论文部分内容阅读
采用了热模拟实验机研究了Al-Cu-Mg-Ag耐热铝合金的热压缩变形行为。实验的温度和应变速率分别为340～500℃,0.001～10 s-1。分别用了本构方程和人工神经网络来对Al-Cu-Mg-Ag合金的流变行为进行了分析和模拟。神经网络的结构是3-20-1;输入参数是温度,应变速率和应变;输出参数是流变应力。结果表明该合金的流变曲线出现加工硬化、过渡、软化和稳态流变这4个阶段,流变应力随着应变速率的增加而增大,随着变形温度的下降而减少。用所建立的神经网络模型预测了变形温度和应变速率对流变应力的影响,预测的结果与热压缩变形的基础理论吻合得很好,而且该模型可以很好地描述Al-Cu-Mg-Ag合金的流变应力,在应变速率为0.001～10 s-1的条件下,其平均相对误差分别为3.68%,3.98%,1.53%,3.53%和2.04%。这表明神经网络的预测性能优良,具有很强的推广能力。同时通过本构方程和神经网络的预测结果比较看出神经网络模型的相关系数比较高,而且神经网络比本构方程有更好的预测性能。神经网络可以预测不同应变下的相应的流变应力,但是本构方程只可以根据不同的应变速率和温度来预测峰值应力。
The hot compression deformation behavior of Al-Cu-Mg-Ag heat-resistant aluminum alloy was investigated by using a thermal simulator. Experimental temperature and strain rate were 340 ~ 500 ℃, 0.001 ~ 10 s-1. The rheological behavior of Al-Cu-Mg-Ag alloy was analyzed and simulated by constitutive equation and artificial neural network respectively. The structure of the neural network is 3-20-1; the input parameters are temperature, strain rate and strain; the output parameters are flow stress. The results show that the rheological curve of the alloy appears in four stages of work-hardening, transition, softening and steady-state rheology. The flow stress increases with the increase of strain rate and decreases with the decrease of deformation temperature. The effect of deformation temperature and strain rate on the flow stress was predicted by the proposed neural network model. The predicted results are in good agreement with the basic theory of hot compression deformation, and the model can describe Al-Cu-Mg-Ag The average relative errors of the flow stress of the alloy are 3.68%, 3.98%, 1.53%, 3.53% and 2.04% at the strain rate of 0.001-10 s-1 respectively. This indicates that the neural network has good predictive performance and strong promotion ability. At the same time, the correlation coefficient between the neural network model and the neural network is higher than the predictive result of the constitutive equation and neural network, and the neural network has better prediction performance than the constitutive equation. Neural networks can predict the corresponding flow stress at different strains, but constitutive equations can only predict peak stresses based on different strain rates and temperatures.