Convolutional Neural Networks for Fault Diagnosis and Condition Monitoring of Induction Motors
FICC 2023: Advances in Information and Communication
Intelligent fault diagnosis methods using vibration signal analysis is widely used for fault detection of bearing for condition monitoring of induction motors. This has several challenges. First, a combination of various data preprocessing methods is required for preparing vibration time-series data as input for training machine learning models. in addition, there is no specific number(s) of features or one methodology for data transformation that guarantee reliable fault diagnosis results. In this study, we use a benchmark dataset to train convolutional neural networks (CNN) on raw vibration signals and feature-extracted data in two separate experiments. The empirical results show that the CNN model trained on raw data has superior performance, with an average accuracy of 98.64%, and ROC and F1 score of over 0.99. The results suggest that training deep learning models such as CNN are promising substitution for conventional signal processing and machine learning models for fault diagnosis and condition monitoring of induction motors.
Bearing fault diagnosis, Condition monitoring, Convolutional neural networks
Information Systems and Technology
Fatemeh Davoudi Kakhki and Armin Moghadam. "Convolutional Neural Networks for Fault Diagnosis and Condition Monitoring of Induction Motors" FICC 2023: Advances in Information and Communication (2023): 233-241. https://doi.org/10.1007/978-3-031-28073-3_16