Industrial Predictive Maintenance: Al-Driven Temporal Modeling for Remaining Useful Life Prediction

Predicting the Remaining Useful Life (RUL) of industrial components is a key requirement for implementing effective predictive maintenance strategies in modern manufacturing environments. In practice, however, estimating RUL from sensor measurements is challenging due to the nonlinear nature of degradation processes and the presence of multiple operating conditions that influence system behavior over time. In this study, we propose a data-driven deep learning framework that combines convolutional feature extraction with bidirectional temporal modeling to improve RUL prediction performance. The proposed approach first employs a one-dimensional convolutional neural network to capture degradation-related patterns from multivariate sensor signals. These features are then processed by a Bidirectional Long Short-Term Memory (BiLSTM) network in order to model temporal dependencies along the degradation trajectory. An additive attention mechanism is further integrated to enable the model to focus on informative operational cycles that contribute more significantly to the final RUL estimate. To improve robustness across varying operational regimes, a condition-aware normalization strategy is incorporated during the preprocessing stage. The proposed framework is evaluated on all four subsets (FD۰۰۱-FD۰۰۴) of the NASA CMAPSS dataset using standard Root Mean Square Error (RMSE) and NASA scoring metrics. Experimental results indicate that the model achieves consistent prediction performance across datasets with different fault modes and operating conditions, while maintaining stable convergence behavior and moderate computational complexity. These findings suggest that the proposed architecture provides a reliable and scalable solution for data-driven prognostics in predictive maintenance applications.

کلیدواژه ها:

Remaining Useful Life ، Predictive Maintenance ، CNN ، BiLSTM ، Attention Mechanism ، CMAPSS