Fiber Orientation Prediction at Multiple Thicknesses in a PPMC Tensile Coupon

Abstract

This study evaluated the ability of artificial intelligence tools to reconstruct local fiber orientation distribution (FOD) at multiple thicknesses of a prepreg platelet molded composite (PPMC) specimen. A deep convolutional neural network (DCNN) was employed to accurately predict FOD at three thickness regions and the total thickness of a molded specimen by using thermally induced strain on the surfaces of the specimen. The developed DCNN architecture was trained with thousands of synthetic finite element morphologies of PPMC plates. The DCNN performed well on the synthetic data and predicted the FOD in the specimen with a 0.071 mean absolute error (MAE) for the entire specimen thickness and 0.128, 0.145 and 0.125 MAE for the top, middle and bottom thirds of the specimen thickness, respectively. The proposed methodology predicted the spatially varying FOD at multiple thickness levels in PPMC parts and can be used for a nondestructive evaluation process to detect erroneous fiber orientation in PPMC components.