Multimodal Data-Driven Design of Low Pressure Die Casting Gating System for Aluminum Alloy Cabin

Abstract

The design of gating systems for intricate and sizable thin-walled castings necessitates the utilization of empirical equations and a significant number of iterative experiments. This procedure is known to be demanding in terms of time and labor.Based on the machine learning method, this paper proposes a design strategy of "casting material - casting structure characteristics - gating system process parameters" for the low-pressure die casting gating system process, taking the large aluminum alloy complex cabin as the research object. This strategy enables the intelligent design of casting gating systems through a non-sequential multi-input and multi-output model with a common layer capable of handling multimodal data. The reliability of the machine learning model design results was verified using the prediction results of the casting simulation software as a quality indicator. To address the challenge of limited interpretability in neural networks, a phase field model is employed to investigate the origins of casting defects arising from variations in the solidification process of different materials. This analysis aims to elucidate the underlying mechanisms through which casting material impacts the gating system. Additionally, this approach serves to underscore the intelligent nature of the proposed strategy for gating system design. The casting model designed based on this strategy is smoothly filled and solidifies sequentially in EasyCast software. This strategy improves the traditional low-pressure die casting process design method, which has long cycle time, high cost, and low efficiency. It also provides an important reference for the design and development of intelligent low-pressure casting process.

Keywords: Machine learning, low pressure casting, gating system design, EasyCast, solidification phase field