Mitigation of sand mold related metal casting defects through virtual manufacturing

Metal casting is an important manufacturing process and most castings are made by small to mid-sized foundries. Sand molded castings account for 85% of all iron and steel castings produced and approximately 5 out of 30 major casting defects including gas blows and sand burn-on are related to poor sand mold atmosphere control. Mold atmosphere control should become more crucial in the coming years as 3D printed molds with finer features and higher binder levels become more widespread and as quality of foundry shop air comes under tighter regulation by the Environmental Protection Agency [3,4]. We propose to develop a relatively cheap, fast, robust, geometry specific computational modeler of sand mold atmosphere which should allow to significantly reduce sand molded castings defect rate, increase foundry production rates and reduce foundry energy consumption. This will be accomplished by insisting on mathematical models of binder outgassing and transport detailed enough to give physical fidelity, by utilizing more sophisticated numerical methods available with modern scientific software, to allow for accurate solutions of mathematical models on coarse space-time grids. This will be implemented on HPC resources and use an intuitive, problem-specific graphical user interface front-end that can be easily understood by the foundry process engineer. The distribution base of the mold dynamics casting defect prediction program would cover roughly 1000 US based iron and steel foundries and an additional number of non-ferrous alloy foundries. This tool would become part of the defect prediction software suite which already includes prediction of other major casting defects such as metal shrinkage porosity. Our proof-of-concept computations show that aggressive computational run time goals for the simplest useful mold atmosphere models are achievable. We are targeting one-to-two hour run times for typical low- to mid-range HPC cluster resources, that can be made available to a foundry cost-effectively. In Phase I of the project, we plan to demonstrate that these goals are realistic for a mid-sized iron casting sand mold. Phase II would focus on more sophisticated models which would rely on distributed computational techniques to maintain short target runtimes. The numerical sophistication needed for the project will come from the Lawrence Berkeley National Lab Chombo software library with its planned Phase I extensions and physical fidelity of the models and the utility of the user interface will be ensured by the continued interaction with the metal casting community via the American Foundry Society.