Generating realistic estimates of AFP tow geometry for gap and overlap predictions

Abstract

Gap and overlap defects are formed during automated fiber placement (AFP) layup due to stochastic variations in tow position and geometry. There is utility in predicting the size and frequency of these defects since they affect the manufacturing rate and structural performance. Two methods which reflect the measured process variability were implemented to simulate realistic tow geometry. A Monte Carlo (MC) simulation used independent random sampling from the distributions of the sources of variation. A Markov Chain Monte Carlo (MCMC) simulation used the Random Walk Metropolis algorithm which samples these distributions while considering the neighborhood of past samples thus better preserving spatial continuity. Simulated tows were assembled into virtual laminae. Tow-level validation using spatial-frequency Fast Fourier Transform (FFT) analysis demonstrated that the MCMC method reproduces the dominant wavelengths and spectral characteristics of experimentally measured tow waviness, whereas the MC method introduces unrealistic high frequency variations. Lamina-level comparisons showed that while both methods predict similar gap and overlap area percentages, only the MCMC method accurately captures defect length distributions. The proposed framework provides realistic AFP layup geometries for use in mesoscale modeling and for informing process improvement efforts.