Study On Residual Life Prediction Of Aeroengine Compressor Blades Based On Residual Stress

Aero-engine rotor blade is one of the key components of the aero-engine,which has an important impact on the reliability and safety of the aero-engine.In order to improve the stability and service life of the blade,the residual compressive stress field is usually applied on the blade surface in engineering.In this paper,a certain compressor rotor blade is taken as the research object.Aiming at the residual life prediction of the aero-engine blades,the current status and shortcomings of current methods for predicting the residual life of aeroengine blades are analyzed,and a method for predicting the residual life of aero-engine blades based on residual stress is proposed.Before the macro or micro cracks can be detected in the blade,the residual stress on the blade surface can be obtained by detecting the change of the micro-lattice structure of the blade material.The residual stress is used as the characteristic parameter to construct the prediction model of the residual life,and the residual life of the blade can be predicted.The main work includes:Firstly,the characteristic parameters of the blade residual life prediction model are determined.In this paper,the residual stress at the specific detection points on the blade surface is taken as the characteristic parameters of the blade life prediction model.The selection of the residual stress detection points will directly affect the prediction effect of the model.When determining the detection points of residual stress,127 cases of blade failure were analyzed by statistical analysis method.At the same time,the 3D model of the blade was built by Solid Works,static analysis and modal analysis of the blade were carried out by finite element analysis technology,the load distribution of the blade was determined.Comparing with the simulation results,it is found that there is a good consistency between the concentrated area of blade load distribution and the common damage parts of blade.On this basis,seven residual stress detection points are determined on the blade surface,and the residual stress values of the detection points are used as the characteristic parameters of the blade life prediction model.Then,the residual stress at the detection point is tracked and measured by X-ray diffraction method,and the database of characteristic parameters is established.The method of residual stress measurement by X-ray diffraction was studied.The residual stress values of 144 blades working for 0 h,300 h,600 h and 900 h were measured by X-ray diffraction.A total of 3444 detection points were measured and 492 sets of characteristic parameter vectors were obtained.The results show that there are some numerical residual compressive stresses uniformly distributed on the surface of the new blade;the residual stresses on the blade surface will gradually decrease with the blade working;the residual stresses on the vulnerable and concentrated parts of the blade will decay more rapidly;and the residual stresses on the damaged parts of the blade with micro-damage will decay significantly.Finally,two models for predicting the residual life of blades are constructed.Based on the analysis of the target model of the residual life algorithm and the characteristics of training samples,two machine learning algorithms,BP neural network and support vector machine,are used to construct the residual life prediction model of blades.When building BP neural network model,according to the requirement of generalization performance of prediction algorithm,the Bayesian regularized BP neural network residual life prediction model is constructed,the optimal network size and key network parameters are determined,and the prediction performance of the model is analyzed.When constructing the residual life prediction model of support vector regression machine,two algorithms models of e-SVR and v-SVR are constructed by the LIBSVM software package respectively.The free parameters of the two models are optimized,and the prediction performance of the two models under the optimal parameters is compared and analyzed.The v-SVR method with superior performance is selected to construct the residual life prediction model of support vector regression machine.The two algorithms are simulated by MATLAB,and the two models are compared and analyzed.The results show that the two models can accurately predict the residual life of blades.The performance of support vector machine model is better when the number of samples is small,and Bayesian regularized neural network model is better when the number of samples is large enough.