Abstract: This study presents the integration and simulation of a hybrid deep learning framework for induction motor fault diagnosis. The framework combines Convolutional Neural Networks (CNN) for feature extraction with Bidirectional Long Short-Term Memory (Bi-LSTM) for temporal sequence learning. Vibration data representing both normal and imbalance fault conditions were sourced from a benchmark induction motor dataset, pre-processed through downsampling, and balanced using the Synthetic Minority Oversampling Technique (SMOTE). To ensure reliable performance estimation, the model was evaluated under three validation schemes: Stratified K-Fold cross-validation (K=5), Nested Cross-Validation with hyperparameter tuning, and TimeSeriesSplit validation to preserve temporal order. Results show that the CNN-Bi-LSTM achieved the highest average accuracy of 93% ± 0.01 under Stratified K-Fold, outperforming classical baselines such as Support Vector Machines (85%), K-Nearest Neighbors (89%), and Multilayer Perceptron (92%). Nested CV confirmed the model’s generalization ability during hyperparameter optimization, while TimeSeriesSplit demonstrated its adaptability to sequential data streams, a critical feature for real-time monitoring. These findings confirm that integrating CNN and Bi-LSTM yields a robust predictive maintenance framework capable of capturing both spatial fault signatures and temporal dependencies. The hybrid model demonstrates strong potential for deployment in industrial environments, offering reliable and accurate fault detection that supports proactive maintenance strategies.
Keywords: Induction motor, Fault diagnosis, CNN-Bi-LSTM, Predictive maintenance, Stratified K-Fold, Nested Cross-Validation, TimeSeriesSplit.
Title: Integration and Simulation of a CNN–Bi-LSTM Hybrid Model for Fault Diagnosis in Induction Motors
Author: Odeh A.A, C.M. Onuigbo, P.C. Ene, Opata C. E, C. Okechukwu
International Journal of Electrical and Electronics Research
ISSN 2348-6988 (online)
Vol. 13, Issue 4, October 2025 - December 2025
Page No: 1-21
Research Publish Journals
Website: www.researchpublish.com
Published Date: 06-October-2025
