Performance of reduced and real-time models for simulations of material processes.

The performance of real-time data models and their associated machine-learned trained databases are dependent on the complexity of fluid-thermal-mechanical interactions for material process applications. Additionally, sophisticated multi-dimensional search spaces for various chemical, thermal and mechanical material parameters are promoting critical challenges for the quick predictions and suitable decision making. The purpose of this research work is to investigate the performance of real-time models for thermal-mechanical material processes in terms of accuracy and generality. The objectives are to study the issues of optimized data structures and best algorithms to determine the best data training and query optimization techniques including neural network (NN) and genetic algorithm symbolic regression (GASR). The formation of effective snapshot matrices and their size\quality for real-time models being created are also considered for the case of best stored information. These matrices are designed to capture the most important aspects of the processing parameters including rate of changes, initial conditions and equipment's limits. Furthermore, the accuracy and reliability of these real-time models for additive manufacturing and extrusion processes are investigated using real-world case studies. [ABSTRACT FROM AUTHOR]