Physics-informed Tandem Neural Network for Rapid and Accurate Temperature History Prediction in Polymer Composite Extrusion-based Large-Format Additive Manufacturing
We propose a Physics-Informed Tandem Neural Network (PITNN) that couples an inverse model, inferring latent boundary conditions from a 30-second temperature observation, with a forward PINN to predict full temperature histories in large-format additive manufacturing. After fine-tuning on experimental data, the model achieves 2.2–2.5 °C mean absolute error (R² > 0.98) with a 2.3 ms inference time, enabling real-time layer time optimization and closed-loop thermal control.