No abstract

Degradation is an unavoidable phenomenon in industrial systems. Hidden Markov models (HMMs) have been used for degradation modeling. In particular, segmental HMMs have been developed to model the explicit relationship between degradation signals and hidden states. However, existing segmental HMMs deal only with univariate cases, whereas in real systems, signals from various sensors are collected simultaneously, which makes it necessary to adapt the segmental HMMs to deal with multivariate processes. Also, to make full use of the information from the sensors, it is important to differentiate stable signals from deteriorating ones, but there is no good way for this, especially in multivariate processes. In this paper, the multivariate exponentially weighted moving average (MEWMA) control chart is employed to identify deteriorating multivariate signals. Specifically, the MEWMA statistic is used as a comprehensive indicator for differentiating multivariate observations. Likelihood Maximization is used to estimate the model parameters. To avoid underflow, the forward and backward probabilities are normalized. In order to assess degradation, joint probabilities are defined and derived. Further, the occurrence probability of each degradation state at the current time, as well as in the future, is derived. The Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset of NASA is employed for comparative analysis. In terms of degradation assessment and prediction, the proposed model performs very well in general. By sensitivity analysis, we show that in order to improve further the performance of the method, the weight of the chart should be set relatively small, whereas the method is not sensitive to the change of the in-control average run length (ARL).

With the development of automation technologies, data can be collected in a high frequency, easily causing autocorrelation phenomena. Control charts of residuals have been used as a good way to monitor autocorrelated processes. The residuals have been often computed based on autoregressive (AR) models whose building needs much experience. Data have been assumed to be firstorder autocorrelated, and first-order autoregressive (AR(1)) models have been employed to obtain residuals. But for a pth-order autocorrelated process, how the AR(1) model affects the performance of the control chart of residuals remains unknown. In this paper, the control chart of exponentially weighted moving average of residuals (EWMA-R) is used to monitor the pth-order autocorrelated process. Taking the mean and standard deviation of run length as performance indicators, two types of EWMA-R control charts, with their residuals obtained from the pth-order autoregressive AR(p) and AR(1) models, respectively, are compared. The results of the numerical experiment show that for detecting small mean shifts, EWMA-R control charts based on AR(1) models outperform ones based on AR(p) models, whereas for detecting large shifts, they are sometimes slightly worse. A practical application is used to give a recommendation that a large number of samples are necessary for determining an EWMA-R control chart before using it.

Accurate prediction of remaining useful life (RUL) plays an important role in the formulation of maintenance strategies. However, due to the diversity of the failure mode of equipment, there are significant differences between the degradation data, which greatly affects the accuracy of RUL prediction. In this case, an ensemble prediction model considering health index-based (HI-based) classification is proposed in this paper. Firstly, the stacked autoencoder (SAE) is employed to construct the HI. Then, the time window is used to sequentially process the HI sequence, so that many data segments with the same length can be achieved. To differentiate the data with the similar degradation process, K-means and Xgboost are selected to construct offline and online data classification models respectively. Finally, according to the results of the data classification, the ensemble model that integrates multiple machine learning methods are separately trained and then adaptively used for RUL prediction. In addition, integrating multiple methods helps to improve the generalization ability of the model. The NASA C-MAPSS dataset is applied to verify the effectiveness of the proposed method, and the results show that the proposed method achieves a higher prediction accuracy and shorter computational time than other existing prediction models.