Dynamic synthesis augmented TimeGAN and adaptive temperature control for microwave heating

Abstract

Microwave ovens are valued for their convenience and efficiency; however, many models still face issues with heating accuracy. While simulation analyses have made progress in addressing these challenges, the complexity and time requirements of multi-scenario data collection remain a challenge, as the lack of sufficient real-world data hinders the effective evaluation of model performance. To address this issue, we propose the Dynamic Synthesis Augmentation-TimeGAN (DSA-TGAN), which integrates a Discriminative Guided Warping (DGW) module to generate data that captures both the primary features of the heating process and additional perturbation information, effectively simulating the variations in microwave heating. The generated data serves as a pseudo-training set for TimeGAN, which is trained through an adaptive framework to produce sufficient experimental data. Additionally, we demonstrate that fine-tuning the pre-trained DSA-TGAN with a small amount of data from different microwave models enables successful transfer learning. Leveraging the synthetic data and feature analysis algorithms, we developed a process-adaptive temperature control method that enhances the accuracy and stability of microwave heating. Experimental results confirm that the DSA-TGAN model achieves the goals of high-quality data synthesis and effective transfer learning, significantly enhancing microwave heating performance. In addition, the proposed data augmentation model can be widely used in other microwave heating fields such as chemical processing and material synthesis.