Research On Prediction Model Of Friction Performance Of Water-Lubricated Bearings Based On Random Forest Algorithm

As an important part of marine propulsion system,the tribological performance of marine bearings has a direct impact on the navigation safety of ships.With the increasingly strict requirements of environmental protection,water-lubricated stern tube bearings are more and more widely used.The tribological performance of stern tube bearings directly determines the reliability of water-lubricated bearings and marine propulsion systems.Polyurethane,as the traditional matrix material of waterlubricated bearings,needs to be improved in the process of ship operation.In order to solve these problems,it is found that the wear resistance of polyurethane can be improved by filling it with polyethylene wax material.A large amount of design is needed in the process of mixing test,which increases the time cost.Therefore,this dissertation proposes a prediction model based on Random Forest classification algorithm,in order to reduce the test cost and optimize the design of new composite materials with wear resistance.This dissertation selects the best one from two kinds of algorithms(classification algorithm and regression algorithm),and selects the representative subdivision algorithm for comparative analysis.From the calculation results,it is found that the random forest classification algorithm has the advantages of both,and has better robustness to the data fitting performance of small data volume.Therefore,this dissertation builds a prediction model based on the random forest classification algorithm.The model constructed in this study is fitted by algorithm and data.The algorithm compares and selects Random Forest classification algorithm.The test object of data is water-lubricated bearing made of polyurethane matrix material and polyethylene wax,and five groups of friction data tested under different working conditions are adopted.Firstly,the data are analyzed and screened,and the running-in data of each group of working conditions were eliminated.The data of normal wear stage are merged into a group of data sets,and the random forest classification algorithm is used to build the data model.The grid search method is used to optimize the parameters of the algorithm,root Mean Square Error(MSE)and ROC curve method are used to verify the reliability of the model,and the model is further optimized by the five-fold cross-validation method.The final model shows that the mean value of ROC curve is 0.85,and the MSE value is 0.18,that is,the prediction accuracy of the training set is 0.85,and the error of the test set is 0.18,and the training effect of the model is good.Through continuous learning to further improve the above model,that is,to optimize the model through supplementary training data,this dissertation designs a database through My SQL and Navicat,and designs a calculation software that encapsulates the algorithm.In order to verify the effect of the model predictions,this article also designs five groups of different working condition of the new test,taking the average of the working condition of each group,will receive value compared with the model prediction,find that the results of error rate is below 6%,the main cause of the error is due to the pressure fluctuation in the process of water lubricated bearing in the test.The prediction model established in this dissertation is reasonable and can predict the tribological properties of the tested materials.