Aging heat treatments in high temperature NiTiHf shape memory alloys (SMAs) create nano-sized precipitates and modify their phase transformation behavior. Depending upon specific ap-plication requirements, the phase transformation behavior in the SMA may be tailored through performing appropriate aging heat treatments. However, identifying the correct heat-treatment for the desired modification in the phase transformation behavior can be time-consuming and may require costly experiments. Can micromechanical modeling speed up this process by predicting the microstructure to behavior linkage? What factors need to be considered to improve accuracy? What tools can improve the computation cost? The modified phase transformation behavior with heat treatment is a result of the new microstructure. Micromechanics connects the microstructure of a material to its behavior and is useful in the context of NiTiHf SMAs for identifying desired microstructures, and thereby the heat treatments. We developed a full-field finite element based micromechanical modeling framework for modeling the microstructure-behavior linkage in precipitation hardened NiTiHf SMAs. Representative volume elements (RVEs) from 3D transmission electron microscopy-based reconstructions of the material microstructure were used. With the new modeling framework, the NiTiHf SMA behaviors of different compositions and processing were analyzed. Consequently, compositional and processing effects affecting phase transformation in NiTiHf SMAs were identified. The full-field micromechanical modeling can be computationally expensive, and the size of RVEs modeled can affect the prediction. Additional investigations were performed to develop faster tools of computation and to determine size RVEs. Fast Fourier Transform (FFT) based solution methodology, in the SMA micromechanical model, was found to improve computation time. Using the FFT, larger RVEs and multiple realizations were analyzed in the SMA micromechanical model and the RVE size statistics were studied. Consequently, for capturing RVE size statistics, we developed a new general formulation using principles of perturbation theory. The dispersion of individual RVE behaviors was formulated as perturbation from the ensemble average behavior, and hence the ensemble statistics can be derived in terms of perturbations. The new RVE size methodology was demonstrated for SMA micromechanical model, and found to be capturing the RVE statistics in a wide range of precipitate volume fractions. Further, we investigated the potential of data-based machine learning tools for faster predictions in new RVEs using full-field simulation data. The machine learning models were explored at three levels of complexity: (a) with precipitates, (b) without precipitates and (c) with anisotropy without precipitation. For (a) with precipitates, the RVEs were represented using 2-point statistics and reduced using principal component analysis and taken as input to a machine learning model to target as the effective response. The model predicted responses in new volume fraction RVEs and were compared with full-field simulation. In (b) without precipitates, we investigated the ability of a machine learning model to predict complex partial transformation responses. The machine learning model was built using experimental major actuation responses of a NiTiHf SMA, and its ability to interpolate the behavior to predict minor cycle responses was investigated. In (c) with material property anisotropy, the ability of a machine learning model to capture the anisotropy in single crystal SMA responses was investigated. Simulations from a crystal plasticity model were used for training, and the ability of the machine learning model to predict responses in new orientations was investigated. A micromechanical modeling framework was successfully demonstrated for predicting the behavior of precipitation hardened NiTiHf SMAs. The computational cost of the modeling frame
