Abstract
Abstract
The ultimate aim of casting simulation is to recommend process parameter values that result in the best possible casting quality. However, commercial casting simulation is currently used as a trial-and-error tool, mostly comparing between possible scenarios with no guarantee of realism. Three techniques are presented to overcome these discrepancies for solid investment casting of jewellery as a test bed. The first technique consists in recording temperature versus time profiles by thermocouples embedded in the casting. These profiles are used as calibration reference regarding the overall heat transfer coefficient at the melt-mould interface in simulation. The second technique consists in metallographic validation of simulation findings in terms of correlating microstructure and defects of the casting with simulation results concerning temperature distribution and porosity. The third technique consists in introducing genetic algorithms as an optimization tool. A small number of simulation sample runs link process parameters of interest, such as mould temperature and melt temperature, with results characterizing casting quality, e.g. porosity. These samples are used to train a neural network, thereby creating a generalized meta-model of the process, which is directly used as the fitness function of the genetic algorithm. Combining the three techniques described above caters for realistic and practical casting process planning using commercially available casting simulation software as a tool.