Simulation-based selection of optimum pressure die-casting process parameters using neural nets and genetic algorithms

Neural network meta-models are suggested in this work in order to generalise from examples connecting input process variables to output variables, such as filling time, solidification time, defects, etc, and can help in avoiding excessive numbers of what-if scenarios examined on the casting simulation software.

Abstract

Pressure die-casting condition selection mainly relies on the experience and expertise of individuals working in production industries. Systematic knowledge accumulation regarding the manufacturing process is essential in order to obtain optimal process conditions. It is not safe a priori to presume that rules of thumb, which are widely used on the shop floor, always lead to fast prototype production calibration and to increased productivity. Thus, neural network meta-models are suggested in this work in order to generalise from examples connecting input process variables, such as gate velocity, mould temperature, etc., to output variables, such as filling time, solidification time, defects, etc. These examples, or knowledge, are gathered from experiments conducted on casting simulation software, which are designed systematically using orthogonal arrays (DoE). They could also be based on experiments from industrial practice. Neural models derived in this way can help in avoiding excessive numbers of what-if scenarios examined on the casting simulation software, which can be very time-consuming. Furthermore, they can be employed in the fitness function of a genetic algorithm that can optimise the process, i.e. yield the combination of input parameters which achieves the best output parameter values.