On An Integrated Experimental and Computational Approach to Derive Phenomenological Equations to Predict Properties in Titanium Alloys

The development of tools to predict the mechanical properties based upon compositional and microstructural inputs in multi-component, multi-phase Ti-based alloys represents a significant challenge. One such solution is the development of high-fidelity databases and the subsequent application of non-linear modeling tools such as neural networks based upon a Bayesian framework to extract the underlying composition-microsructure-property relationships. This approach has resulted in successful tools for the prediction of properties but often is based upon complex equations that do not appear to be phenomenological.  Thus, one must use new approaches in parallel with neural networks to derive the phenomenological equations.  This talk will highlight the development of such rules-based models for the prediction of the tensile and fracture toughness properties of Ti6Al4V at room temperature. These models have been successfully used to isolate the influence of the individual microstructural features on the mechanical properties. This talk will then provide the framework, including integrated genetic algorithms and Monte Carlo approaches, to determine the first generation of the more robust phenomenological models.