Study on Degradation-based Prognostics and Health Management Methods of Storage Electromechanical Equipment

Abstract

Prognostics and health management (PHM) as an important safeguarding technology, has been widely used in many engineering fields for providing theoretical and technical support for ensuing the normal and safe operation of equipment, reducing the cost of maintenance, and improving the productivity. Since the failure event data is both cost and time consuming, and the individual product performance is significant different, traditional reliability assessment cannot meet the needs of specialized equipment health assessment. As an inherent attribute, the generation and aggravation of degradation throughout the entire life cycle of the electromechanical equipment, which is the direct cause of performance decreasing and system fault. Effectively extracting degradation characteristics, accurately identifying degradation status, precisely establishing degradation model for individuals, and timely predicting the development trend of degradation is able to improve product safety and usability, and avoid failures and significant disasters. Therefore, the study on degradation-based PHM methods and applications have important significance on both theoretical and practical perspectives.

The implementation of degradation-based PHM of electromechanical equipment is directly dominated by the degradation mechanism of the key component and susceptible to the external factors. However, the complicated degradation mechanism of the electromechanical system, especially for complex and precise instruments and systems, makes it difficult to directly obtain degradation characteristics under specific usage conditions. Dynamic external factors such as the environmental factors, operation conditions, and loads increases the instability of degradation and makes it difficult to determine the degradation level and predict the degradation trend under varying working conditions. To overcome the above limitations, our study carries research on the degradation-based PHM methods and applications of electromechanical equipment, thoroughly studies the degradation mechanism of complex electromechanical equipment and degradation modeling method at variable operation conditions, overcomes the theoretical and technical problems during the implementation of degradation-based PHM, and finally realizes the failure prognosis and remaining useful life (RUL) prediction of electromechanical equipment based on degradation information.

Aiming at the problems that the degradation indicator of complex electromechanical equipment is difficult to directly obtain and the degradation characteristics is not clear, an indirect degradation feature extraction and quantitative identification method based on degradation mechanism analysis is proposed. Taking the hard disk drive (HDD) as research object, the dead-disk interface dynamics was employed to extract degradation information. A two-degree-of-freedom head-disk interface contact model was established to conduct force analysis, and the Archard wear model based correlation model that correlates head-disk contact force and the vibration response spectrum was proposed that considers the effect of nonlinear contact force, adhesion force, and contact force on slider vibration. The dissipated energy of different vibration modal were extracted as the degradation indicators, respectively. Finally, correlation analysis was employed to confirm the dominant vibration modal of slider wear degradation and a quantitative identification model of slider wear degradation was established by regression analysis.

The response of incipient degradation stage is weak and sometimes confused together with other responses which makes it difficult to distinguish. Compared to severe defect, the energy of the response signal of incipient degradation is low and easily be neglected when classification, especially the background noise is large, however, detecting the defect at early stage is able to provide more time margin to make maintenance decision. To solve these problems, an enhanced trace ratio linear discriminant analysis (ETR-LDA) for identifying incipient degradation states is proposed. ETR-LDA inherits the excellent properties of the trace ratio linear discriminant analysis (TR-LDA) that utilizes the Fisher's discriminant criterion to search for the most separable projection space and introduces a trace ration operator to simplify high-dimensional matrix operation. By introducing the confusing patterns to the original objective function of TR-LDA as a regular term, ETR-LDA is supposed to improve the separability of the new projection space. The modified objective function is proved to be able to converge to the global optimal solution while simultaneously ensuring the value global between-class distance and the confusing between-class distance, which overcomes the limitation of the confusing patterns on global separability. The identification results of HDD slider wear degradation show that the proposed method can effectively distinguish the incipient degradation status of the slider and improve the accuracy of health monitoring at early stage.

The health indicator and degradation trends of key components of electromechanical equipment are prone to working loads and working conditions, which increases the difficulty of the implementation of PHM. To amend that, a condition monitoring and degradation modeling methodology at varying working conditions is proposed. Taking lithium-ion battery as research object, its whole life cycle degradation data of at different discharge rates was acquired via designing and carrying out a switching discharge rates degradation test. Based on the capacities of a battery at different discharge rates at a specific time, the divergence of capacity caused by discharge rate is revealed and depicted by introducing the Peukert's law. From a whole life cycle perspective, a modified Peukert's law with dynamic Peukert coefficient is proposed, which follows a second-order polynomial respect to battery degradation and eliminates the battery performance divergence on cycle number scale. A condition monitoring and health evaluation framework for lithium-ion battery based on dynamic Peukert's law is proposed, which establishes the mapping relationship of battery degradation status under different operating conditions. The proposed model reveals the degradation characteristics of lithium-ion batteries under varying operating conditions, provides convenience for better understanding of battery performance, and is an effective solution for lithium-ion battery condition monitoring and health management under varying operation conditions.

Considering the randomness, individual variability, and the non-linear characteristics caused by the accelerated degradation test in the degradation process, a stochastic process based RUL prediction model with adaptive failure threshold is proposed. Taking the lithium-ion battery as research object, the Wiener process is utilized to characterize the randomness of the degradation increment within a specific time step, and the dynamic Peukert's law is employed to establish the non-linear mapping relationship between the battery health indicators at accelerated degradation test and normal stress degradation test which helps to obtain the dynamic failure threshold updated adaptively with degradation. An adaptive parameter updating method combining Rauch-Tung-Seriebel (RTS) smoothing and expectation maximization (EM) for nonlinear degradation model is proposed, which solves the problem that the parameter estimation during the nonlinear degradation is sensible to the initial value and susceptible to outliers. The RUL prediction results of lithium-ion batteries under different discharge rates verify that the proposed method can effectively predict the RUL under normal stress with the help of accelerated degradation data.

Our study conducts research within the scope of degradation-based PHM from the aspects of degradation mechanism exploration, degradation feature extraction, varying working condition degradation modeling, and RUL prediction. The research focuses on typical electromechanical equipment, that are HDD and lithium-ion battery, solves problems such as feature extraction and fault diagnosis at incipient stage, degradation condition monitoring and modeling at varying working conditions, and RUL prediction via accelerated degradation data, and provides effective solutions for degradation-based PHM of electromechanical equipment.

Date of Award	7 Aug 2020
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Kwok Leung TSUI (Supervisor), Yanyang Zi (Supervisor) & Yanyang Zi (External Supervisor)