Abnormal identification of lubricating oil parameters and evaluation of physical and chemical properties based on machine learning

Abstract

This paper firstly proposes an abnormality identification model for online oil monitoring parameters. The SVR model was established to predict the output power, and the information entropy of the deviation between the predicted value of the regression model and the actual monitoring value was used to identify the abnormal parameters of the oil. The physical and chemical properties of the oil was evaluated and predicted comprehensively. Principal component analysis (PCA) was used to eliminate the correlation between parameters, and then the deviation degree of each principal component is calculated after feature transformation. After adding the deviation degrees of each principal component according to their weights, the performance evaluation index of the oil is obtained. At the same time, a prediction model of oil performance trend based on Gated Recurrent Unit is proposed. The feasibility and applicability of this model are verified by comparing with the results of LSTM neural network and ARIMA model. The evaluation of the physical and chemical properties of the oil can detect the deterioration of the oil condition at the early stage, avoiding or reducing accidents, so as to ensure the safe and reliable operation of the unit.