China Region Telephone

18688976300
News
Service Hotline

18688976300

Mobile: 18688976300 (Mr. Peng) 15222276387 (Mr. Bai)

Email: pengwl@honred.com

QQ: 176496342

Address:Building A1, Runteng Industrial Park, Shiyan Street, Bao'an District, Shenzhen

Removal Torque SPC and Trend Alerting: Metrics, Sampling and Out-of-Control Rules

Date: 2026-11-21       Views: 11

A torque transducer on a removal station is not process control. The sensor answers whether this one screw came out correctly; a control chart answers whether the batch has drifted. What sits between the two is not hardware but statistical discipline: which metric, how many parts, when to alarm, and who does what afterwards.

A passing rate is not a controlled process

A per-screw verdict proves one screw was fine; a control chart proves the batch has not drifted. Removal differs from fastening in one fundamental way. When driving, the tool applies the torque and repeatability is high. When removing, the torque is decided in reverse by the state of the joint, so cure level of threadlocker, thread wear, incoming lot and bit wear all act on it, and all of them drift slowly with time. Yield is a lagging result; a trend metric is what catches the shift early.

Choose the metric: one-sided spec means upper limit and spread

Removal torque is a one-sided specification, so the chart should watch the upper limit and the spread rather than the average alone. The upper limit protects the workpiece, since a screw that comes out too hard has usually seized and the thread or the boss takes the damage. The lower limit is not really a specification at all; it marks a screw that spun free, which belongs in a defect count rather than in a distribution. Use an Xbar-R chart at subgroup size five, and switch to Xbar-s once subgroups reach ten. Limits have to be computed from roughly twenty-five stable subgroups in a pilot run, never copied from the drawing tolerance, otherwise the chart only restates the specification.

Sampling: small and frequent beats large and rare

Five consecutive parts every two hours catches drift that a hundred parts once a shift will miss. The drivers on a removal station are bit wear, air supply fluctuation and incoming lot variation, all slow variables, and consecutive sampling is what separates them from single-part anomalies. Pair subgroup size five with a range chart and move to a sigma chart above ten; capability targets are usually 1.33 and 1.67 for automotive work. A small screw removal equipment used as a rework cell should log every cycle, so subgroups can be rebuilt from history instead of relying on a clipboard.

Three of the eight rules carry the value

Trend rules catch a shift far earlier than a point crossing the limit. Seven consecutive points on one side of the centre line, six consecutive points rising or falling, and two out of three points beyond two sigma on the same side are the three that fire most often on a removal station; a single point outside the limit is usually the result rather than the warning. Two disciplines go with them. Parts already judged as stripped or seized must be removed before plotting, because they pull the mean and inflate the limits. Defect rate belongs on its own p-chart rather than inside the torque distribution. Comparability matters as much as sample size, since a subgroup should come from the same screw position on the same model, which is why an automatic vision screw removal machine that records model and position every cycle makes the chart readable at all.

Closing the loop from alarm to action

An alarm without a mapped action is just a log entry. Two levels work well. Yellow means a single trend-rule hit, answered by re-measuring five parts and checking bit wear. Red means repeated trend hits or a point outside the limit, answered by stopping the line, changing the bit, checking air dew point and pressure, and re-verifying the transducer. Every alarm needs a timestamp, model, screw position and disassembly count to be usable in an 8D report. Small fasteners also need bit diameter and pick-up method in the action list, because when a M0.8-M2.0 screw removal solution changes nozzle between models the torque distribution shifts as a whole, and the chart has to be segmented by recipe or every changeover will raise a false alarm.