Information fusion feature preprocessor based on FRFT for analog circuits fault diagnosis

This paper presents a new fault feature preprocessor method for analog circuit fault diagnosis. An information fusion method based on fractional Fourier transform (FRFT) is introduced to extract features from voltages of the circuit under test (CUT). Firstly, the voltage signals gathered from test nodes of the CUT are preprocessed by FRFT, the fractional order p of the FRFT changes from 0 to 1 with a given step. Then, we gain the amplitudes of the transformed signals in fractional space and extract the mutual information entropies as features by a defined division scale. After normalization, the extracted features are used to train a neural network to diagnose faulty components in the CUT. The proposed feature preprocessor method is applied to two CUTs and is compared with three ordinary preprocessing methods in analog circuit fault diagnosis. The experiment results reveal that the proposed method can simplify the structure of the network and improve the diagnosis performance.