Centrifugal pump impeller defect identification by the improved adaptive variational mode decomposition through vibration signals

Abstract

An ameliorated salp swarm algorithm (ASSA) is proposed to enhance the exploration and exploitation stages of the basic salp swarm algorithm using the concept of opposition based learning and position updation. These concepts not only resolve the issue of slow convergence but also reduces the computation time and circumvent strucking in local minima. Based on ASSA, an improved adaptive variational mode decomposition (VMD) method has been proposed to identify the impeller fault in the centrifugal pump. The optimal combinations of VMD parameters: mode number and quadratic penalty factor, are selected adaptively to decompose the vibration signal. On decomposition, the sensitive mode is identified for the extraction of fault features. The basis of identification of sensitive mode is a maximum value of weighted kurtosis. The ranking of fault features is done by Pearson correlation coefficient (PCC). The ranked features train the extreme learning machine (ELM) model and further, the model is tested for fitness evaluation. The overall training accuracy of the ELM model is found to be 100% with 0.0012 seconds of training time. The testing accuracy was found to be 97.5%. Results obtained at twenty-three classical benchmark functions and the Wilcoxon test validate the efficiency and superiority of the proposed ASSA algorithm in the diagnosis of centrifugal pump impeller faults.