基于BP人工神经网络的TC17钛合金显微组织-力学性能关系预测

Prediction of Correlation between Microstructure and Tensile Properties in Titanium Alloys Based on BP Artificial Neural Network

摘要:

钛合金的性能对其组织状态十分敏感，与组织的多种显微特征呈现非线性的交互关系。本研究在定量分析钛合金显微组织的基础上，采用BP人工神经网络方法建立了TC17钛合金组织与力学性能的关系模型。该模型输入的显微组织特征参数包括：α相体积分数、α相厚度和不同形态α相的体积分数，输出的力学性能包括抗拉强度、屈服强度、延伸率和断面收缩率。结果表明，该模型具有很好的预测精度和泛化能力。应用贝叶斯正则化和动量梯度下降学习法较好地解决了传统BP人工神经网络训练高精度和预测低精度的过拟合现象。此模型的建立对构建TC17合金利用组织预报力学性能的专家知识库具有重要作用，而且对钛合金专家系统的整体开发具有重要指导意义。

Abstract:

Titanium alloys’ properties are sensitive to the microstructure very much, which have nonlinear interactive relationship with the microstructral characteristics. In this study, a model was developed for the prediction of the correlation between microstructure and tensile properties in titanium alloys using artificial neural network (ANN). The inputs of the neural network were quantificational microstructure parameters, including thickness of α-laths, volume fraction of α-laths and Ferret Ratio. The outputs of the model were the tensile properties, including ultimate strength, yield strength, elongation and reduction of area. The model was based on back-error propagation (BP) neural network, and trained with the data collected from isothermal compression experiments of Ti17 alloys. A very good performance of the neural network was achieved such as prediction accuracy and generalization ability. Bayesian regularization and gradient descent learning method can solve the super-fitting problem of high-accuracy training and low-accuracy prediction of traditional BP artificial neural network. The model can be used for prediction of tensile properties of Ti17 alloys according to its microstructural features. Modeling this correlation is fairly necessary to build a robust expert database in titanium expert system.