A joint classification-regression method for multi-stage remaining useful life prediction. (January 2021)

• Record Type: Journal Article
• Authors: Wu, Ji-Yan; Wu, Min; Chen, Zhenghua; Li, Xiaoli; Yan, Ruqiang
• Is Part Of: Journal of manufacturing systems. Volume 58(2021)Part A
• Page Start: 109
• Page End: 119
• Publication Date: 2021-01
• Subjects: Prognostic technique -- Remaining useful life -- Multi-stage -- Machine learning; Manufacturing processes -- Periodicals; Production engineering -- Data processing -- Periodicals; Robots, Industrial -- Periodicals; Production, Technique de la -- Informatique -- Périodiques; Robots industriels -- Périodiques; Electronic journals; 670.42
• DOI: 10.1016/j.jmsy.2020.11.016
• Languages: English
• ISSNs: 0278-6125

Abstract

Highlights: Propose a novel multi-stage remaining useful life (RUL) prediction scheme for predictive maintenance in manufacturing and engineering systems. Develop a joint classification-regression framework to leverage specific features in different machine healthy stages. Conduct extensive experiments on real industrial sensory data to validate the improvements over the existing RUL prediction algorithms. Prove the flexibility of the proposed framework by adopting different machine/deep learning algorithms for the healthy stage classification and stage-level RUL prediction. Abstract: Remaining useful life (RUL) prediction plays an important role in increasing the availability and productivity of industrial manufacturing systems. This paper proposes a joint classification-regression scheme for multi-stage RUL prediction. First, the time domain and frequency domain features are extracted from various types of raw sensory data (e.g., acoustic, current, vibration and temperature) to constitute the training data set. Second, the system health stage is classified based on the trained model and real-time sensory data. Third, we perform stage-level RUL prediction with regression algorithm to estimate overall useful life. Distinct from the existing RUL estimation algorithms, the proposed multi-stage remaining useful life (MS-RUL) prediction effectively integrates the machine/deep learning based classification and regression to improve overall estimation accuracy. We conduct the performance evaluation with sensory data from real manufacturing systems. Experimental results demonstrate that the proposed MS-RUL achieves approximately 6.5% accuracy improvements over the state-of-the-art algorithms in the RUL prediction.