基于Arrhenius和神经网络模型对镍基合金热变形行为的研究

Research on hot deformation behavior of a novel P/M superalloy based on Arrhenius and artificial neural network model

为研究新型高温合金的热变形行为,在应变速率为0.001~1 s^(-1),变形温度为1 080~1 140℃的条件下进行热压缩,得到了真应力-应变曲线并且建立了Arrhenius本构方程。但基于典型Arrhenius本构方程的实验数据与预测数据存在较大误差,Arrhenius本构方程的平均相对误差绝对值(AARE)为23.36%。然后采用应变补偿法对Arrhenius模型进行修正,AARE降至9.92%。此外,较低应变速率下的预测数据比较高应变速率下的预测数据更加精确。为了进一步更精确的预测,采用反向传播人工神经网络(BP ANN)模型对应力-应变曲线进行预测。结果表明,在典型的Arrhenius和应变补偿型模型中,ANN模型具有最高的精度和效率,AARE缩小至1.75%。

Nickel-based superalloy strengthened by'phase(L12 structure)has excellent high temperature strength,great plasticity and outstanding damage tolerance.These properties make it irreplaceable in many areas,such as aerospace industry and nuclear power industry.Meanwhile,a series of problems such as segregation,inho-mogeneity and poor hot working performance coming from traditional casting/forging superalloy have been solved by powder metallurgy process.Nowadays,powder superalloy has become the first choice of key hot end compo-nents such as turbine disc of advanced aero-engine and powder superalloy has become an important symbol of en-gine advancement.To obtain desired mechanical properties,superalloy is usually fabricated by extrusion,hot forg-ing and other processes.Therefore,it is of great importance to study the hot deformation behavior of superalloy.To study the hot deformation behavior of superalloy,constitutive model is applied to describe the relationship be-tween flow stress and deformation parameters.Among many constitutive models,Arrhenius-type constitutive mod-el is widely and commonly used in describing deformation behavior of nickel-based superalloy.Through the regres-sion analysis,the relationship between flow stress and strain rate,deformation temperature is established and deformation parameters can be optimized to acquire superalloy with homogeneous microstructure and excellent mechanical properties.However,because of the complex none-linear characteristics of deformation parameters on flow stress,it is difficult to predict the flow behavior precisely.Thus,tools to describe this relationship more accurately are urgently needed.Deep learning tools,such as artificial neural network(ANN),is the promising way to solve this problem.ANN has the ability of self-learning,self-adaptation,strong nonlinear function approximation and fault tolerance.Meanwhile it does not rely on mathematical models and deformation mechanisms.Through the adjustment of the internal connections between a large number of nodes,ANN can achieve the purpose of information processing and predict more accurate than other constitutive models.The effect of deformation temperature and strain rate on hot deformation behavior of the PM superalloy was investigated.The change in flow stress during hot deformation actually is the competition between work hardening(dislocation accumulation,dislocation interaction,etc.)and softening mechanism(DRV,nucleation,grain growth,etc.).The AARE and R of typical Arrhenius-type model is as large as 23.36%and 0.9658.After the modification in strain,the AARE and R drops to 9.92%and 0.9887,respectively.While BP-ANN model is adopted,the value of AARE and R slumps to 1.75%and 0.9995.BP-ANN performs better in dealing with the complex relationship between deformation parameter and flow stress.