Background and Motivation HUSSIN

1 1 CHAPTER 1: INTRODUCTION

1.1 Background and Motivation

Operational equipment such as pumps, conveyors, motors and others generate a large number of signals that can be monitored. As the parts of the equipment move and rotate, they produce vibration, sound and may change temperatures and pressures. In addition, the condition of the oil used as a lubricant also has a significant effect on the working condition of the equipment . These signals can act as maintenance indicators, which could be used to describe the key relationship between equipment condition and a maintenance decision. Using these condition-monitoring signals, we could assess equipment condition in its present operating environment and maintenance actions are carried out only when necessary. This could result in a safe, effective and economical maintenance operation. Furthermore, the importance of condition monitoring in maintenance has been increasingly recognized, due to the availability of modern condition monitoring technologies. With these technologies, continuous conditional indicators are provided, which can help maintenance departments to develop, measure and improve maintenance actions in the organization. However, most of these developments focused on the technical aspects of condition monitoring Rao, 1995; 2001; 2002; 2003, such as advanced tools and techniques in monitoring technologies, signal processing, data acquisition and interpretation. These are mainly for diagnosis on issues of what to do, but the issue of when to do it received less attention. Yet, deciding when to do it preventive repair also requires some justifications. It is a common practice that a certain threshold value has to be set for a chosen condition monitoring parameter of the equipment. The threshold level may be set up based upon manufacture recommendations, personal experiences or other subjective criteria to provide a warning that a significant change has occurred and immediate action needs to be taken. From the condition-based maintenance perspective, this threshold level may not be optimal. This is due to the fact that each machine is an individual, which may behave differently even though they are supposed to be identical, 2 so such a common threshold level is not appropriate in most cases. Furthermore, setting up a common threshold level may pose a difficulty for maintenance actions, and would incur extra costs. For example, if we set up a lower threshold value, it may result in needing more early replacements, and could waste much of the useful remaining lifetime of the equipment. In contrast, if we set up a higher threshold value, it will result in an increase of machine failures. Thus, from an economical or safety point of view, the basic idea is to use all the monitored data current and past of a particular system, and make corresponding decisions to maintain the production equipment based upon cost, safety or other criteria. There is obviously a need for an appropriate model to aid such a decision support for plant maintenance managers. It is noted however, only a very few tools are available in the market and only a small amount of research has been devoted to this area. Thus, developing a suitable model for effective decision-making in condition-based maintenance is regarded as an important addition to the subject. To achieve our primary goal as described above, this research aims to predict the underlying state of a piece of production equipment, given its observed condition monitoring measurements at each monitoring point to date. How to define the underlying state of the equipment is a difficult issue and in this thesis, we used residual time and wear as examples. The residual time is chosen due to the fact that it represents an important characteristic in deciding an appropriate maintenance decision such as when to replace the equipment Reinertsen, 1996. Similarly, cumulative wear could also be used, as it is a direct indication of the deterioration process. If these underlying states can be predicted, maintenance actions including manpower, equipment and tools, and spare parts can be planned and scheduled Al-Sultan and Duffuaa, 1995. In this study, by using available monitored-condition information, we believe that prediction of the underlying state could be much better than the conventional method, which uses only the current age to predict the remaining life or wear of a machine. It is noted that the result of the underlying state prediction can only be described by a probability distribution due to the fact that it is random and unknown. This poses the question of how we can obtain such a probability distribution of the underlying state. This is the main question addressed in this thesis. It is also highlighted that this 3 distribution is a key element in the subsequent maintenance decision model that we aimed to achieve. However, there are several critical challenges, which make the establishment of the underlying state distribution a difficult problem. The first challenge is how we can define the failure, based upon the chosen underlying state and its relation to the observed monitoring data. That is, how well the condition monitoring data reflects the deterioration or failure process. Hence, understanding and modelling the processes of deterioration and failures themselves become essential in this research. The second challenge is that we may have rich sources of condition-monitoring data but very little failure information. Also as reported by Ascher et al. 1995 the data captured from the field suffers from many problems and it makes the data manipulation task both important and challenging. Hence, most of the effort has been placed on understanding, manipulating and preparing the data for development of the model. The third challenge is that we may have a good theoretical model, but can it be implemented in real applications? Scarf 1997 surveyed the available papers on modelling condition-based maintenance and appealed for the applicability of the models in practice. Applicability in this case implies how the complexity and computation time can be reduced. Therefore, in developing the model, a few assumptions have been made, not only to simplify our modelling but also to ensure practical advantages. To overcome all the challenges stated above, the objectives of this research are as follows: 1. To investigate the appropriateness of the defined state used in the model to quantify the equipment condition. 2. To identify ways in which established model can be improved. 3. To explore approximation solution for the analytical model. 4. To investigate some application of the model subjected to the chosen condition monitoring data and their relations toward maintenance planning and scheduling. 4 The following section contains an outline of the thesis, and gives an overview of the work.

1.2 Organization of the Thesis