软计算技术在航空钛合金硬度预测中的应用研究

Research on the Application of Soft Computing Technology in Hardness Prediction of Aviation Titanium Alloys

摘要: 钛合金在航空航天材料中一直占据重要地位。而硬度作为钛合金的重要性质之一，对硬度的研究具有一定的实际价值。本文为研究航空钛合金棒件在受电磁脉冲工艺强化时硬度增加值随工艺参数的变化而变化的规律，建立基于BP神经网络的回归预测模型。并为克服BP神经网络容易因陷入局部最优解而意外终止的情况，提出利用具有全局最优性的遗传算法和粒子群算法对BP神经网络连接神经元节点的权值和阈值进行优化。结果表明，混合算法分别将MAE值从0.75769降低到0.69921、0.68881；将RMSE值从0.89908降低到0.84161、0.83932；将R值从0.99486提升到0.99632、0.99874。平均误差降低率分别为7.06%、7.86%，R值提升率分别为0.15%、0.39%。验证了混合算法模型的可行性，虽不能完全预测，但对实验进行仍具有一定的指导作用。

Abstract: Titanium alloys have always occupied an important position in aerospace materials. Hardness, as one of the important properties of titanium alloys, the research on hardness has certain practical value. This paper studies the law that the increase in hardness of aviation titanium alloy bars changes with the process parameters when they are strengthened by electromagnetic pulse technology, and establishes a regression prediction model based on BP neural network. And in order to overcome the situation that the BP neural network is prone to unexpectedly terminate due to getting trapped in the local optimal solution, it is proposed to optimize the weights and thresholds of the neuron nodes connected to the BP neural network by using the genetic algorithm with global optimality and the particle swarm optimization algorithm. The results show that the hybrid algorithm reduces the MAE value from 0.75769 to 0.69921 and 0.68881 respectively; Reduce the RMSE value from 0.89908 to 0.84161 and 0.83932; Increase the R value from 0.99486 to 0.99632 and 0.99874. The average error reduction rates were 7.06% and 7.86% respectively, and the R value increase rates were 0.15% and 0.39% respectively. The feasibility of the hybrid algorithm model has been verified. Although it can not predict completely, it still has a certain guiding role for the experiment.