Hybrid Machine Learning and Physical Modeling of Feedstock Deformation During Robotic 3D Printing of Continuous Fiber Thermoplastic Composites

TL;DR

This paper presents a hybrid physics-based and data-driven modeling approach to predict feedstock deformation in robotic 3D printing of continuous fiber thermoplastic composites, addressing manufacturing defects.

Contribution

It introduces a novel hybrid model combining Kelvin-Voigt viscoelastic and neural ODE techniques to improve deformation prediction during printing.

Findings

Model accurately reproduces prepreg behavior across a wide temperature range.

Hybrid approach demonstrates robustness and generalization in real printing scenarios.

Experimental and numerical methods identify key deformation phenomena.

Abstract

Feedstock deformation during 3D printing of continuous fiber composites is a critical challenge in path planning and a main driver in the generation of manufacturing defects. The proposed work addressed the feedstock deformation during the deposition through several experimental and numerical pathways. The experimental setups and numerical simulations are used to identify the main driving phenomena in the deformation of feedstock through residual stress relief and drying, crystallization, and thermal stresses. A hybrid physics-based and data-driven modeling effort is performed, using Kelvin-Voigt viscoelastic modeling of the composite prepregs and a stabilized neural ODE for the modeling of drying and crystallization. The identified hybrid models from DMA and DSC experiments are used in robotic 3D printing to validate the deposition of a composite prepreg in real printing settings. The results show the ability of the model to reproduce the prepreg behavior far above the temperature used in the training, showcasing its robustness and generalization capability.