Random Forest Based Fault Diagnosis Algorithm And Experimental Research

Rotating machinery is one of the most widely used large-scale power drive equipment,and it usually runs in harsh environment.Due to the complexity of its subsystems,the variability of the working states and the improper operations,the sensor signals obtained during the fault diagnosis of the rotating machine are mostly interfered by environment.Therefore,the accuracy improvement of rotating machinery fault diagnosis has been investigated by researchers.Random forest algorithm with the idea of combining weak classifiers is used in this thesis,to solves the problem of low accuracy in fault diagnosis using single classifier.However,the conventional random forest algorithm cannot guarantee the accuracy of fault diagnosis problems with a large number of unlabeled samples.Thus,the semi-supervised learning is integrated into the traditional random forest algorithm to improve the classification accuracy by increasing the number of labeled samples.The main contents are as follows:1.When a single classifier is used for rotating machinery fault diagnosis,there is a problem that the accuracy rate is low.In this dissertation,we present a fault diagnosis method for rotating machinery gears based on the random forest.First,the gear sensor signals under multiple modes and multiple fault conditions are collected,with time domain characteristics extracted as the input features of the random forest.Then,the constructed random forest model is used to diagnose the gears.Finally,the classification results are compared with the classification results of the support vector machine method.The fault diagnosis results show that the random forest algorithm has higher prediction accuracy than the support vector machine method.2.When the traditional random forest is used for fault diagnosis with only a small number of labeled samples available,there is a problem that the accuracy rate is low.An improved random forest algorithm is presented for fault diagnosis of rotating machinery.Firstly,a large number of unlabeled samples are divided into two parts,denoted respectively as unlabeled sample Ⅰ and unlabeled sample Ⅱ.For unlabeled sample Ⅰ,graph-based semi-supervised learning is used for label prediction.Then,the unlabeled sample I with a predictive label and labeled samples are used to train multiple decision trees.If the classification result is consistent with the one of label prediction,then the unlabeled sample Ⅰ is added to the labeled samples and used for building a random forest model.While,the data of unlabeled sample Ⅱ are utilized for testing of the obtained random forest model.Finally,the developed random forest algorithm is applied to an experimental platform of rotating machinery.It is shown from the simulation results that,for the case of samples being noise pollution and with unsatisfying labels,the new developed algorithm can improve the fault classification accuracy than the traditional random forest.