Machine Learning-Based Prediction of Time Required to Reach the Melting Temperature of Metals in Domestic Microwaves Using Dimensionless Modeling and XGBoost

Resumen

A novel and cost-effective methodology is introduced for the precise prediction of the melting time of metals and alloys in a 700 W domestic microwave oven, using a hybrid SiC-graphite susceptor to ensure efficient heating without direct interaction with microwaves. The study includes experimental trials with multiple alloys (Sn-Bi, Zn, Zamak, and Al-Si, among others) and variable masses, whose results made it possible to construct a dimensionless model, trained with XGBoost on easily measurable thermophysical properties (specific heat, density, thermal conductivity, mass, and melting temperature). The model achieves high accuracy, with a relative error below 5%, and metrics of MAE = 4.8 s, RMSE = 6.1 s, and R2 = 0.9996. The generalization of the model to different microwave powers (600-1100 W) is also validated through analytical adjustment, without the need for additional experiments. The proposal is implemented as a Python application with a graphical interface, suitable for any academic or teaching laboratory, and its performance is compared with classical models. This approach effectively contributes to the democratization of thermal testing of metals in educational and research settings with limited resources, providing thermodynamic rigor and advanced artificial intelligence tools.