Identification of electronic component faults using neural networks and fuzzy systems

The authors describe the development of a two-step procedure for identifying fault components in electronic circuits containing both analog and digital components. A neural network uses circuit input and output voltage values as inputs to the network and has individual output nodes corresponding to potential faulty components. When a set of tests (input/output patterns) from a faulty board are applied to the neural network, either one or many faulty components will be indicated. If a test points to one component, then that component is bad and no further diagnosis is necessary. If a test indicates that more than one component may be bad, then further work using a fuzzy system is required to identify the faulty component. Data from a 50-component printed circuit board were used to test this neural/fuzzy faulty component detection system