Prediction of particulate content in oil based on SPA vibration feature selection

Abstract

Aiming at the non-stationary characteristics of oil pressure vibration signals containing particulate, a method for predicting particulate content in oil was proposed based on vibration characteristic frequency extraction by vibrational mode decomposition (VMD), variable selection using successive projections algorithm (SPA) and T_S fuzzy identification combined. Firstly, the pressure vibration signal was decomposed by VMD and a series of narrow-band characteristic frequency matrices were obtained. Then, variables were selected using SPA to construct the feature vector matrix. Finally, the feature vector matrix was used as the input of T_S fuzzy identification to identify the content of particulate in oil. The results showed that the VMD reconstruction of the original oil sample pressure signals could well characterize the main variation of the original signal; the 19 variables were selected from the characteristic frequency of the vibration signal from the oil pressure using SPA, the 19 pressure vibration characteristic frequency of 11 sample sets SPA selected was taken as the input variable of T_S identification model; for each set of sample, the predicted output of the content of particulate in oil was obtained, model prediction decision coefficient is 0.8637, the root mean square error is 0.1979, a reasonable prediction effect was obtained.