Data-driven surface temperature prediction for variable tool geometries in automated fiber placement

Abstract

Accurate surface temperature prediction is critical for ensuring quality control and process optimization in automated fiber placement (AFP). While traditional heat transfer modeling approaches rely on finite element analysis (FEA) and numerical methods, they often struggle to generalize across different tool geometries and heating mechanisms because they are typically tailored to specific conditions and require substantial reformulation when conditions change. This study introduces a data-driven modeling approach to predict applied surface temperature during AFP layup. A polynomial regression model was developed using experimental data collected from infrared (IR) and pulsed light (PL) heating systems across various processing parameters, including heater power, layup speed, distance-to-surface, and p-angle (AFP end-effector head tilt relative to the nip-point). A 10-fold cross-validation demonstrated strong predictive accuracy, yielding coefficient of determination, $R^2$, values of $0.914$ and $0.916$ for the IR and PL models, respectively. A manufacturing case study further demonstrated the ability of the model to predict temperature variations across flat and complex tool surfaces, while flux knockdown experiments were used to quantify temperature distribution effects. Experimental validation using thermocouple measurements confirmed the accuracy of the model in predicting surface temperature, with a mean percent error of $3.01\%$, highlighting the model's potential for real-time AFP process monitoring. While the model effectively captures key thermal behaviors, future work will focus on incorporating two- and three-dimensional thermal effects, integrating physics-based modeling, and expanding validation to laser-assisted AFP heating. This research advances machine learning-driven heat transfer modeling in AFP, paving the way for intelligent composite manufacturing.