A new Wasserstein distance- and cumulative sum-dependent health indicator and its application in prediction of remaining
- PDF / 2,148,933 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 96 Downloads / 167 Views
(2020) 42:479
TECHNICAL PAPER
A new Wasserstein distance‑ and cumulative sum‑dependent health indicator and its application in prediction of remaining useful life of bearing Jiancheng Yin1 · Minqiang Xu1 · Huailiang Zheng1 · Yuantao Yang1 Received: 21 June 2019 / Accepted: 13 August 2020 © The Brazilian Society of Mechanical Sciences and Engineering 2020
Abstract The safety and reliability of mechanical performance are affected by the condition (health status) of the bearings. A health indicator (HI) with high monotonicity and robustness is a helpful tool to simplify the predictive model and improve prediction accuracy. In this paper, a new health indicator based on the Wasserstein distance (WD) and cumulative sum (CUSUM) is proposed. First, a standard normal signal is simulated as the reference. The second step is to calculate the frequency distribution histogram of the reference signal and that of monitoring signals for the bearing. The next step is to obtain the WD between the frequency distribution histogram of the reference signal and that of the monitoring signal. Finally, the fluctuation of the WD is amplified by applying the CUSUM. The performance of the proposed HI is evaluated by testing three run-tofailure datasets. The results show that the proposed HI has better monotonicity and robustness and can be effectively used to predict the remaining useful life of bearings. Keywords Health indicator · Wasserstein distance · Cumulative sum · Remaining useful life · Life prediction
1 Introduction As an important component of rotating machinery, the health of bearing has an important impact on the reliability and safety of rotating machinery [1]. Hence, the effective and reliable remaining useful life (RUL) prediction of bearing is essential to formulate a timely maintenance schedule and improve the reliability of rotating machinery. To date, most of the prediction methods can be generally divided into either model-based or data-driven methods. Compared with model-based methods [2, 3], the data-driven methods do not need to construct a complicated physical model and have been extensively applied in the RUL prediction of bearings. For example, Bastami et al. [4] utilized the artificial neural network and wavelet packet features to estimate the RUL of bearing. Zhu et al. [5] predicted the RUL of bearing by using the multi-scale convolutional neural Technical Editor: Wallace Moreira Bessa. * Minqiang Xu [email protected] 1
Deep Space Exploration Research Center, Harbin Institute of Technology, Harbin 150001, China
network. Ren et al. [6] proposed a prediction framework for bearing based on auto-encoder and deep neural networks. However, instead of using direct monitoring signals, the above-mentioned data-driven prediction methods all need to utilize the indirect indicators to reflect the health status of bearing. Therefore, how to develop effective health indicators (HI) is crucial to simplify the data-driven-based prediction models and improve the accuracy of prediction [7]. The HI can be divided into p
Data Loading...
