Analog Circuit Incipient Fault Diagnosis from Raw Signals using Multi-Layer Extreme Learning Machine

Analog circuits play important roles in modern electronic systems. Incipient fault diagnosis of analog circuits is a recognized challenging research direction due to the difficulty of fault feature extraction and identification. This paper proposes an early fault diagnosis algorithm for analog circuits based on multilayer extreme learning machine (ML-ELM).Its basic idea comes from the Auto-encoder(AE) and the Extreme Learning Machine (ELM). The proposed method which combines the characteristics of both, so it has the ability of feature extraction and fast training speed. In this method, the time domain sampling signal of the circuit can be directly used as the fault sample, and the unsupervised feature extraction is carried out layer by layer through the deep network, the diagnosis results are obtained after the supervised classification by the traditional ELM algorithm. The whole process does not rely on multi-step iterative and reverses fine-tuning. The experimental results of the Sallen-Key band-pass filter circuit and Leapfrog low-pass filter circuit show that the method has not only high diagnosis accuracy, but also fast diagnosis speed. Because this method gets rid of the tedious manual feature extraction and complex signal processing, it improves the training efficiency and makes the method more universal.