Application of SaRT–SVM algorithm for leakage pattern recognition of hydraulic check valve

Check valves are key components in hydraulic systems. The cross-port leakage in check valves is a common fault that affects their performances. The vibration and pressure fluctuations excited by leaks are weak, therefore leakage is difficult to be identified and classified by intelligent algorithms and non-destructive testing methods. To maximize the performance of leak pattern recognition, we had improved the sequential minimal optimisation algorithm for enhancing the classification performance and tested it with the University of California Irvine Machine Learning Repository. Furthermore, combining with the search and rescue team (SaRT) algorithm, we propose SaRT-SVM algorithm. Two important parameters γ and C of support vector machine (SVM) were optimised and compared with response surface and other algorithms. We analysed the SaRT–SVM method for leakage pattern recognition and validated the robustness of the developed method by applying the method on multiple fault samples of each fault mode, additional different noises, and another independent data collection. The results showed that the SaRT–SVM algorithm exhibited excellent classification performance and robustness when applied to the leakage pattern recognition of hydraulic check valves under the influence of different noises.