Deep LSTM Enhancement for RUL Prediction Using Gaussian Mixture Models

• Published in: Automatic control and computer sciences Vol. 55; no. 1; pp. 15 - 25
• Main Authors: Sayah, M., Guebli, D., Noureddine, Z., Al Masry, Z.
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 2021; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Clustering; Computer Science; Control Structures and Microprogramming; Life prediction; Machine learning; Modular systems; Probabilistic models; Propulsion systems; Useful life; RUL; deep learning; Gaussian mixture; expectation-maximization; LSTM network; prognostics and health management
• ISSN: 0146-4116; 1558-108X
• DOI: 10.3103/S0146411621010089

This paper introduces a new deep learning model for Remaining Useful Life (RUL) prediction of complex industrial system components using Gaussian Mixture Models (GMMs). The used model is an enhanced deep LSTM approach, for which Gaussian mixture clustering is performed for all collected sensors data and operational monitoring information. This distribution-based clustering using the hyperparameter ε leads to an adequate deep neural network for RUL prediction. An expectation-maximization algorithm was implemented to configure the deep LSTM network for RUL estimation. The proposed Gaussian mixture Clustering-based deep LSTM model for useful life prediction of the industrial components is trained and tested on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) datasets. The experiments of the enhanced deep LSTM model show clearly the relevance of using Gaussian mixture clustering for quality improvement of RUL prediction through deep LSTM models. ( https://github.com/sayahmhgithub/EnhancedLSTM4RUL.git ).