An expert system to identify anodic coating process defects. Part 1 : The contribution of neural networks

Abstract

The origin of process defects that may occur in the production of anodized finishes is categorized and the literature on process defects is reviewed. The author suggests from personal experience that in many plants steps taken to overcome problems due to the occurrence of defects is largely empirical and based on prior experience It is considered that this is a situation in which a systematic approach using computer-based information technology has practical advantages. After briefly discussing expert systems that have been used in metal finishing it is argued that these have limitations when applied to the identification of process defects since there is a degree of uncertainty existing as to the conditions that prevailed when the defect was produced. A neural networks program, considered to be particularly suited to evaluating problems in condition of uncertainty, has been adapted for the identification of defects. The primary classification is based on whether the defect is below or within the anodic coating, or associated with sealing. Having made this primary identification the user is directed to a file which relates to a specific process stage at which the defect was produced. After entering those features that describe the defect the program will identify the defect and indicate the probability of the classification being correct. Examples are given of the application of the program to defect identification.