Organization of the Thesis

4 The following section contains an outline of the thesis, and gives an overview of the work.

1.2 Organization of the Thesis

This thesis is organized in nine chapters. In Chapter 1, we first introduce the common problems arising in condition-based maintenance and their challenges, which motivate us to carry out this research. In Chapter 2, a brief introduction to maintenance and management strategy in undertaking maintenance actions is presented. We introduce the concept of condition monitoring and condition-based maintenance in detail. We investigate the literature and current monitoring techniques used within industry. It is noted that the scope of the literature review is mainly concerned with the modelling aspect of decision making in condition-based maintenance, thus several approaches and concepts related to this issue are discussed. Chapter 3 presents a new development of a conditional residual time model that is different from the literature as reviewed in Chapter 2. A discrete state space is used to define the condition of the operational equipment. The methodology and formulation used in this development are discussed, and simulation studies with numerical results are presented. In this chapter, we show how the development of this model can be used to predict the initial point of a random fault in a system. The process of model fitting and testing using an actual dataset is shown in Chapter 4. Also in Chapter 4, we had an attempt to compare our results with a statistical process control based method, which had been developed in a previous study conducted by Zhang 2004. In Chapter 5, we investigate numerical approaches as an alternative solution to the model developed in Chapter 3. Approximate approaches, such as grid based and particle filtering, are discussed. We demonstrate these approaches using simulated and actual datasets from Chapter 4. Chapter 6 presents further developments for predicting the conditional residual time using actual oil monitoring data. The data collected is not organised, with missing 5 values and many unexplained, so we have to re-organise the data into a format that is suitable for our model. One of the processes of re-organising the data is by transforming the collected data into a measure known as the total metal concentration. In general, three components of data are available to us such as the metal elements, lubricant performances and contaminant indicator while conducting the oil analysis programme. In this chapter, we only used the metal elements component indicator as it provides important information about the wear of the internal engine parts. Although there are many metal concentrations in the oil sample, not all of them are useful. Hence, a technique to reduce the dimension or size of metal elements is discussed. In addition, several procedures dealing with incomplete data are also presented. The model developed is fitted to the data and the numerical results are given. We carried out several tests to show the robustness of the model developed, and produced significant results. In Chapter 7, the model developed in Chapter 6 is enhanced with more monitoring information. Using all three components presented in the oil-monitoring data, we grouped them into two groups. Here, we consider that these two types of condition- monitoring information are not correlated with each other but have different relationships with the residual life. The assumption and formulation for the new model are discussed, with a focus on interpreting and preparing the data required in the new model. As the lubricant performance and contaminant indicators are significantly correlated, we carry out an independent component analysis to separate each variable. The model is supplied with the actual data and the numerical results are given. Several other tests are also conducted. As seen from Chapters 3, 4, 6, and 7, the residual time is used to characterize the failure or deterioration process of the production equipment. But, in Chapter 8, we attempt to model the deterioration process using a measure called ―wear‖, without using the residual time concept. To do this, we introduce a model that uses a continuous random variable to represent the process of deterioration of the system at any monitoring point. This is done using a beta distribution and allows us to have a more generic wear model that can be applied to many situations. This model is tested with the same oil data used in Chapter 6, and the numerical result is given. Furthermore, an analysis between this model and the residual time model in Chapter 6 is made. 6 Finally, in Chapter 9, the findings of the thesis are summarised and recommendations for future work are also discussed. 7 2 CHAPTER 2: LITERATURE REVIEWS

2.1 Introduction