Inspection of Tapered Rollers for a Global Bearings Manufacturer
It was decided to use a Deep Learning AI powered inspection technique since the defects were qualitative and across a wide range of roller SKUs. The key steps followed in this workflow consisted of image collection, image annotation, Deep Learning model selection/training, deriving an optimized Edge inference model, deployment on the production floor and, finally, maintenance.
Qualitas worked collaboratively with the customer to collect and annotate a sufficient number of good (G) and not-good (NG) images of the tapered rollers, showing both the cylindrical and larger flat surfaces. A few hundred images were thus collected and processed. This image data was used to train the chosen Deep Learning AI model iteratively till acceptable performance was achieved. A key consideration was to keep false positive and false negative predictions sufficiently low across the wide variety of SKUs for a range of subjective surface defects.