Web process inspection using neural network classification of scattering light

The authors note that web process inspection requires the rapid examination of vast amounts of data. They examine two resulting issues: the sensory system and the computer processing required to detect faults in the sheet material accurately. It was found that scattered coherent light from the surface of the material being processed could be directly conditioned by a photodetector so as to produce a small set of features which are then examined by a neural network trained to find unsatisfactory surface conditions. A surface inspection system using measurement of the angular distribution over a 25° cone angle of the scattering was constructed, calibrated, and evaluated for inspection of coated sheet steel samples. Features created by a simulated segmented photodetector are inputs to a neural network which uses classification based on Kohonen´s LVQ2 (learning vector quantization-2) algorithm. The system was evaluated with CrO₂ coated steel samples. Classification by fault or no fault categorized 133 samples correctly out of 135, while there were seven errors in one attempt at classification into the various common surface faults out of the same number of test samples and nine in another