Actual path prediction and correction in tailored fiber placement process through a machine-learning algorithm

Abstract

Tailored Fiber Placement (TFP) is an additive manufacturing process for composite materials. It is based upon the embroidery technique, where continuous fiber rovings are positioned in a textile base material, fixed by a stitching yarn, and are later impregnated in a matrix. Its main advantage consists of the high design freedom, allowing the production of variable-axial reinforcements, depositing fibers only where desired and with optimized orientation. However, when dealing with small dimensions, deviations can be clearly observed between the target path (designed path) and the actual path, effectively stitched by the TFP machine. Among the sources for these deviations, one can mention fiber’s waviness, buckling, materials involved, influence of input parameters, i.e., the tension applied at the roving, the distance between stitching points, speed, and other systematic deviations inherent to any manufacturing process. The present work focused on measuring these deviations and analyzing the influence of the stitching parameters on them. It was observed that the distance between stitching points plays a major role in this regard. After collecting the stitching data with varying inputs, a machine learning algorithm was trained to predict the deviations and to propose a second target path, whose objective is to minimize the deviations between the first target path and the actual path. The machine learning algorithm was able to minimize the average deviation over the target path by up to 55%.