Remaining Useful Life Estimation of Slurry Pumps Using the Health Status Probability Estimation Provided by Support Vector Machine

The slurry pump is one of the most important machines and widely applied in oil sand industries, mining, waste treatment, etc. The mixtures transported by the pumps include the solids as well as the liquids with different volume and hardness that make the pumps work under abrasive and erosive environment. This would cause the continuous wear of the components, especially the impeller, in the pumps. As a result, the efficiency and useful life will be greatly reduced. Every unexpected failure of slurry pumps could cost companies high up to millions of dollars. To avoid this problem, traditional scheduled maintenance strategies are usually adopted but it can not warn the impending failure and sometimes the components are replaced when they are still in healthy status. Consequently, effective condition monitoring and online health status assessment methods are of great significance and should be developed to conduct timely and effective slurry pump fault diagnosis and prognosis. In this paper, an effective data driven technique for estimating the remaining useful life of slurry pumps are developed based on the health status probability estimation obtained by Support Vector Machines (SVMs). The signals collected by the sensors installed on an industrial slurry pump are used for analysis. The results show that frequency band selection and the position of sensors have some effect on the useful information acquisition and that SVM has superior performance in industrial data processing.