Fusion-Based Neural Generalization for Predicting Temperature Fields in Industrial PET Preform Heating

TL;DR

This work tackles predicting 2D temperature fields during microwave preheating of PET preforms with limited data. It combines transfer learning and model fusion to achieve robust generalization across material (virgin vs. recycled PET) and geometry variations, leveraging 2D axisymmetric simulations to reduce data demands. Two case studies demonstrate substantial generalization gains using a fusion-based approach, outperforming models trained from scratch on larger datasets. The approach offers a scalable, data-efficient path toward intelligent thermal control in manufacturing, with potential applicability to other complex physical processes requiring limited-data modeling.

Abstract

Accurate and efficient temperature prediction is critical for optimizing the preheating process of PET preforms in industrial microwave systems prior to blow molding. We propose a novel deep learning framework for generalized temperature prediction. Unlike traditional models that require extensive retraining for each material or design variation, our method introduces a data-efficient neural architecture that leverages transfer learning and model fusion to generalize across unseen scenarios. By pretraining specialized neural regressor on distinct conditions such as recycled PET heat capacities or varying preform geometries and integrating their representations into a unified global model, we create a system capable of learning shared thermal dynamics across heterogeneous inputs. The architecture incorporates skip connections to enhance stability and prediction accuracy. Our approach reduces the need for large simulation datasets while achieving superior performance compared to models trained from scratch. Experimental validation on two case studies material variability and geometric diversity demonstrates significant improvements in generalization, establishing a scalable ML-based solution for intelligent thermal control in manufacturing environments. Moreover, the approach highlights how data-efficient generalization strategies can extend to other industrial applications involving complex physical modeling with limited data.

Figures (7)

• Figure 1: (a) Molded and defective PET bottles; (b) Preforms with different designs.
• Figure 2: Schematic representation of the microwave heating optimization process for PET preforms. The default temperature profile (left) shows uneven heating typically observed without slabs modifications. A custom-designed microwave cavity with adjustable dielectric slabs (center-left) enables fine-tuning of the electromagnetic field distribution. The optimized slab con-figuration (center-right) is tailored to the preform geometry, leading to a significantly im-proved and more uniform temperature profile (right). . Adapted from García-Baños et al. garcia2022.
• Figure 3: Model fusion workflow combining outputs from variant-specific models into a unified generalized predictor.
• Figure 4: Comparison of MLP architectures: (a) standard and (b) with skip connections ("Add") for improved gradient flow and stability.
• Figure 5: (a) Heat capacity vs. temperature for virgin and recycled PET. (b) Resulting temperature profiles along preform length under different heat capacity assumptions.