Development of Linguistic Rules Diagnosis of Failure in Centrifugal Pump for Use in Expert System

Abstract

Operational failures in centrifuge pumps could be hydraulic or mechanical. However, most of these mechanical and hydraulic failures are connected cause of their operational nature and finding the right cause is due to considering numerous mechanical and hydraulic signs and parameters in pumps. On the other hand, due to non-linear and fluctuant behavior of pumps in the matter of time and not precise and clear data from a complicated system such as a petrochemical unit installations, investigating the failure possibilities in pumps by exact mathematical equations is so hard. However, the expert staff in maintenance could find the cause of failure by getting benefit of imprecise and verbal rules of "if-then" and the awareness of the non-clearance in system. However, it should be noted that this process takes time and also it is so dependent on the knowledge of expert staff, also human error can cause wrong decision-making and it declines reliability in the system. Therefore, this study is looking forward to find a collection of imprecise and verbal rules by using the reference book of "OREDA", also the pump handbook and the knowledge of expert staff in maintenance. With the acquisition of knowledge and Extraction of linguistic rules from taking effect between critical failure modes of mechanical and hydraulic parameters of process pump is created the relationship between these parameters and verbal rules, also by using a neural network of MLP category for the cases to train the neural network. Obtained results in companion with the average on accuracy, precision and recall parameters would be the approval of our method.