Diagnosis of multifaults in analogue circuits using multilayer perceptrons

It is shown, by means of an example, how multiple faults in bipolar analogue integrated circuits can be diagnosed, and their resistances determined, from the magnitudes of the Fourier harmonics in the spectrum of the circuit responses to a sinusoidal input test signal using a two-stage multilayer perceptron (MLP) artificial neural network arrangement to classify the responses to the corresponding fault. A sensitivity analysis is performed to identify those harmonic amplitudes which are most sensitive to the faults, and also to which faults the functioning of the circuit under test is most sensitive. The experimental and simulation procedures are described. The procedures adopted for data preprocessing and for training the MLPs are given. One hundred percent diagnostic accuracy was achieved, and most resistances were determined with tolerable accuracy.