Ensemble Learning Based Decision-Making Models On The Aero-Engine High-Pressure Rotor Vibration Failure

The aero-engine is an extremely complicated and precise mechanical system.Whether any component works normally is related to the stability and safety of the entire system.The failure of a component will cause a huge disaster.Moreover,the research and development cycle is long and costly.It is a crystallization of modern technology.Its development process involves the cutting-edge of various fields such as technology,materials science,and computer science.As the source of power for an aircraft,aero-engines also directly determine various aircraft performance and levels.In the actual operation of the aircraft,due to the changing operating conditions,there are many factors that cause structural vibration.There is a close connection between various components inside the engine.If some of the components are abnormal or fail,they will have a cascading effect on the entire aircraft system.This paper analyzes the vibration faults of aero-engines caused by the imbalance of high-pressure rotors.During the research,it is found that the main difficulty in constructing a fault diagnosis model of aero-engines is the processing of unbalanced data and the extraction of dimensionality reduction of multidimensional feature data.At the same time,through communication with the institute,two methods of single structural feature parameter analysis and combined structural feature parameter analysis are used for fault diagnosis.Among them,the single structural feature parameter analysis adopts softmax classifier method for analysis and research,and the combined structural feature parameter analysis is conducted by integrated learning.The failure diagnosis rate of the model constructed by research and analysis of single structural feature parameters is difficult to meet the requirements,so an integrated learning method is used for fault diagnosis.This method uses the borderline-smote method to balance the data,and proposes a weight-based Gini coefficient feature extraction method for feature extraction.Finally,the extracted features are used to construct an aero-engine fault diagnosis model.Through research and analysis of vibration fault data of aero-engine high-pressure rotor provided by a research institute.By using a weight-based Gini coefficient feature extraction method to extract features with a high degree of correlation with the fault label from the sample space,and at the same time,the model of ensemble learning is used to construct the model.Mainly studied the classification effects of three more advanced boosting ensemblealgorithms such as xgboost,gbdt and adaboost,discussed the advantages of different algorithms,used three algorithms and used decision trees as the base learners to construct three classifiers.And using the bench data of a research institute for testing,it is found that the fault diagnosis rate of the fault diagnosis model constructed by the xgboost algorithm can reach more than 95% and the AUC value is also significantly higher than the other two methods.And the failure accuracy of the model constructed with the combined structural feature parameters is also significantly higher than the model constructed with the single structural feature parameters.It shows that this method has a good application in the vibration fault diagnosis of aero-engine high-pressure rotor.