Method for Detecting Residual Stress in Machined Parts

• 1
  Key takeaways
• 2
  What residual stress actually is and why the method for detecting residual stress matters
• 3
  Choose the method for detecting residual stress by depth and access
• 4
  Where the stress comes from in a CNC process
• 5
  When detection is worth it and when it is not
• 6
  Common errors that ruin good measurements
• 7
  How to run a hole-drilling measurement step by step
• 8
  Method for detecting residual stress: comparison by use case

Residual stress is the stress that remains in a body when no external load and no temperature gradient is acting on it. Two things create it during machining: uneven plastic deformation, and phase change. Cutting pushes metal instead of cleanly shearing it, so the surface layer is stretched while the layer beneath resists. Heat from the tool adds a thermal gradient on top of that.

The result is a self-balancing system. Tension in one zone is matched by compression somewhere else. That balance is why a part can measure perfectly on the CMM at 20 °C and still move a day later. Remove material from one side during a second operation and the balance shifts.

Stress itself is not automatically a defect. Shot peening deliberately puts the surface into compression to delay fatigue cracking. The problem is uncontrolled stress: high tensile stress at a fillet, at a weld toe, or under a ground surface. That is where cracks start.

So the purpose of any method for detecting residual stress is not to prove a part is stress-free. It is to quantify magnitude and direction, track how they change between processes, and decide whether the part is safe or needs a relief step.

• 1
  MagnitudeReported in MPa or ksi. Comparison only works if units and depth match.
• 2
  DirectionPrincipal stress axes. A value without orientation is hard to act on.
• 3
  Depth profileSurface value versus subsurface value. They often have opposite signs.

There is no universal instrument. Each technique trades depth, spatial resolution, and whether the part survives. Start by asking what depth matters. If the failure risk is at the surface, such as a ground bearing journal, you need a near-surface method. If the risk is a thick casting or a weld, you need bulk measurement.

Hole-drilling is the workhorse. You bond a three-element strain rosette on the surface, drill a small blind hole at the rosette center in incremental steps, and record the strain relieved at each step. The standard hole is Ø1.6–2.0 mm to a depth roughly equal to the hole diameter. From the relieved strains you compute the original stress. It is a semi-destructive method, so it belongs on coupons, spare parts, or a sacrificial boss on the casting.

X-ray diffraction measures the spacing of crystal planes. Stress changes that spacing, and the shift in the diffraction peak gives strain. Because X-rays penetrate only 10–30 μm in steel, this is a surface method. It is quick, it works on small features, and it is the usual choice for validating a grinding or turning recipe. Electropolishing in steps extends it into a depth profile.

Ultrasonic testing uses the fact that wave speed in a material depends on stress, a phenomenon called acoustoelasticity. It penetrates the full section and it is genuinely non-destructive. The catch is calibration: you need a reference sample of the same alloy, heat treatment, and grain structure in a known stress state, or the absolute numbers drift.

• 1
  Hole-drillingDepth to about 1× hole diameter. Semi-destructive. Good for thick sections.
• 2
  X-ray diffraction10–30 μm deep. Non-destructive. Best for surfaces and thin layers.
• 3
  UltrasonicFull section. Non-destructive. Needs a matched reference block.
• 4
  Magnetic and BarkhausenFerromagnetic only. Fast comparative screening on the shop floor.

Roughing removes a lot of material fast. A 6061 block hogged down with a 16 mm end mill at 3,000 rpm and a heavy radial engagement leaves a heavily deformed surface layer. That layer is usually in tension. Finish passes are lighter, so they remove some of it, but they also introduce their own deformation.

Heat is the second driver. If the tool edge dulls and the chip load drops, rubbing replaces cutting. Surface temperature climbs, the top layer expands against colder metal below, and on cooling it ends up in tension. You can often see this as a discolored or smeared surface before any instrument confirms it.

Fixturing adds a third component. Clamping a thin wall to a flat plate forces it straight. After unclamping, it springs back and the machined feature is out of position. That is not classical residual stress inside the material, but it behaves the same way on the CMM and engineers usually group it with the same problem.

For asymmetric parts, like a housing with a large pocket on one face, material removal unbalances the internal stress field. One side relaxes more than the other and the part bows. Detecting this early, on a first-article coupon, is cheaper than scrapping a finished batch.

Detection is expensive and slow. Run it when a failure would be costly or dangerous: aerospace brackets, medical implant components, pressure-boundary parts, and any thin-wall geometry where distortion has already bitten you once. In those cases a single coupon measurement can justify a process change.

Skip full measurement when the part is thick, lightly loaded, and has no tight flatness or roundness callout. A 40 mm steel plate machined on all faces will not move enough to matter. Spending a week on diffraction would not change the drawing or the price.

The middle ground is comparative testing. Measure one reference part from a known-good run, then measure each new setup the same way. You are not chasing absolute values, you are watching for a shift. That is fast, cheap, and catches the problems that actually reach production.

There is a fourth option that engineers forget: change the process instead of measuring the outcome. A stress-relief heat treatment between roughing and finishing, a climb-milling finishing pass with a sharp tool, and a light final cut of 0.2–0.3 mm all reduce locked-in stress before it becomes a measurement problem.

At GreatLight we machine parts up to 4,000 mm and hold ±0.005 mm, and the practical control for stress is sequencing. Rough, relieve, rest, finish. On thin-wall aluminum we often leave 0.5 mm on the finishing pass and let the part sit overnight before the final cut. That single night removes more distortion than any instrument can detect after the fact.

The first error is measuring a surface that was just machined and never cleaned. Cutting fluid, oxide, and burrs all affect bonding or diffraction. A wiped, lightly etched surface is the baseline. If the surface is smeared from a dull tool, the reading describes the smeared layer, not the part.

The second is ignoring the reference state. Ultrasonic and magnetic methods give relative readings. If the reference block has a different heat treatment or grain size, the calibration is wrong and the numbers look plausible but are meaningless. Always cut a reference from the same lot.

The third is reporting one number. Stress is directional. A surface can be at 150 MPa tension in the rolling direction and 40 MPa compression across it. Report the principal values and the depth, or the data cannot be compared with the next batch.

The fourth is testing after the part has already moved. If a plate has warped, the stress has partly relieved itself. The reading now describes the new state, not the condition that caused the problem. Measure before final unclamping or on a twin coupon that follows the same route.