Degradation-based Prognostics and Health Management (PHM) for Engineering Systems

Abstract

Prognostics and health management (PHM) methods are popularly used in engineering disciplines for the evaluation of system reliability, the prediction of advent failure, and the mitigation of failure risks. Recent advancements in information and sensor technologies enable the collections of massive condition monitoring (CM) data, which offer unprecedented opportunities for investigating system deterioration behaviors, diagnosis approaches, and PHM techniques using data-driven methods.

This thesis focuses on developing novel methodologies for addressing multiple challenging PHM problems arose from engineering systems with hard failures, on the basis of system deterioration analysis by integrating statistical knowledge and data-driven principles. Three specific degradation-based PHM research problems described as follows have been investigated:

i. Remaining useful life prediction methods for hard failures

In the literature, the joint modeling framework is commonly utilized with the proportional hazards (PH) model for hard failure prediction, which requires a strong PH assumption but may not always hold. To be general, a joint model with extended hazard (EH) is proposed for remaining useful life (RUL) prediction in this thesis.

Besides the modeling of time-to-event data using PH or EH functions, the characterization of degradation signals highly impacts the prediction accuracy. Stochastic processes for degradation signals are widely studied in soft failure cases, such as the Gamma and Wiener processes. However, stochastic processes have not been considered for degradation signals in hard failure problems including those exhibit similar stochastic behaviors. In the proposed joint model with PH, the Wiener process with drift is first adopted to model the degradation signals.

ii. Degradation-based reliability-centered predictive maintenance policy

The effective utilization of predictive information, such as RUL prediction results and the probability of future failure, in system scheduling for hard failures has not been sufficiently studied. In this thesis, a reliability-centered predictive maintenance policy is proposed to maintain systems with high reliability and low cost rates on the basis of its deterioration process.

iii. Predictive analytics of system operating conditions: a wind turbine application
A predictive analytics method is developed for inferring the operating conditions of a complex engineering system, the wind turbine. Based on a large volume of data, data-mining algorithms are first utilized to evaluate the operating conditions of wind turbines. Next, the predictive analytics of its remaining functional time is conducted.

Extensive case studies and simulation experiments are conducted to evaluate proposed PHM methods. The main framework and methods developed in this thesis are possible to be applied in various engineering fields, including manufacturing, energy, healthcare, etc.

Date of Award	14 Aug 2018
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Zijun ZHANG (Supervisor), Qiang ZHOU (Co-supervisor) & Min XIE (Co-supervisor)