A Fusion Approach with Application to Oil Sand Pump Prognostics

In industrial field, slurry pumps are widely used to transport mixtures of abrasive solids and liquid in wet mineral processing operations. As working under adverse environment, the performances of slurry pumps are often degraded or the pump system even fails unexpectedly. Therefore, significant resources are invested in programs maintenance to avoid unscheduled downtimes and ensure that the required performance of system is maintained at the maximum efficiency. This work is developed from a particular need in oil-mining industry to monitor the health of slurry pumps. In this study, relevance vector machines (RVM) are utilized to predict the remaining useful life (RUL) of field impellers combined with two-summed exponential function. To solve the non-stationary problem emerged in the data, a novel feature extracting process is designed. Finally, one field dataset is applied to evaluate the effectiveness of the proposed prognosis model. The application result shows good performance on degradation trend and remaining useful life prediction of the pump impellers. Hence the model can well solve the problem when to replace the related components before they area absolutely out of service to avoid the sudden downtime.