Detection of abnormal sound in industrial equipment based on classification and outlier calculation
Fucai Hu, Shaowei Li, Linke Zhang, Xiaoxi Guan, Fangyuan Cheng, Yongsheng Yu · 2023
Aiming at the problem that there are many abnormal sound types in industrial equipment and few fault samples can be obtained, a method for detecting abnormal sound of industrial equipment is proposed. Firstly, we focus on two improved neural network models: MobileNetV2 based on classification method and Dense AutoEncoder based on outlier calculation method. Combined with the characteristics of neural network, we propose an improved MFCC feature extraction method. In order to improve the training accuracy, the comprehensive perturbation of the training samples is increased by the data augmentation; finally, the influence of hyperparameters on the training effect is analyzed. Experiments shows that the improved method proposed in this paper achieves an excellent recognition rate.