A Hybrid Diagnosis Model With Experimental Verification for Composite Faults in Deep-Sea Mud Lifting Pumps

Abstract In deep-sea oil and gas exploitation, the reliability of mud lifting pumps directly affects the operational safety. However, the coupled operation of multiple pumps may easily induce composite faults such as abnormal vibration and excessive torque, which the conventional diagnostic methods struggle to address effectively. This article proposes an improved support vector machine (SVM) diagnostic method based on the collaborative optimization of linear discriminant analysis (LDA) and adaptive differential evolution (ADE). It uses LDA to reduce the dimensionality of the dataset and extract fault-sensitive features, and dynamically optimizes the kernel function parameters of SVM through the ADE algorithm, thereby constructing the LDA–ADE–SVM hybrid diagnostic model. To verify its effectiveness, a ten-stage centrifugal pump fluid–structure coupling model was built. Eighty-one groups of orthogonal experiments were designed to obtain 81 groups of simulation data. Compared with SVM, LDA–SVM, and LDA–PSO–SVM, the LDA–ADE–SVM achieved an accuracy of 96.30% with a calculation time of 13.93 s, outperforming the other algorithms. A dual-pump test platform was built to simulate dynamic working conditions. Fifty-four groups of orthogonal experiments were completed to collect 9294 samples, which were divided into groups at a ratio of 7:3 for comparison. The LDA–ADE–SVM showed significant advantages with an accuracy of 98.27% and a calculation time of 1028.16 s. This verifies its adaptability and robustness, providing support for offshore drilling and oil-gas development safety.