A Fault Detection Method for Analog Circuits Based on the Wavelet Features and One-class KNN

The fault detection method for analog circuits based on multi-classification learning can not only determine the circuit state, but also locate the faulty components and identify the fault classes. However, in the absence of fault samples, only the fault detection method based on one-class learning can accurately identify the circuit states. To ensure the reliability of electronic equipment, a fault detection method based on the wavelet features and one-class KNN is proposed in this paper. This method can obtain the detection threshold by using only the normal samples of the analog circuits to identify the circuit states effectively. Firstly, the wavelet features are calculated to mine the key information of the signals. Then, to enhance the efficiency of fault detection, principal component analysis (PCA) is used to reduce the dimension of wavelet features. Finally, the normal samples are adopted to train the fault detection model named one-class K-nearest neighbor (KNN), and the detection threshold is empirically determined from the training samples, therefore monitoring the circuit states. Furthermore, the four-op-amp biquad high-pass filter circuit is chosen as the experimental circuit to implement the simulation experiments of fault detection. Experimental results demonstrate that the proposed fault detection method presents better performance on identifying circuit states compared with other typical methods.