Root Crack Incipient Gear fault detection using Machine Learning Algorithms

Abstract

Rapid advancements in sensors and computing technology have motivated stupendous progress in evolving the next generation of Industry 4.0, where machines really speak about their health state in real time. Low-speed machinery, such as the churn gearbox in a chemical plant and the wind turbine main bearing, is considered to be the most difficult to analyse for incipient faults because these faults remain hidden in strong environmental noise during machine operation. Early detection of these faults at the incipient stage is paramountto increasing the overall availability of the whole plant and machinery. Vibration sensor-based fault diagnosis techniques were found to be the most popular among researchers working in the condition monitoring field as well as the most adopted framework in industries. However, acoustic analysis, a non-contact type of sensing technique, possesses huge potential in the fault diagnosis of rotating machinery. This article describesthe framework for fault detection of a slow-speed gearbox working under variable speeds and loads using data fusion from vibration and acoustic signals at the statistical feature level. Maximal Overlap Discrete Wavelet Transform (MODWT) is used as the signal processing technique to remove the noise from the vibro-acoustic signals and extract the fault-related information. Feature monotonicity of the extracted features from MODWT-processed signals is considered as the selection criteria for selecting vibro-acoustic features containing maximum fault-related information. A performance comparison of sensor data fusion over vibration and acoustic sensors when used alone is also presented. To validate the performance of the proposed framework, five different artificial intelligence (AI) models have been used, and the results have been compared.