Multi-fault diagnosis of electronic circuit boards using neural networks

A report is presented on the feasibility of using artificial neural networks to recognize faults in electronic circuit boards. In particular, the issue of networks trained for single-fault diagnosis being used to recognize multiple simultaneous faults is investigated. The research concentrated on determining the number of physical test points needed. The effect of using more than one test pattern to identify the fault, and the effect on generalization of the number of nodes in the hidden layer. The study has shown that neural networks can be used for automatic knowledge acquisition in the diagnosis of electronic circuit board failures. The results are not always 100% reliable, and the user must be willing to accept this lower reliability. The fault detection rate is highest with neural networks having 96 nodes, and the false alarm rate is lower. As the number of nodes in the input layer decreases, the fault detection rate decreases and the false alarm rate increases. Increasing the fault threshold values decreases the number of false alarms and also decreases the fault detection rate. The results indicate that it is important to use more than one test vector reading. It is shown that as long as the number of nodes in the hidden layers is sufficient to do the classifications. there is little impact on the fault detection rates