Reliability Analysis Of Blade Crack Propagation Life Based On Improved Response Surface Method

In the research of the fatigue life reliability of the turbine blade,Comprehensively considered failure criteria and edge load,according to the reliability theory,the intelligent algorithm,and neural network response surface method,we separately propose four Response Surface Methods in order to improve the calculation accuracy and calculation efficiency: the Mind Evolutionary Algorithm & BP Response Surface Method(MEA-BPRSM),the Restricted Boltzmann Machine & BP Double Response Surface Method(RBMBPDRSM),the Mind Evolutionary Algorithm &BP Extremum Response Surface Method,(MEA-BPERSM)and the Restricted Boltzmann Machine & BP Extremum Response Surface Method(RBM-BPERSM).Thus,this paper demonstrates the four Response surface Methods are valuable reference for the research on aero-engine turbine blade.(1)Reliability analysis of blade fatigue life based on response surface method and thinking evolutionary algorithm-BP neural network.Firstly,through the thermal-structure coupling numerical simulation,we obtain the distribution characteristics of the maximum stress and maximum strain of the blade.Then,taking the blade speed,temperature,material density and the fatigue-related parameters as input random variables,and using blade fatigue life as the output response,we get a small sample data set with the Latin Hypercube Sampling technology to establish the MEA-BPRSM mathematical model again and again;Finally,applying Monte Carlo method,we made a large number of simultaneous sampling of MEA-BPRSM and solved the reliability probability,and the result shows the comprehensive reliability of blade fatigue life is 99.73%.(2)Reliability analysis of blade crack propagation stress intensity factor based on restricted Boltzmann machine-BP neural network dual response surface method.First of all,taking the turbine blade material density,speed,elastic modulus,aerodynamic pressure,and gravity are analyzed as random input variables,and considering the maximum stress and maximum strain distribution characteristics of the blade based on the simulation on the blade with the thermal-structural coupling.The theoretical research confirms the location where the crack appears is the maximum stress,or the blade roots.Continuing we calculated the stress intensity factors at different stages of crack growth.Then,we obtained a small sample data set by using Latin hypercube sampling technology,and were trained to establish the RBM-BPDRSM Mathematical model;Finally,using the Monte Carlo method to sample the RBM-BPDRSM in large quantities and calculated the reliability probability.the result shows that the reliability of the mode I crack growth stress intensity factor for blade crack growth was 99.68%,and the mode II crack growth was 99.82%.(3)Reliability analysis of turbine blade crack propagation life based on improved neural network extreme response surface method.First,the same way to take the density,speed,elastic modulus,aerodynamic pressure,and gravity of the blade as input variables,and we found the stress and strain distribution characteristics of the crack growth of the turbine blade through deterministic analysis.The blade crack growth life is used as the output response to dynamically output Respond to all the minimum values in the analysis time domain as a new output response,therefore we got a small sample data set,and normalized the data;then,we use the small sample data set to train to establish the MEA-BPERSM and RBM-BPERSM mathematical models respectively;finally,we used the Monte Carlo method to sample a large number of MEABPERSM and RBM-BPERSM mathematical models,and obtained their reliability probabilities respectively is 99.81% and 99.69%.