A tolerance method for industrial image-based inspection

Although image-based inspection has been applied to a wide range of industrial applications, inspection accuracy remains a challenging issue due to the complexity involved in industrial inspection. The common method adopted in industry is to use a qualified image as a template image to inspect each live image on a pixel-by-pixel basis. In this paper, a tolerance method is presented to replace the template image method. The said tolerance is formed by two indices computed from a sample image, instead of using the whole image for inspection. To ensure an accurate tolerance zone, a neural networks method is used to take the noise and uncertainties in the images under inspection into consideration. To improve neural networks training speed, the Taguchi method is adopted to select a minimum number of the sample images needed for training. Once a tolerance zone is obtained, live images are inspected against it. If the indices of a live image fall inside the tolerance zone, the part is regarded as good, otherwise defective. Three examples are given: one for auto part inspection, one for label inspection, and one for machining part inspection. The inspection accuracy achieved is above 94%.