Time-frequency analysis and support vector machine in automatic detection of defect from vibration signal of centrifugal pump

Abstract

Centrifugal pumps operate at moderate to high speed. Contamination in the fluid in terms of solid particles and chemically reactive substances causes damage to the impeller, casing, and seals. A defect in bearing due to improper lubrication, adverse loading and manufacturing defect may also affect the performance of the pump. Hence there is a need to develop a reliable procedure for defect identification in the centrifugal pump. A robust automated signal processing algorithm is proposed for the purpose. Features sensitive to defective conditions are extracted from raw signal and scale marginal integration graph. The genetic algorithm (GA) is used to find the optimal parameters of support vector machine (SVM). Using the optimal parameters, training of SVM is carried out for the learning of defective conditions of the pump. After training, features are applied to SVM for the identification of the defective condition of the pump. The performance evaluation of the proposed method is made using receiver operating characteristics graph and is found to be reliable. The overall recognition rate of the proposed method in identifying the specific conditions of the pump is 96.66%. In this work, an attempt is also been made to reduce the training time of GA-SVM model.