A CNN-CLSTM Approach for Condition Monitoring and Fault Diagnosis of Induction Motor on Manufacturing

Abstract

Numerous industrial applications rely on induction motors due to their numerous advantages, especially three-phase induction motors. Their safe and reliable operation is, thus, very important. The motor is prone to problems and breakdowns, which can lead to long periods of inactivity and substantial operational and financial losses. So, early fault detection is critical for vehicle safety. Precise sequentially is required throughout the method selection, preprocessing, feature extraction, and model training processes. One non-linear signal processing method used in data preparation is Cepstrum analysis, which is the integral derivative of the logarithm of the input signal's absolute value of its DFT. With the use of ICA, a multivariate random signal can be converted into a signal with components that are statistically completely independent of each other. This process is known as feature extraction. When training a CNN-CLSTM model, feature extraction takes precedence. This new method outperforms two cutting-edge algorithms: CNN and CLSTM. Accuracy improved significantly, reaching 97.38%, according to the results.