Predicting bearing damage in electric motors: a comparative study using combined neural networks

The bearing damage within the motors can lead to unexpected failures. This study aims to develop an efficient model for automated prediction of bearing damages within motors, leveraging advanced deep learning techniques combined with traditional machine learning algorithms. The proposed method operates by extracting intricate features from input bearing damage data using a Fine-tuned Convolutional Neural Network (CNN) combined with a Recurrent Neural Network (RNN) model. These extracted features are then integrated into Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbor (KNN), and CART models for predictive analysis. The proposed integrated model was validated using the Paderbone University Bearing Fault Benchmark Dataset, which contains six classes of bearing conditions. The developed Fine-tuned CNN-RNN model achieved an exceptional performance, attaining 96% accuracy across all six bearing damage classes. Comparative analysis with prominent algorithms, including CART (99% accuracy), SVM (96% accuracy), KNN (96% accuracy), and RF (99% accuracy), underscored the superiority of the proposed model's integrated approach. The results demonstrate the efficacy of the Fine-tuned CNN-RNN architecture in accurately predicting and diagnosing bearing damages within motors. This approach has the potential to enhance predictive maintenance and reliability engineering practices, contributing to the improved operational reliability and performance of electric motor systems.