Fault prediction method of analog circuits based on MLSTM

Abstract

Abstract According to statistics, most of the electronic equipment fault cases are related to the analog circuit, which may lead to the failure of the whole system due to the drift or performance degradation of the parameters of an electronic component in the analog circuit. However, because of the continuity of the parameters of each component, the lack of test nodes and the nonlinearity problems, the analog circuit becomes more difficult to prognose. The fluctuation and inconspicuousness of the degradation trend in analog circuits lead to the difficulty in extracting characteristic parameters and low accuracy of fault prediction. Therefore, a fault prediction method for analog circuits based on Multi-scale and Long-Short Term Memory (MLSTM) is proposed in this paper. Multi-scale feature learning method can divide the original data into different scales to improve the learning accuracy of small sample data. LSTM can well identify the temporal sequence of the widely separated events in the noise input stream. In this paper, the multi-scale feature learning method is combined with LSTM to improve the accuracy of fault prediction. Finally, the DC-DC switching power supply is taken as a case for verification, and the proposed method is compared with other algorithms to verify its accuracy and applicability.