This thesis details the use of a multivariate approach to study the reliability of mechanical equipment in a cement production plant. The cement plant, located in Kenya, contains a large number of mechanical equipment for which stoppage records, production totals, and vibration measurements have been recorded. The primary objective of this research is to build integrated predictive models, incorporating all three data sources, that can be used to predict the reliability and behavior of the equipment. The motivation for this research stems from the cement plant, which has a desire to improve their existing maintenance strategies based on insights about equipment reliability. This thesis will describe the process of building an analysis framework, and demonstrate the use of this framework to generate maintenance decision support.
The methodology of this thesis is tailored to the specific objectives of the research problem, but can be applied to similar problems in future research. The data collection and pre-processing steps are the foundation of the analysis, during which the structure of the data is assembled in preparation for analysis. During the descriptive analysis, preliminary insights are acquired from each of the data sources, and critical elements of the plant are selected as the focus of later models. After critical equipment is identified, the data are integrated to provide a complete record of all stoppage events, production rate, and vibrations observations. The integrated data provide a vast opportunity for reliability modelling, which is explored through the use of non-parametric, semi-parametric, and fully parametric survival analysis techniques. Additionally, several classification models are used to identify the extent to which the integrated data is able to predict future maintenance actions following failure.
The analysis and results are presented with the intent to demonstrate the implications of each model with respect to maintenance decision support. Although each model is estimated based on a specific subset of the plant, the analysis framework can be repeated for any equipment for which data is available.