faster to program, more rigid, and easier to inspect.
Use simultaneous five-axis when the surface is continuously curved and the tool axis must follow the surface normal, as on impeller blades or mold cores.
How do we keep long runs from drifting?
Warm up the machine to a stable temperature before the first part, then re-probe the work offset on a fixed interval. Log the temperature so you can see a trend before it becomes scrap.
For runs over a few hundred parts, schedule a mid-run inspection and a tool change before the wear curve turns up.
Send us the part that keeps drifting
Upload a drawing or a STEP file. We review the geometry, the tolerance stack, and the five-axis setup, then return a quotation and a free DFM analysis within 12 hours.
12-hour quote100% inspectionNo minimum orderNDA on request
More CNC process notes
We publish setup notes, tooling trials and inspection data from the factory floor.
