Modeling of Microstructure Property Relationships in Ti-6Al-4V

Fuzzy logic neural network models were developed to predict the room temperature tensile behavior of Ti-6Al-4V. This involved the development of a database relating microstructure to properties. This necessitated establishing heat treatment processes to develop microstructural features, mechanical testing of samples, creating rigorous stereology procedures, developing numerical models to predict mechanical behavior, and determining trends and inter-relationships relating microstructural features to mechanical properties. Microstructural features were developed using a Gleeble??? 1500 Thermal-mechanical simulator. The system used computer controlled resistive heating equipment to provide rapid cooling and heating abilities. Samples were obtained from mill annealed plate material and both alpha + beta forged and beta forged materials. A total of 72 samples were beta solutionized and heat treated using different heating and cooling conditions. Rigorous stereology procedures were developed to characterize the important microstructural features. The features included Widmanst??tten alpha lath thickness, volume fraction of total alpha, volume fraction of Widmanst??tten alpha, grain boundary alpha thickness, mean edge length, colony scale factor, and prior beta grain size factor. Chemical composition was also determined using standard chemical analysis and microscopy techniques. The samples were tested for yield strength, ultimate tensile strength, and elongation at room temperature. The samples were imaged using various microscopy techniques. Results from the tests and the characterization were used to develop fuzzy logic neural network models to predict the mechanical behaviors and develop relationships between the microstructural features (using CubiCalc RTC???). Results were compared to standard multi-variable regression models. The fuzzy logic neural network models were able to predict the yield, and ultimate tensile strength, within acceptable error ranges with a limited number of input data samples. The models also predicted the elongation values but with larger errors. The models also provided trends detailing the relative importance of the input parameters and the inter-relationships between the features. Of particular importance, the models identified the importance of the Widmanst??tten alpha lath widths, the mean edge length of the Widmanst??tten alpha laths, the colony scale factor, and the prior beta grain size to the tensile behavior.