25 Conveyor Belt Wear Life Modelling Callum Webb Professor Melinda Hodkiewicz School of Mechanical and Chemical Engineering Assistant Professor Nazim Khan School of Mathematics and Statistics Stephen Muller, Rick Wilson BHP Billiton Iron Ore Abstract BHP Billiton Iron Ore (the client) owns and operates over three hundred conveyor systems with a total length exceeding two hundred kilometres across its operations in the Pilbara region. The current approach to estimating remaining life of conveyor belts is based on extrapolating thickness worn per unit of time. This has several limitations, such as failing to consider changes in conveyor utilisation or downtime. Five alternative measures of wear rate for conveyor belts have been formulated and investigated. Analysis of data from 114 belts installed on 22 different conveyor systems has shown that these alternatives provide better support for predicting remaining belt life compared to the traditional approach. In particular, throughput based wear rates produce models with greater explanatory power and capture changes in conveyor utilisation and downtime. Predictive models based on all six wear definitions have been developed using multiple linear regression, providing tools for forecasting belt replacements based on conveyor design and operational parameters. 1. Introduction Currently, the client’s approach to planning belt maintenance is based on historical belt replacement data and staff experience. The client wishes to reduce the risks inherent in this process by developing an evidence based understanding of the factors that influence belt wear, and a tested approach to planning replacement. Conveyor belt thickness measurements are conducted periodically as part of routine inspections and maintenance. While this raw data provides trends to estimate remaining life, it has several limitations: • Utilisation in terms of conveyor throughput and running time is not taken into account, compromising the accuracy of predictions. • The data currently considered are not sufficient to allow for accurate and confident forecasting of replacements. • Wear rates can only be estimated as thickness data becomes available. • Belt life on similar, new installations cannot be predicted. The aim of this project is to develop a data-based tool for predicting remaining belt life, and a better understanding of the factors that affect the rate of conveyor belt wear. Multiple regression techniques similar to those used by Badisch, Ilo and Polak (2009) and Krishnaswamy and Krishnan (2002) were used to develop predictive models.