Stamped Panels: Camera Matrix Replaces 80 Percent Sampling
Stamped panels are wide parts typical of an automotive press line, and their completeness affects welding. Manual sampling covered only part of the run, most products moved on unchecked, and sample selection was random, so a defective part might never be picked. This cell deploys a camera matrix on the press line and inspects stamped panels 100 percent inline. Inspection accuracy rises from 80 percent under manual sampling to above 95 percent, with a 6 second cycle per panel.
Eighty percent is a statement about the sample
Sampling accuracy is easily misread. Eighty percent does not mean 80 percent of the batch is good; it means human judgment was right on 80 percent of the sampled parts. Escape risk is the product of coverage and judgment accuracy, and when coverage is low, perfect judgment cannot stop a part that was never picked.
Stamped parts come in large batches, have modest unit value and cause batch rework at welding; probabilistic passing is most dangerous here.
Coverage is the main variable in escape rate
Raising coverage to 100 percent lowers escape probability, the most direct step here. Under full inspection every unit has a record, anomaly tracing and responsibility have evidence, and the criterion no longer swings with shift or operator state. The precondition is that inspection must not slow the line, or coverage rises while output falls and the arithmetic still fails.
One camera cannot span the panel, so the matrix follows the format
One camera struggles to cover a wide panel; forcing the field loses resolution. The cell uses several industrial cameras in a spatial matrix across the panel, each covering a region, with stitching and zone judgment completing the full-face check. The matrix scales: a wider panel needs more cameras, not one ultra-high-resolution unit. Coverage must be confirmed by calibration so the whole panel lies inside the inspectable area.
Six seconds against the press cycle
The cycle is 6 seconds per panel, synchronized with the press line, subject to site acceptance results. Imaging, comparison, presence judgment and output must all finish in that window, so software schedules cameras together and proceeds in parallel, not one camera after another. The equipment sits inline and interfaces with existing transfer mechanisms, so the retrofit does not affect the main line and parts need no offline wait.
Interception has to keep up with the verdict
A failing panel is intercepted while passing panels flow on. The signal links to line control, and anomalous panels are separated at a designated position without disturbing flow. This is often overlooked: if interception lags or needs manual handling, half the value leaks away at the exit. The press line also brings vibration and cycle fluctuation, so judgment cannot be disturbed by the environment, and imaging and triggering must be designed for those conditions.
The accuracy gain comes from coverage
Accuracy rises from 80 percent under sampling to above 95 percent, and rework cost caused by escapes falls by more than CNY 10 million a year, subject to site acceptance results. The gain comes from expanded coverage, not a smarter algorithm; the saving comes from defective parts not reaching later processes. For that arithmetic to hold, full inspection must run stably over the long term. Occasional and sustained full inspection have very different effects on rework cost.
Questions about replacing sampling with full inspection
Multi-camera matrix or one high-resolution camera?
Look at panel format and feature size. When the format is too wide for one camera's resolution, the matrix is more flexible and easier to scale; if one camera can cover it, one unit is simpler.
Is presence-only inspection enough?
It depends on the main escape mode. Typical escapes are missing parts and skipped operations, covered by presence and key-feature integrity; appearance defects need separate items.
What if surface features are weak?
Use lighting design to raise feature contrast so judgment has a usable difference. If a feature has no readable contrast at all, the approach must be reassessed.
Will 100 percent inspection slow the press line?
Inspection runs inline with no extra transfer, and the 6 second cycle matches the press cycle. Whether that is enough depends on actual line speed and item count.