Machine Learning Based Gear Oil Temperature Anomaly Analysis And Lubrication Performance Evaluation

Gear box is the core mechanical component of double fed wind turbine,which plays the role of increasing speed and power transmission.Thirty percent of the failures are caused by lubrication failure.The analysis of wind turbine gear oil’s abnormal temperature is of great importance,as well as evaluating the performance of both new and in-process oil,enhancing lubrication,refining the quality,and extending the service life of wind turbine gear oil.This paper builds a gear oil temperature anomaly analysis and lubrication performance evaluation model based on machine learning.Based on the data sources of wind turbine SCADA system,gear oil infrared spectrum and physical and chemical properties experimental data,and friction and wear properties experimental data,machine learning is adopted to mine useful information from the above data.Establishing recognition and early warning models of gear oil’s abnormal temperature,analyzing physical and chemical properties of infrared spectrum gear oil,and evaluating tribological properties of gear oil additives,the accuracy and effectiveness of these models were then verified through comparative analysis.The main work contents and results are as follows:(1)Carry out abnormal gear oil temperature identification and early warning.The research object was the oil temperature of the gearbox in the wind turbine SCADA system.Firstly,LOF and DBSCAN were used to deal with the outliers in the data.Then,GRA analysis method was used to extract useful feature information from high dimensional information,and a TCN-Attention gearbox oil temperature prediction model was established.Comparing with other models and analyzing the recognition results of abnormal temperature of fan gear oil,it is verified that this model can detect the potential trend of abnormal temperature of gear oil and give early warning.(2)Combine infrared spectrum detection technology with machine learning to analyze the physical and chemical properties of gear oil.The PLS physical and chemical properties prediction model was established,and the BPSO-BSOA combined optimization method was designed to screen the characteristic wavelength of the feature data of the whole region of the infrared spectrum of the gear oil.Verification of the efficacy of the BPSO-BSOA-PLS amalgamated optimization model in the examination of the physical and chemical characteristics of the gear oil was finally accomplished.(3)Four kinds of extreme pressure antiwear additives commonly used in gear oil were selected to design experiments and test their tribological properties.GRNN and SVM prediction models of average friction coefficient and wear volume of gear oil were established.GWO method is used to optimize the smoothing parameter σ of GRNN model to improve the predictive ability of the model.Through model testing and LOOCV analysis,it is proved that the GWO-GRNN model has excellent performance,good prediction accuracy and generalization ability.