Application of Bayesian neural network modeling to characterize the interrelationship between microstructure and mechanical property in alpha+beta-titanium alloys

Titanium alloys, especially alpha+beta titanium alloys are used extensively in the aerospace industry because of their attractive balance of properties. The mechanical properties of these materials are very much sensitive to their microstructure. Microstructure in these alloys can be controlled essentially through alloy composition and various thermomechanical processing routes. Microstructures in these alloys are characterized in terms of size, distribution and volume fraction of both alpha (HCP crystal structure) and beta (BCC crystal structure) phases. The above-mentioned features can coexist and span different length scales. The interrelationships between the microstructure and mechanical properties are characterized qualitatively in the literature. Physics based models are difficult to implement due to the presence of a wide variety of microstructural features with different length scales and mutual interaction of these features. The modeling of such properties is much more complex when composition is added as an additional degree of freedom.;In this work neural network models with a Bayesian framework have been employed to characterize the microstructure and mechanical property interrelationships in alpha+beta Ti alloys based on Ti-xAl-yV (4.76 alpha+beta Ti alloys based on Ti-xAl-yV (4.76<x<6.55; 3.30<y<4.45) with controlled variations in interstitial oxygen (O) and Fe (0.07 wt% O< 0.20; 0.11<wt% Fe<0.41). These alloys are subjected to various heat treatments and thermomechanical processing conditions such as beta annealing and alpha+beta processing to obtain a range of microstructure and mechanical properties. The important microstructural features in alpha+beta processed alpha+beta titanium alloys are equiaxed alpha grain size, volume fraction of equiaxed alpha grains, width of the alpha lamellae in transformed beta matrix and important features in beta heat treated alpha+beta titanium alloys are size of alpha colony, width of the alpha lamellae, prior beta grain size, volume fraction of colony and grain boundary alpha thickness. A database is populated with the above-mentioned quantified microstructural information, composition and mechanical properties. The mechanical properties predicted in this study are tensile properties and fracture toughness.;Based on the controlled virtual experiments conducted using neural networks on alpha+beta processed alloys suggested important microstructural features that will affect tensile properties are size of the equiaxed alpha grain and volume fraction of equiaxed alpha. The controlled virtual experiments on beta heat-treated alloys suggested important microstructural features such as width of the alpha lamellae, alpha colony size and prior beta grain size have negative influence on tensile properties.;The virtual experiments conducted on alloys which are processed in the alpha+beta phase field suggested that the size of the equiaxed alpha is an important variable which increases the fracture toughness. In beta-processed alloys, important microstructural features such as size of the alpha colony decrease the fracture toughness while width of the alpha lamellae and prior beta grain size increase the fracture toughness.;The alloying elements such Al, O and Fe improve the yield strength of both alpha+beta processed and beta processed alpha+beta titanium alloys. The O and Al have negative influence on fracture toughness while Fe has positive influence on fracture toughness.;The examination of the region beneath the fracture surface of alpha+beta processed alloy suggested occurrence of the microcracks within the equiaxed alpha particle clusters. The frequency of occurrence of the microcracks is increased when two neighboring equiaxed alpha grains have common or near common basal plane. The detailed dislocation analysis on regions near the microcrack indicated presence of extensive basal slip.