Constitutive relationship of (Ti5Si3 +TiBw)/TC11 composites based on BP neural network. (August 2022)

• Record Type: Journal Article
• Authors: Liang, Zhong; Yu, Fu; Yinyang, Wang; Yongdong, Xu
• Is Part Of: Materials today communications. Volume 32(2022)
• Publication Date: 2022-08
• Subjects: Discontinuously reinforced titanium matrix composites -- BP artificial neural network -- Arrhenius equation -- Constitutive model; Materials science -- Periodicals; 620.11
• DOI: 10.1016/j.mtcomm.2022.103973
• Languages: English
• ISSNs: 2352-4928

Abstract

Abstract: The Gleeble3500 thermal simulator was used to (2vol%Ti5 Si3 +5vol%TiBw)/TC11 composites with network reinforcement structure at a deformation temperature of 1183–1363 K and a strain rate of 0.01–10 s-1 to perform a deformation of 60% constant temperature compression experiment. The thermal deformation behavior of composites and the stress-strain curve of thermal compression deformation were studied through the results of constant temperature compression experiments. In addition, based on the Arrhenius equation and the BP neural network model, the constitutive equations of flow stress σ, strain rate ε ̇, and deformation temperature T were established. The results show that as the strain rate increases and the deformation temperature decreases, the flow stress of the composites increases, and the stress-strain curve shows three stages of work hardening, rheological softening and stable rheology. Based on the Arrhenius model, a constitutive equation was established to fit and predict the true flow stress. The average relative error, the correlation coefficient and, the root mean square error are 7.4%, 0.99526, 14.5, successively. The prediction has a rough generalization ability and has deviation errors. On the other hand, based on BP Neural network model, a constitutive equation was established to fit and predict the true flow stress. The average relative error, the correlation coefficient and, the root mean square error are 2.5%, 0.99675, 3.7 in sequence. The prediction accuracy is high, and the generalization ability is strong. Thus, It is suitable for the establishment of nonlinear constitutive equations of materials. Graphical Abstract: ga1