Strengthening mechanical performance with machine learning-assisted toolpath planning for additive manufacturing of continuous fiber reinforced polymer composites

Abstract

Additive manufacturing of continuous fiber composites enables the realization of complex yet optimal design, fully leveraging the transversely anisotropic mechanical properties of fibers by aligning the fiber direction with the principal stress direction. This offers a new design and manufacturing pipeline to enhance the structural efficiency of composites under specific working conditions. However, the computational efficiency of existing methods based on finite element analysis for calculating principal stress fields is low and unsuitable for low-volume and high-mix type production of additive manufacturing. Herein, this study proposes a machine learning-assisted toolpath planning method to reliably and efficiently generate the continuous fiber toolpath that strengthens the mechanical performance of composite structures. The method constructs a convolutional neural network enhanced with a self-attention mechanism to accurately predict regular principal stress direction field for complex geometries with given working conditions. Subsequently, toolpaths are extracted from a scalar field whose gradient is locally orthogonal to the stress direction, followed by redundant point removal, regrouping, and continuity operations to ensure the toolpaths satisfy the manufacturing constraints. Additionally, criteria for assessing both the mechanical performance and manufacturability of the toolpath are developed. By comparing the average computation time for 100 samples, it is demonstrated that the proposed method improves computational efficiency by 87.3 % compared to existing methods. Furthermore, when applied to various structures, the direction prediction error remains within 10°, and the differences in stiffness of the toolpath-integrated structures and manufacturability of the toolpaths are both within 10 %, demonstrating the reliability of the method for complex and varying geometries.