Abstract: Abnormal sound detection is an important part of compressor quality control. Aiming at the characteristics of rare compressor fault sample signals, a method is proposed to extract the common characteristics of normal signal samples of refrigerator compressors by using self-encoder to realize fault detection. On the basis of a large number of normal signal samples, common features are extracted. Due to the different reconstruction errors of normal and faulty samples, the optimal threshold can be determined as the fault classification standard. The research results show that: under the condition that the fault signal samples are sparse, the common characteristics of the normal signal samples of the refrigerator compressor are used to detect the fault signal samples, and the judgment accuracy rate can reach 97.4%.

Key words: Autoencoder, Acoustic signal, Common feature, Fault sparse