320 Faulty Welds Traced by a Keyence Traceability System on a Toyota Line

Set up the 320-weld scenario

Three Keyence data-matrix reader gates sit along a Toyota body-in-white line. During one production shift, the readers capture part identity while the weld monitoring system records 320 welds outside the allowed resistance-and-current envelope.

Each shell has a serial identity of its own. The line controller stores the weld schedule as well, including the weld IDs and the robot cell assigned to each operation. When the Keyence readers feed the part serials into that record, the plant can retrieve the affected shells, the exact weld points, and the Fanuc cell associated with the trend.

Containment becomes narrower because the 320-weld excursion points to a rack of shells already inside the plant. The suspect weld IDs are already attached to those records. Serial mapping sends quality technicians to known physical parts without placing unrelated production on a broad hold.

How the count becomes findable

The Keyence readers provide identification and routing, while the spot-welding monitors on the guns judge weld acceptance. Those monitors send dynamic resistance curves to the SIMATIC panel, where the curves are compared with a reference band. In this case, 320 consecutive curves move toward higher initial resistance, the monitor assigns those welds to quarantine, and the traceability layer exposes the matching shells in the plant.

Cap wear and the fixture stack-up

The main fault is electrode cap wear on one gun in a robotic weld cell. The cap is a copper-alloy consumable, and its working face changes shape as production runs.

When the face mushrooms outward, the contact area grows. The larger contact patch lowers current density at the faying surfaces beneath the level assumed by the weld schedule.

A smaller weld nugget follows if the schedule still expects a sharper cap profile. The transformer can continue to deliver the programmed amperage, yet the monitor sees a curve that resembles an under-current condition. The Siemens logic and the Fanuc motion program remain as commissioned. The consumable has moved past its useful dressing interval.

The previous dressing routine is driven by cycle count. After a fixed number of spots, the gun moves to the tip dresser, the face is cleaned and reshaped, and production restarts. That routine behaves cleanly when the part mix stays close to the duty profile used to set the interval.

A cell that spends more time welding thick-gauge reinforcement panels loads some guns harder, both thermally and mechanically. Those caps wear faster than the fixed counter predicts, so the process can drift before the scheduled dress event arrives.

The locating fixture explains why some borderline welds cross the limit sooner. Under certain part-stack tolerances, a reinforcement panel can sit with a slight gap. That gap shifts the resistance path enough to push weak-margin welds out of range, so the cap-wear trend and the fixture condition appear in the same 320-weld containment list.

The fixture revision is modest: a presence-detect pin and a clamp-confirm sensor. The cell withholds the weld schedule until the stack-up is confirmed flush. Variation that the weld monitor had been catching after the operation is removed before the weld fires.

Under the revised maintenance strategy, the resistance curve becomes the wear signal. When the leading edge crosses a threshold, the SIMATIC panel calls for a dress cycle or a cap change while the weld is still inside the acceptance window. The degradation appears in monitor data the panel already trends, so no new instrumentation is required and the maintenance action follows measured wear without changing the gun hardware.

The Fanuc retrofit under ISO 13849 guarding

Changing the dresser trigger from fixed count to process condition brings the safety architecture into the retrofit review. The Fanuc arm and dresser station are behind guarding rated to ISO 13849, with a performance level assigned to light curtains and interlocked access gates. Any increase or change in dresser actuation has to preserve the safety-related control functions.

In this setup, the condition-based request is written in the standard control logic and passed to the dresser through the same safe-torque-off gating that already governs the axis. The SIMATIC safety controller continues to manage access and motion permissives at the assigned performance level. The dress decision remains in the standard PLC task and can request action only when the safety state permits it, keeping the maintenance change from altering the machine risk profile during recovery from the weld issue.

Cycle-time cost of condition-based dressing

Throughput concern shows up downstream from the body station. A handling cell that stops more often for cap dressing can disturb the Fanuc palletising cycle time budget. The body line feeds a buffer, the palletiser draws from that buffer, and irregular dress stops make the level swing.

Assume a 60-second station cycle and a cap dress that takes 4 seconds of gun downtime. With the fixed schedule, dressing every 400 spots on a gun making one weld per part creates a 4-second event every 400 cycles. The average overhead is about 0.01 seconds per cycle.

During a high thick-gauge mix, the condition-based trigger on the worst-duty gun pulls the interval to roughly every 250 spots. Average overhead rises to about 0.016 seconds per cycle. In the cycle-time spreadsheet, the extra load barely moves the line.

The disturbance comes from clustering. Thick-panel batches can make several guns request dressing within a short window. If the cell combines those requests in one cycle, the downtime can reach 12 to 16 seconds. The downstream buffer then starves, and the palletiser waits for parts.

Scheduling logic in the SIMATIC panel handles the peak load. Condition-triggered dress requests are staggered across guns, so only one gun dresses in a given cycle. The total dressing load is spread across more cycles, turning a 12-to-16-second stall into a single 4-second event that the buffer can absorb.

Average cycle loss can look acceptable while the plant still sees intermittent downstream starvation and blames the palletiser. Comparing the dress-request log with the buffer-level trend ties the delay to clustered cap maintenance.

Traceability after containment

The 320 flagged welds remain inside containment, and recovery avoids a full teardown of the cell. The weld monitors generate the degradation signal first. The SIMATIC panel stores and trends the curves, the guarded Fanuc cell accepts the new dress trigger without compromising safety functions, and Keyence identification ties the serial records back to the physical shells.

The threshold value cannot be treated as a clean cell-wide constant. The worst-duty gun produces a clear resistance-curve trend that trips the limit cleanly. On lighter mixed-duty guns, the same setting can call the dresser on noise.