Meta-Learner Blended Ensemble of Multi-Scale GRU-CNN for Remaining Useful Life Forecasting

With advancements in engineering and technology, modern systems are becoming progressively more complex. These composite systems are employed in each and every industry- from healthcare to automation and manufacturing. To ensure their sustainability and reliability, integrating effective Prognostics and Health Management (PHM) systems has become necessary. This allows for the prediction of the system's condition and for making advanced maintenance decisions in order to reduce downtime and enhance system reliability. This prevents breakdowns and helps to forecast failures and estimate the Remaining Useful Life (RUL) of components. This work presents a hybrid model combining Gated Recurrent Units (GRUs) and Convolutional Neural Networks (CNNs) for accurate RUL prediction. The multi-scale CNNs are utilised for spatial feature extraction, and the GRUs are employed for temporal dependencies. This helps the model effectively handle complex datasets. Furthermore, a blending ensemble of multiple base models and a meta-learner was applied. Particle Swarm Optimization (PSO) is used to optimise the hyperparameters. The model demonstrates superior results on NASA's turbofan engine degradation dataset, offering a promising solution for predictive maintenance and estimation of RUL in diverse applications.