Application of genetic algorithm to enhance the predictive stability of BP-ANN constitutive model for GH4169 superalloy

• Published in: Journal of Central South University Vol. 31; no. 3; pp. 693 - 708
• Main Authors: Zheng, De-yu, Xia, Yu-feng, Teng, Hai-hao, Yu, Ying-yan
• Format: Journal Article
• Language: English
• Published: Changsha Central South University 01.03.2024; Springer Nature B.V
• Subjects: Artificial neural networks; Back propagation networks; Compression tests; Constitutive models; Engineering; Genetic algorithms; Mathematical models; Metallic Materials; Nickel base alloys; Parameters; Plastic flow; Stability; Superalloys; Thermal simulation; Thermal simulators; True stress; GH4169 superalloy; stress–strain; genetic algorithm; artificial neural network; 遗传算法; 人工神经网络; backpropagation; 应力-应变; GH4169高温合金; 反向传播
• ISSN: 2095-2899; 2227-5223
• DOI: 10.1007/s11771-024-5591-x

In order to better characterize the plastic flow behavior of GH4169 superalloy, isothermal compression tests of GH4169 superalloy at different temperatures and strain rates were carried out using Gleeble 1500 thermal simulator. The back propagation artificial neural network (BP-ANN) constitutive model of GH4169 superalloy was established based on true stress–strain data, and the relationship between the prediction stability of the constitutive model and the model parameters was further investigated. The prediction results show that the BP-ANN model outputs were highly influenced by the model parameters. To address this issue, genetic algorithm (GA) was used to optimize the BP-ANN constitutive model, and the GA-BP-ANN integrated constitutive model was presented. The optimization results show that the GA-BP-ANN integrated constitutive model greatly enhances the prediction stability and improves the generalization ability of GH4169 superalloy’s BP-ANN constitutive model.