New vision technologies provide an opportunity for fast detection and diagnosis of quality problems compared with traditional dimensional measurement techniques. This paper proposes a new use of image processing to detect quality faults using images traditionally obtained to guide manufacturing processes. The proposed method utilizes face recognition tools to eliminate the need of specific feature detection on determining out-of-specification parts. The algorithm is trained with previously classified images. New images are then classified into two groups, healthy and unhealthy. This paper proposes a method that combines Discrete Cosine Transform (DCT) with either Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA) to detect faults, such as cracks, directly from sheet metal parts.

This content is only available via PDF.
You do not currently have access to this content.