Development of a Multi-Physics FEM model with Voxel based porosity algorithm for defect prediction in LPDC components

Abstract

The automobile wheel market is expected to grow by almost 5.52% from 5 years now. Majority of the market of the automobile wheels is occupied by Aluminium alloys because of numerous advantages, one of which includes light weight. The most preferred manufacturing process for the Aluminium alloy wheels is the Low Pressure Die Casting. However, the process is accompanied by various defects, some of which reflect significantly on the fatigue and impact properties are Shrinkage, gas porosity, oxide film and oxide inclusions. The severity of the defects not only depends upon the type or quantity, but also on the location of the defect. Some of the visible defects can be reworked, few of which that do not pass the quality conformance are re-melted. It would be much economical to design the mold, cavity, die cooling circuit and appropriately set the process parameters, which will reduce the defects and its severity. Various physics that are involved in LPDC process are Heat Transfer, Free-Surface and Turbulent flow, Phase transformation and Alloy composition. A Multi-physics simulation can incorporate all the mentioned physics required for LPDC. Various input set-point parameters that can contribute in Shrinkage porosity are Mold pre-heat temperatures, Pressure Cycle, Cooling channel flow-rates, delay and duration. The FEM-FVM simulation tools provide an FEM output like phase fractions, from which shrinkage porosity is calculated via post-processing techniques. On the other hand industrial softwares apply some simplified or approximate solutions to calculate porosity, which may not always be true. A better approach here would be to use FEM-FVM tool to calculate phase fractions, and then apply a post-processing algorithm based on the shrinkage physics to track the porosity quantity and location. Once a simulation data with porosity results are obtained, a Machine Learning algorithm can help provide predictive and prescriptive analytics. An optimization algorithm can also be developed based on the simulation results to suggest a recipe (input set-point parameters) for minimizing shrinkage porosity. The current thesis focuses on developing a Multi-physics model for LPDC of automobile wheel, followed by post-processing voxelization and density based algorithms to locate and quantify shrinkage porosity. It also discusses a Multi-Disciplinary design optimization approach for suggesting a recipe for minimal defects in the wheel. A machine learning approach is also discussed to predict and prescribe the wheel quality.