Minimum measurements at minimum set of test nodes for analog circuit fault diagnosis

A new technique for isolating all separable hard faults in analog circuits using minimum measurements at a minimum pre-selected set of test nodes is presented. The spectrum of the circuit response to a sinusoidal input test signal is simulated at all circuit nodes. A clustering algorithm is then applied to evaluate the separability performance of all simulated measurements. Uniquely classified faults are isolated each in one set while faults having similar responses are grouped in one ambiguity set. The proposed algorithm is then applied in two consequent steps. In the first step, selection of an optimal minimal set of test nodes is achieved. In the second step, selection of minimum measurements at the pre-selected set of test nodes is realized. The resultant overall classification accuracy is comparable to that obtained using all measurements at all circuit nodes. The proposed strategy is demonstrated with a three-stage active filter circuit example. Moreover, classification results are verified using a learning vector quantization neural network.