The project has its background in a strong industrial drive to manufacture both larger and more complex die-cast aluminum components. This places entirely new demands on the gating system design for the tools that shape the component. The goal is to develop a physics-based machine learning model that can predict and optimize gating parameters for die casting. This will reduce development time, increase component quality, and decrease resource consumption for die-cast aluminum components. By integrating physical models and principles with machine learning algorithms, more robust and realistic simulations, improved optimization, and better decision-making can be achieved. The approach of the project is to develop and link flow dynamics calculations with machine learning algorithms, thereby providing designers with entirely new tools for design support during the development of gating systems for die casting. The project will continue up to the demonstration of a real component. The results will be beneficial to the entire die casting industry, as the developed method can be used to design gating systems for any die-cast components. Furthermore, the results are expected to have a positive impact on the manufacturing sector as a whole, through increased awareness, and thereby contribute to industrial digitalization in Sweden. The method development will be carried out in several steps. First, melt flow and solidification models will be developed and validated for die casting, using the open source OpenFOAM computational fluid dynamics (CFD) software. The validation will be conducted using already existing data as well as previous work at RISE. Verification simulations in commercial casting simulation software will be carried out by the industry. The generated flow data will be used for the training of AI models, for example in the encoding step of a Convolutional Neural Network (CNN) or the first step of an Autoencoder, in order to identify the most significant variables (latent variables), which in this case correspond to the geometric design — the converging step. In the following step (the decoder step), alternative and physically constrained solution proposals for the optimized ideal design are generated — the diverging step. The project consortium consists of: -RISE Research Institutes of Sweden: Project leader and responsible for research activities. -Volvo Cars: Contributing to research, process simulations, CAE. -Husqvarna AB: Process simulation, validation. -AGES Kulltorp/UB Verktyg: Process simulation, prototyping of demonstrator.