Centrifugal pumps (CPs) are crucial components in many plant operations. However, they are susceptible to failures due to mechanical faults and/or fluid flow abnormalities. These faults not only affect the CP system but also affect the systems delivering flow to it or receiving flow from it. Therefore, it is extremely crucial to recognize the faults and estimate their severity during operation, so that a corrective action may be initiated. In the present work, an attempt has been made to develop a flexible algorithm based on support vector machine (SVM) suitable to classify CP faults, like the suction and discharge blockages (with varying severities), impeller defects, pitted cover plate faults and dry runs. Also, a combination of mechanical faults (impeller defects and pitted cover plate faults) and suction and discharge blockage faults are considered. For the sake of classification, the CP vibration data and the motor line-current data are generated in time-domain for each fault experimentally. Furthermore, industrially operating CP signatures cannot be immune to noise generated from other operating equipment in the premises. Hence, to assess the robustness of the developed methodology, signals are corrupted by adding 5%, 10% and 25% additive white Gaussian noise. The developed algorithm is tested with corrupted data. The efficiency of fault predictions obtained while testing with noisy and non-noisy data are compared. The results are very promising and carry a high potential for industrial applications.

This content is only available via PDF.
You do not currently have access to this content.