Application Research On Bearing Fault Diagnosis Based On Improved XGBoost In Big Data Environment

Rolling bearings are very common and widely used important parts in various largescale machinery and equipment.They play a very important role in various types of actual production.However,due to the characteristics of being vulnerable to damage,bearing failure will affect the stability of the overall unit.This led to a series of later problems.In order to prevent the safety problems of the whole machine caused by bearing failures,it is required to carry out maintenance work on rolling bearings to prevent hidden dangers.Therefore,it is of great practical significance to determine the type of bearing failure in advance and ensure its normal operation.Facing the era of big data,this paper takes bearing faults as the main research direction.Firstly,it explains the research background and significance of fault diagnosis in the big data environment;then summarizes and analyzes the related applications of big data technology and the theoretical process of bearing fault diagnosis.At the same time,this paper takes the three basic processes of bearing fault diagnosis as the starting point of research.The main research contents are as follows:（1）Aiming at the problem of the noise in the bearing vibration signal obtained by the sensor affects the fault diagnosis result,the LMD（Local Mean Decomposition）method is used to eliminate the noise,according to the characteristics of FPA（Fixed Point Algorithm）.A method of joint signal noise reduction is proposed.And aiming at the problem of the sensitive features in the original vibration signal are difficult to extract,and the constructed feature set is difficult to fully express the type of bearing fault,which affects the accuracy of later fault diagnosis.A dimensionality reduction method based on KPCA（Kernel Principal Component Analysis）is proposed.（2）Facing massive fault data,aiming at the problem of low diagnostic accuracy of traditional fault diagnosis model based on improved XGBoost（eXtreme Gradient Boosting Decision Tree）is constructed.The model takes the characteristics of multiple fault types as input,and improves the classification accuracy through XGBoost based on the parameter-optimized SVM.Among them,the improvement of XGBoost and the selection of two feature set processes are used to achieve the improved effect.The data processing process in the paper is applied on the Spark-Big Data platform,and scientific calculations are performed in parallel processing.In order to further illustrate the feasibility and advantages of the fault diagnosis model,the rolling bearing data provided by CWRU（Case Western Reserve University）is used for training and simulation.Finally,the diagnosis of this paper and the superiority of this model is proved by the fault bearing provided by the company and the comparison between 4 different fault diagnosis models.