Data-driven fast modeling of induction-heated feeder operation for metal casting processes

• Develop physics-based finite element models for induction-heated feeders. • Perform billet heating tests to validate the physics-based modelling method. • Simulate the feeder operation systematically and generate its operational database. • Develop a data-driven surrogate model based on the operational database. • Evaluate the performance of the fast data-driven model for the feeder operation. Electromagnetic induction heating of open-top feeders is considered an energy-efficient method to ensure continuous melt feeding while the metal product part solidifies during the casting process. Induction-heated feeders enable the use of smaller feeders, resulting in higher casting yields. To optimize its operation for high efficiency, it is essential to have a fast and reasonable estimation of induction heating power to the melt inside the feeder, which can be affected by melt properties (electric resistivity and magnetic permeability), induction coil operating parameters (current intensity and frequency), and process conditions (melt level position and temperature). Hence, a data-driven surrogate model is constructed to predict the induction heating power applied to the melt inside the feeder based on an operational database generated through numerical simulations using a verified physics-based induction-heated feeder model. The physics-based model is built using the commercial multiphysics modeling software package COMSOL, which incorporates the detailed geometry of the induction feeder, the physics of magnetic fields generated by induction coils, and the induction heating inside the melt. To verify the accuracy of the physics-based model, several billet heating tests are designed specially so that the induction heating power to the billet can be extracted. The results from the experiments and the simulations are compared. Physics-based simulations, rather than physical experiments, are employed to generate the operational database for data-driven, fast modeling. The effects of melt properties, induction coil operating parameters, and metal casting process conditions on induction heating power to the melt are analyzed using the data-driven surrogate model.