A Fault Detection Approach Based on Sound Signal Analysis for Equipment Monitoring

Abstract

Machine fault detection has great practical significance. Compared with the detection method that requires external sensors, the detection of machine fault by sound signal does not need to destroy its structure. The current popular audio-based fault detection often needs a lot of learning data and complex learning process, and needs the support of known fault database. The fault detection method based on audio proposed in this paper only needs to ensure that the machine works normally in the first second. Through the correlation coefficient calculation, energy analysis, EMD and other methods to carry out time-frequency analysis of the subsequent collected sound signals, we can detect whether the machine has fault.

Research on the Modeling Method of Multi-Sensor Based Fault Diagnosis Models

“…In this study, we used a BP neural network to establish a fault diagnosis model based on multiple oil characteristics [9] , which works as shown in Figure 2. The input parameters include oil characteristics such as water content, viscosity, and the number of ferromagnetic abrasive particles in different particle size ranges.…”