Probabilistic Prediction of Location-Specific Microstructure in Turbine Disks

Numerical modeling tools can facilitate the process design, performance evaluation, and lifing prediction of a number of high-end components, including critical rotating jet engine parts. They can predict the component-wide state variables (such as stress, strain, strain rate, temperature) as a function of thermo-mechanical processing. These variables may then be coupled to microstructure evolution (such as grain size, precipitation). Subsequently the component, with its local variation in stresses and microstructure features, may be exposed to virtual tests (such as spin pit tests for jet engine turbine disks), and thus the performance of a component may be predicted as a function of the underlying locally specific microstructure-property relationships. Traditionally,"first order"structure-property relationships (such as strength) have be derived from average, scalar microstructure features (such as mean grain size, mean precipitate size), with encouraging success. A workpiece is initialized with an"as-received"microstructure (grain size, precipitate volume fraction), and thermo-mechanically processed. Classical microstructure models (such as Johnson-Mehl-Avrami-Kolmogorov) act on the average microstructure features and output an average microstructure result. However, higher order properties like fatigue crack initiation, fatigue life, etc require inputs that are more sophisticated than the"average"microstructure features. Whereas the average strength may be derived from the average microstructure features, lifing properties must be derived from the"worst actor"microstructure features those at the"long end of the tail"when graphed as a histogram. In order to predict these"worst actor"microstructure features, more sophisticated microstructure models, and a more robust state variable infrastructure than one which simply stores"average"values must be employed. Thus, during Phase I of this program, a proof-of-concept probabilistic model was developed to demonstrate this capability. Rather than providing scalar state variables of predicted strain, strain rate, temperature, and residual stresses, distributions of state variables, produced as a result of normal variation and uncertainties in the material and the processing conditions, were computed. These distributions then acted on distributions of microstructure features (e.g. grain size), in a probabilistic manner, thereby providing a numerical modeling tool that can compute the location-specific, probabilistic microstructure features and phenomena necessary as inputs to accurately compute higher order properties such as component life. During Phase I, Scientific Forming Technologies Corporation (SFTC) teamed with Carnegie Mellon University (CMU) to link sophisticated microstructure evolution research tools, developed at CMU, with existing FEM and microstructure modeling tools in the DEFORM code. Jet engine OEM GE Aviation provided industrial support. The proof-of-concept probabilistic modeling framework demonstrated in Phase I allows systematic analysis of the variabilities and uncertainties associated with the processing conditions, boundary conditions, material properties and incoming starting grain size distribution of the billet material. Thus, a probabilistic, location-specific microstructure response and residual stress distribution may be derived as a function of thermo-mechanical processing, and used as an input to a probabilistic lifing model. For Phase II, the team is expanding to include industrial supply chain partner Ladish, and research partner UES, to further improve the verification and validation of the thermo-mechanical processing, material modeling and property prediction methodology. Continued enhancement of the microstructure models and probabilistic infrastructure, integration into the commercial code in a user-friendly GUI, verification and validation of the model outputs, and more, are planned.