Predicting Remaining Useful Life (RUL) of Wind Turbine Gearboxes via Hybrid Bayesian-LSTM Network with Adaptive Feature Rescaling

**Abstract:** This paper introduces a novel methodology for accurately predicting the Remaining Useful Life (RUL) of wind turbine gearboxes, a critical component for renewable energy efficiency and cost-effectiveness. Current state-of-the-art approaches struggle to effectively integrate diverse sensor data and dynamically adjust to changing operating conditions leading to RUL prediction inaccuracies. We propose a Hybrid Bayesian-LSTM (HB-LSTM) network that combines the probabilistic modeling capabilities of Bayesian inference with the temporal sequence learning of Long Short-Term Memory (LSTM) networks. Furthermore, we incorporate an Adaptive Feature Rescaling (AFR) module to dynamically normalize sensor data, mitigating the influence of operational variability and enhancing model accuracy. The proposed HB-LSTM-AFR system offers a 20% improvement in RUL prediction accuracy compared to existing deep learning models and is immediately commercially viable for predictive maintenance programs.