Can You Trust Your Removal Torque Data: MSA and GR&R for Torque Measurement Systems

No matter how tight a removal torque window is, if the measurement system alone consumes most of that window, the numbers can only be read as a trend and never used for accept or reject decisions. This is the step most often skipped in removal processes: teams fight to narrow the window from plus or minus ten percent to plus or minus five, without ever verifying that their measurement chain can resolve such a difference.

Prove the measurement system first, then talk about the torque window

Measurement system analysis (MSA) answers one question: of the variation you observe, how much comes from the parts and how much from the gauge and the operator. Gauge repeatability and reproducibility (GR&R) is the most used part of it, splitting variation into repeatability (spread when one person measures one part on one setup several times) and reproducibility (shift caused by changing people, fixtures, or shifts). GR&R practice for tightening torque is well established, but removal loads in the opposite direction and captures a peak, so copying the tightening method produces a falsely good verdict.

Three difficulties specific to removal

The first is peak capture. The breakaway peak lasts only a few milliseconds, and an inadequate sampling rate returns a smoothed value that looks beautifully repeatable while actually filtering out the real variation. The second is the mechanical chain under reverse loading. Removal runs against the thread hand, and clearance between sensor and bit, and between bit and recess, produces an impact just before the peak; the sensor records that impact as part of the peak, so a range that is too small clips it and a range that is too large loses resolution. The third is clamping stiffness. The same screw measured on a rigid fixture and on a compliant one can differ by more than ten percent, because fixture rebound takes part in releasing the stored energy. This is why sampling work often runs on bench-mounted small screw removal equipment, where fixture and frame form one unit and stiffness does not drift between shifts.

How to lay out GR&R: sampling, levels, repeats

Samples must cover both ends and the middle of the window, typically ten representative parts split into low, medium and high levels, rather than all screws at one torque level, which compresses the computed variation and biases the verdict toward optimistic. Use three operators who actually run the removal station, not engineers standing in. Each operator repeats three times per part in randomized order, with previous readings hidden to prevent memory-based correction.

Removal carries one unavoidable limit: repeated cycles on the same screw change thread friction, and the second breakaway torque usually reads lower than the first. A removal GR&R therefore cannot re-tighten at the same location the way a tightening study does. There are two correct routes: use dedicated coupons with stabilized threads that can be cycled, or use a nested design in which each measurement installs a fresh screw from the same lot into the same threaded hole, separating screw-to-screw variation from measurement system variation. Get this wrong and the computed repeatability is really part wear, and the conclusion sends process improvement in the wrong direction.

Decision rules: two thresholds and one index

The tolerance-based rule is the most general: measurement variation below ten percent of tolerance is acceptable; ten to thirty percent is conditionally acceptable depending on application criticality and improvement cost; above thirty percent is not acceptable, and the measurement chain must be fixed before discussing process capability. Also check the number of distinct categories, ndc, which must be at least five, meaning the system can resolve the tolerance band into five distinguishable levels. For a process with a window as narrow as removal, watch one more figure: measurement uncertainty as a share of window width. If it exceeds one third of the window, even a good GR&R number cannot support single-screw decisions.

Pulling variation down in three directions when the study fails

First, range and sampling. Set sensor range at one and a half to three times the peak and never measure small torque with a large-range sensor; derive sampling rate from one tenth of peak duration, which for millisecond-level removal peaks usually means ten kilohertz or more. Second, the mechanical chain. Bit-to-recess fit sets the size of the impact, and a worn bit raises both mean and spread; a floating joint absorbs lateral offset but introduces angular error, so the structure has to limit it. Third, clamping and alignment. Treat workpiece clamping stiffness as a process constant and write it into the work instruction, because different fixtures on different shifts are a common reason removal data cannot be compared.

Machine-side support is straightforward: log removal torque, peak current and descent stroke together, use descent stroke to confirm the bit actually seated, and use peak current to help detect threadlocker residue. An automatic vision screw removal machine goes one step further and confirms recess and screw head condition before the tool engages, taking incoming defects such as a pre-shaved recess out of the torque data. For plants sharing one rework line across the full M0.8-M2.0 range, one practical value of an M0.8-M2.0 screw removal solution is that each size keeps its own parameters and its own decision thresholds, so a single GR&R verdict is never stretched across every size.

Honred Technology (Shenzhen) supports this on the equipment side with independently settable parameters and traceable data: axial force, descent speed and vacuum pick are set separately, and breakaway torque is stored together with images, so a plant doing GR&R or batch analysis has raw records instead of a summary report. For labs and pilot lines built around bench machines that switch small lots frequently, sampling convenience often matters more than cycle time.