A fast soft-fault diagnosis procedure using a first nearest neighbor classification

Abstract

This article proposes a fault diagnosis scheme for second-order switched-capacitor filters embedded in the Infineon PSoC1 reconfigurable microcontroller. The faults considered are single soft faults that diminish the capacitance of the capacitors. Specifically, the scheme focuses on incipient faults, which are the hardest to identify because they produce a small impact on the circuit’s performance. The scheme estimates the value of the degradation of the capacitors by employing a first nearest neighbor (1NN) classifier. Our proposal compares the output test patterns against a dictionary of selected fault patterns (one for each capacitor) using Dynamic Time Warping (DTW) distance measures. The easiness of construction of this dictionary and the conceptual simplicity of the method are the most relevant features of the proposal. The characterization of the scheme is made with MatLab simulations at a transfer-function level to limit the computational cost. The simulation results show that the diagnosis procedure can determine the capacitor that presents incipient faults.