Prediction of Remaining Useful Life in MEMS Devices Using a Stacking Model Approach

Abstract

Accurate estimation of the Remaining Useful Life (RUL) of Micro-Electro-Mechanical Systems (MEMS) devices is essential for enhancing predictive maintenance in industrial settings. This paper proposes a robust stacking ensemble model for Remaining Useful Life (RUL) prediction of MEMS devices, integrating sensor and mechanical component data. Publicly available RUL datasets were augmented using Generative Adversarial Networks (GANs), ensuring diverse input data. Key algorithms-Bayesian Ridge, Random Forest, and XGBoost-were identified using Lazy Predict and combined with a Support Vector Regressor (SVR) as the meta-learner. Hyperparameter optimization was performed using the Deep Deterministic Policy Gradient (DDPG) algorithm, offering enhanced efficiency for large datasets and dynamic retraining. The model demonstrated superior performance based on RMSE and R2 metrics compared to traditional methods. Deployment was achieved via a Flask web application integrated with a CI/CD pipeline using GitHub Actions and Docker, ensuring scalability and reproducibility. This research contributes a scalable and efficient framework for advancing predictive maintenance in industrial automation. © 2025 IEEE.