| Summary: | Correct identifying analog circuit incipient faults is useful to the circuit's health monitoring, and yet it is very hard. In this paper, an analog circuit incipient fault diagnosis method using deep belief network (DBN) based features extraction is presented. In the diagnosis scheme, time responses of analog circuits are measured, and then features are extracted by using the DBN method. Meanwhile, the learning rates of DBN are produced by using quantum-behaved particle swarm optimization (QPSO) algorithm, which is beneficial to optimizing the structure parameters of DBN. Afterward, a support vector machine (SVM) based incipient fault diagnosis model is constructed on basis of the extracted features to classify incipient faulty components, where the regularization parameter and width factor of SVM are yielded by using the QPSO algorithm. Sallen-Key bandpass filter and four-op-amp biquad high pass filter incipient fault diagnosis simulations are conducted to demonstrate the proposed diagnosis method, and comparisons verify that the proposed diagnosis method can produce higher diagnosis accuracy than other typical analog circuit fault diagnosis methods.
|
|---|
