Faulty machine components often get replaced instead of repaired in industries. The involved cost and downtime increase due to this problem, which can be mitigated with proper maintenance planning. This paper proposes a multi-fault diagnosis of two critical rotating components, motor and bearing at a system level. Motor current signatures of all possible muti-fault conditions were acquired at different speeds. The extracted signals were divided into small signal segments by implementing a sliding window with a fixed sample size. Continuous Wavelet Transform (CWT) was then employed on each extracted segment, and different datasets were formed for doing signal, sequence and image classification using Artificial Neural Network (ANN), 1D Convolutional Neural Network (CNN) and 2D CNN models, respectively. A comparative analysis of model performance was performed and found that 2D CNN outperformed other machine learning models. The Motor Current Signature Analysis (MCSA) could be used for system-level multi-fault anomaly detection using machine learning algorithms.
Realistic Condition-Based Anomaly Detection of Multi-Faults in Rotating Machines
2023-08-09
774722 byte
Conference paper
Electronic Resource
English
Quantification of faults in rotating machines
Online Contents | 2007
|Condition monitoring of rotating electrical machines
TIBKAT | 2020
|Condition monitoring of rotating electrical machines
TIBKAT | 2008
|Phase Angle Estimation Algorithms for Detecting Faults by Synchronous Analysis of Rotating Machines
British Library Conference Proceedings | 1995
|