基于改进BP神经网络的TA15钛合金近β锻造组织预测模型

Model for predicting microstructure of TA15 titanium alloy in near β forging based on improved BP neural network

文章以训练结果的误差均方差与误差和降低为目标,通过循环和判断语句改进了MATLAB人工神经网络(ANN)工具箱的BP算法,实现BP网络多结构、多次循环训练,建立了TA15钛合金近β锻造变形参数(变形温度、应变速率和变形量)和变形水冷(WQ)及后续热处理(再结晶退火或高低温强韧化处理)后的组织特征参数(等轴α相的含量、平均晶粒直径和轴比,条状α相的含量和厚度)之间关系的BP人工神经网络模型。结果表明,针对近β锻造组织预报输入参数多,输入-输出参数高度非线性,该模型可以有效避免传统BP模型容易陷入局部极小值点的缺点,可较准确的得到各工艺参数组合下的组织特征参数;模型预测结果可以用于近β锻造不同工艺参数组合下组织特征参数的预报,及其演化规律的分析。

Taking the reduction of the mean square error and the error sum of the training results as the goal,adopting loop and judgment statements,the BP algorithm of the MATLAB ANN toolbox was improved,the multi-structure and repeated training of BP network were achieved,the model based on BP ANN for describing the relationship between deformation parameters（deformation temperature,strain rate and total deformation） and characteristic parameters of the microstructure（the average grain size,the content of equiaxed α,the content and the thickness of striature α） after deforming water cooling（WQ） and subsequent heat treatments（recrystallization annealing or a high and low temperature toughening and strengthening treatment） in near β forging process of TA15 titanium was developed.The results show that：in allusion to excessive input parameters and high nonlinear characteristics of the input-output parameters in near β forging,this model can effectively avoid being prone to fall into local minimum point as in traditional BP model,and can obtain more accurate characteristic parameters of microstructure under different process conditions.The results can be used to predict the microstructure in near β forging with different process parameters,and play a guide to investigate the evolution mechanism of the microstructure in near β forging.