<![CDATA[In Industry 4.0, reliable fault diagnosis is critical for minimizing downtime and preventing catastrophic failures in rotating machinery. However, conventional deep learning models often operate deterministically, lacking the ability to quantify prediction uncertainty—a limitation that hinders risk-based maintenance decisions. This study aims to develop a hybrid deep learning framework that integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Bayesian inference for uncertainty-aware fault diagnosis. The model extracts spatial features from Short-Time Fourier Transform (STFT) spectrograms via CNN, models temporal dynamics from raw vibration signals via LSTM, and quantifies prediction uncertainty using Monte Carlo Dropout (T=50). Evaluated on the benchmark Case Western Reserve University (CWRU) bearing dataset with an 80/20 data partitioning under six operating conditions, the hybrid architecture achieves an accuracy of 99.14% and an F1-score of 0.9914, significantly outperforming standalone CNN (97.42%) and LSTM (84.12%) models. The integration of probabilistic inference enhances decision reliability by providing confidence estimates for each prediction. This work contributes a robust, uncertainty-aware model that effectively captures both spatial and temporal patterns, offering significant implications for safety-critical industrial predictive maintenance systems.]]>
Copyrights © 2026
