A multimodal data-driven design of low pressure die casting gating system for aluminum alloy cabin

Abstract

The design of gating systems for large and complex thin-wall castings requires the application of empirical equations and a substantial number of iterative experiments. In this study, a novel machine learning-based design strategy is proposed for the gating system of large and complex thin-walled castings. The gating system of castings was estimated using a multi-input multi-output neural network model, which incorporated visual, textual, and numerical multimodal features. The R2 value of the model accuracy reached 0.981, and the predicted parameters for the gating system were a slot runner width of 27 mm, a slot runner length of 48 mm, a riser runner diameter of 60 mm, and the number of riser runners of 8. The EasyCast and Procast simulation software were utilized for comparison and verification, resulting in a smooth filling and the achievement of sequential solidification. Finally, the mechanism of casting defects is discussed using the solidification phase-field model of EasyPhase system, elucidating how the alloy system can serve as a distinguishing feature in machine learning models that impact gating system parameters. This strategy is anticipated to enhance the efficiency of low pressure die casting process design, thereby serving as a valuable reference for intelligent process design in low pressure die casting.

Keywords

• Machine learning
• Low pressure die casting
• Gating system design
• EasyCast software
• Phase-field simulation