Artificial Neural Network Approach For FunctionalFault Diagnosis Of Complex Printed Circuit Boards
T. Arslan and A.A. Al-Jumah · WIT transactions on information and communication technologies · 1970
With the continuous increase in the complexity of electronic systems the issue of testing such systems is becoming an increasingly difficult task. The large number of constituent components on a printed circuit board and their internal complexity rules out traditional simulation-based techniques and emphasis upon using techniques which are based on artificial intelligence. This work is a continuation of research carried out by Arslan et al. [2] in which diagnosis is performed using example failure reports produced by Automatic Test Equipments (ATEs). The work of Arslan has also illustrated that expert system-based techniques could be used for the diagnosis procedure, however, the construction of expert systems is a lengthy procedure and their subsequent customisation for different circuits and ATEs require significant modifications to both knowledge-base and rule-base. This task could only be performed by experienced knowledge engineers. In this paper the use of artificial neural networks is investigated in the diagnosis process. A number of neural architectures are investigated mainly based upon Multi-Layer Perceptron and Learning Vector Quantization. Artificial neural networks are designed such that they process test data output from ATEs and indicate possible faults which may cause the failure of a given printed circuit board. The paper describes the neural network architectures investigated and the diagnosis performance achieved. In addition the paper reports on the ability of the neural networks in dealing with noise and duplicate reports that are a characteristic of failures represented in the test data.