Enhanced fault diagnosis and remaining useful life prediction of rolling bearings using a hybrid multilayer perceptron and LSTM network model

Rolling bearings are critical components in many machines, and their performance is vital to ensure the smooth and efficient operation of various industrial systems. Accurate fault diagnosis and prediction of rolling bearings' remaining useful life (RUL) are essential for implementing effective predictive maintenance strategies. This paper proposes a hybrid approach using a feedforward MLP and LSTM network to improve fault detection and RUL estimation. The dataset includes voltage signals recorded under various fault conditions, such as inner and outer raceway defects. The signals were pre-processed using normalization and band-pass filtering to enhance data quality before feature extraction. STFT was employed to transform the voltage signals into time-frequency representations for detailed analysis of transient characteristics. The hybrid MLP-LSTM model leverages the strengths of both architectures: the MLP captures complex, non-linear relationships, while the LSTM handles sequential dependencies. The model's performance was compared against traditional methods such as FCN, SVM, Decision Tree, KNN, LSTM, and CNN-BILSTM, with the proposed model demonstrating superior results. Our findings highlight the model's effectiveness in identifying bearing faults and accurately predicting RUL, leading to more informed maintenance decisions. The proposed approach can significantly reduce unplanned downtime, extend the operational life of critical machinery, and optimize maintenance schedules. The model achieved an accuracy of 99.9 %, sensitivity of 98.90 %, and specificity of 98.16 %, which is implemented by Python, outperforming existing methods and proving its suitability for predictive maintenance in industrial applications.