Failure prediction of critical mine machinery

Abstract

The South African mining industry is under immense economic pressure and has been following a negative trend over the last 30 years. It is thus imperative to regain a position at the front end of technological advances in the global market. Today’s engineering systems have become increasingly complex to be designed and built while the demand for reliability continues. Various studies have involved reliability analyses in industrial fields based on the evolution of the failure rate of equipment. Proper maintenance strategies are required to achieve maximum reliability. Both maintenance and reliability have roots in each other’s territory. Ultimately, optimising maintenance methods will yield increased system reliability. Typically, maintenance strategies in the mining industry are reactive and cyclic, which can lead to failure creep. These strategies can be improved using a proactive approach, which requires forecasting models. Examples of forecasting models include the Weibull models and the Crow-AMSAA (NHPP) model, which can be represented stochastically by the Weibull process. Weibull distributions have proven to successfully model failure patterns for various product failure mechanisms, with the most common being the standard two-parameter model. The Crow-AMSAA model is designed for use on systems where the analysis of the reliability growth process is essential. These models have been widely used in the reliability engineering environment, but has not yet been implemented on equipment operating on a deep-level mine in South Africa. A new practical method was developed for predicting equipment failure using a unique combination of these models, conditional and quantile analyses. The method was applied to mine dewatering pumps, compressors and scraper winches. For each of these components, the base method was adapted to maintain compatibility with the different data set types. The methodology, application and results were documented as three separate articles, which are summarised below. Mine dewatering pumps – Article I (Appendix B). A new failure prediction methodology was developed and validated on four pumps, operating at different depths. The Weibull analysis was applied to previously installed pumps and successfully predicted the failures of newly installed pumps. Furthermore, the results showed a quadratic relationship between the characteristic life and operating depth of a dewatering pump. Significantly, this correlation quantifies all the factors that contribute to the service life of a dewatering pump. Mine compressors – Article II (Appendix C). The base methodology was modified to be more generic and applicable to highly reliable equipment. The new methodology was applied to multiple compressors operating at different compressor houses. Furthermore, the results were validated using the leave-one-out cross-validation method. The model successfully predicted the next component breakdown based on the current survival age of each compressor, which yielded an overall average prediction accuracy of 92%. Mine scraper winches – Article III (Appendix D). The methodology was integrated with the Crow-AMSAA (NHPP) model to predict the number of failures. The ability of the Crow-AMSAA (NHPP) model to make accurate predictions on multi-failure mode systems was leveraged to adjust parameters within the base methodology to predict the actual failure times. An average failure prediction accuracy of 94% was achieved.