Search

• Prognosis for stochastic degrading systems with massive data is investigated. • A prognosis method is proposed from a data-model interactive perspective. • Deep health indicator extraction and model-driven prognosis are handled interactively. • The remaining life is predicted by modeling the progression of the extracted health indicator. • Advantages in the improvement of the prognosis accuracy are justified. Fast development of various advanced sensing techniques has provided the possibility acquiring massive monitoring data of stochastic degrading systems. In this circumstance, deep learning techniques have become the emerging topic in prognostics under massive data. However, the capability of quantifying the prognosis uncertainty is limited by such methods. In addition, existing studies have attempted to combine deep learning methods and stochastic degradation model based methods, but they are basically handled separately, which cannot guarantee that the extracted health indicator (HI) by deep learning method match the adopted degradation model. To address prognostic issues under massive data, this paper proposes a data-model interaction based prognostic method by constructing a closed-loop prediction mechanism between the HI extraction and degradation modeling. The main idea is to construct a novel optimization objective function related with the prognosis performance so as to establish an interactive linkage among the HI extraction, degradation modeling and prognosis. To do so, the extracted HI is hopefully matching the adopted degradation model. In the implementation process, the stacked denoising autoencoder is used to extract the HI from massive data while the Wiener process is used to model the extracted HI for prognosis. Based on the proposed interactive mechanism, these two parts are synchronously optimized. As such, the proposed method hold promise to integrate the advantages of the deep learning in handling the massive data and stochastic process method in quantifying the prognosis uncertainty. Finally, the effectiveness and superiority of the proposed method are demonstrated by a case study associated with turbofan engines. [ABSTRACT FROM AUTHOR]

Condition-based maintenance for a degrading system has been attached great importance in reducing unexpected failures and enabling the safe operation of the system. In this paper, we consider a condition-based replacement problem with observed degradation signals for the determination of the optimal replacement time of the system. The observed degradation signals contain some health related information of the system so that the system's health state may be more accurately estimated and the future degradation progression of the system can be predicted. Based on the predicted degradation state, the concerned replacement problem is formulated in the framework of the Markov decision process and then a new degradation-modeling framework based on maximum likelihood estimation is proposed to analyze the degradation signals and to predict the future degradation state of the system. To solve the proposed condition-based replacement decision problem, we analyze structural properties of the optimal replacement policy and an optimal monotonic control limit solution policy is developed for condition-based replacement. Finally, a case study is provided to illustrate the proposed optimal replacement method. [ABSTRACT FROM AUTHOR]

• Box-Cox transformation is introduced in nonlinear degradation modeling. • Transformed degradation data are modeled by wiener process. • Two-stage estimation procedure is presented to identify the model parameters. • Closed-form remaining useful life distribution is derived for prognostics. • The proposed method is validated by degradation data of batteries and LCDs. Transforming nonlinear degradation paths into nearly linear ones has been widely used for nonlinear degradation modeling and prognostics. However, types of the current transformation functions are difficult to determine. This paper addresses issues in nonlinear stochastic degradation modeling and prognostics from a Box-Cox transformation (BCT) perspective. Specifically, the BCT is first used to transform the nonlinear degradation data into nearly linear data, and then the Wiener process with random drift is utilized to model the evolving process of the transformed data. To determine the model parameters, a two-stage estimation procedure is developed including offline stage and online stage. In the offline stage, the parameters are determined via maximum likelihood estimation method based on the historical degradation data and such estimated values are used to initialize the online stage. During the online stage, the Bayesian method is adopted to update the model parameters using the data of the degrading system in service, in which the hyperparameters are updated by the expectation maximization algorithm. A closed-form solution to remaining useful life with updated model parameters is further derived for prognostics. Finally, case studies for lithium-ion batteries and liquid coupling devices are provided to demonstrate the proposed approach. [ABSTRACT FROM AUTHOR]

• Physics-informed data-driven remaining useful life prediction methods are reviewed. • Physics-informed machine learning based prognosis methods are detailed. • Individual strengths and weaknesses of different kinds of methods are discussed. • Challenges and possible opportunities are provided to steer the future development. Remaining useful life (RUL) prediction, known as 'prognostics', has long been recognized as one of the key technologies in prognostics and health management (PHM) to maintain the safety and reliability of the system, and reduce the operating and management costs. Particularly, thanks to great advances in sensing and condition monitoring techniques, data-driven RUL prediction has attracted much attention and various data-driven RUL prediction methods have been reported. Despite the extensive studies on data-driven RUL prediction methods, the successful applications of such methods depend heavily on the volume and quality of the data, and purely data-driven methods possibly generate physically infeasible/inconsistent RUL prediction results and have the limited generalizability and interpretability. It is noted that there is an increasing consensus that embedding the physics or the domain knowledge into the data-driven methods and developing physics-informed data-driven methods will hold promise to improve the interpretability and efficiency of the RUL prediction results and lower the requirement of the volume and quality of the data. In this context, physics-informed data-driven RUL prediction has become an emerging topic in the prognostics field. However, there has not been a systematic review particularly focused on this emerging topic. To fill this gap, this paper reviews recent developments of physics-informed data-driven RUL prediction methods. In this review, current methods fallen into this type are broadly divided into three categories, i.e. physical model and data fusion methods, stochastic degradation model based methods, and physics-informed machine learning (PIML) based methods. Particularly, this review is centered on the PIML based methods since the fast development of such methods have been witnessed in the past five years. Through discussing the pros and cons of existing methods, we provide discussions on challenges and possible opportunities to steer the future development of physics-informed data-driven RUL prediction methods. [ABSTRACT FROM AUTHOR]