A Novel Machine Learning Solution for the Inverse Heat Conduction Problem with Synthetic Datasets

There are many attempts to solve the Inverse Heat Conduction Problem, and nowadays a growing number of machine learning-based methods have emerged. One major drawback of these methods is that they are very sensitive to the size and quality of the training database. There are several data augmentation techniques for artificially increasing the size of training databases, but these techniques have not yet been investigated in the field of quenching. This paper presents the augmentation methods that can be used, and then evaluates them with a novel experience. As a final tough, we can conclude that modern synthetic data generation can develop the robustness of machine learning methods and play an effective role in the inverse heat conduction problem occurring during the quenching of steel.