Hybrid 1D CNN-RNN Network for Fault Diagnosis in Induction Motors Using Electrical Signals

Abstract

Induction motors are prevalent in many industrial applications due to their robustness, efficiency, and reliability. They are used in various applications, such as pumps, fans, compressors, conveyors, and machine tools. However, faults in induction motors can cause operational and financial losses, and in some cases, they can lead to severe accidents. Therefore, timely and accurate detection of faults is crucial for minimizing the negative impact of these faults. The fault detection methods for induction motors can involve the analysis of various signals such as vibration, current, and voltage. Convolutional neural networks (CNNs) have proven highly effective in many applications but have mainly been applied to two-dimensional data. One-dimensional CNNs offer an excellent alternative for analyzing time sequence datasets since they can work directly with raw signal data without requiring pre- or post-processing. However, the main idea behind 1D-CNNs is to extract spatial features, which can result in the loss of critical temporal features related to time distribution. Recurrent neural networks (RNNs) can effectively capture the temporal dependencies and time distribution in sequences data, making them well-suited to fix the issue. In this paper, we propose a method that combines 1D-CNNs and RNNs called Hybrid 1DCNN-RNN network (HCRN) to analyze the voltage and current signals of a three-phase induction motor. It performs accurate and efficient fault diagnosis, ultimately leading to the more efficient maintenance and reduced downtime for industrial processes.