Study of stochastically varying through the thickness permeability of woven fabric and its effect on void formation in liquid composite molding processes
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Quantifying the effect of geometric, material and process variability on the manufacturing process is the subject of this thesis specifically applied to a class of composite manufacturing processed called as Liquid Composite Molding (LCM). In LCM, the material which is a stack of woven or stitched fabrics, is placed in a closed mold and resin is injected to cover all the empty spaces between the fibers to fabricate a composite part. The 3D permeability network created by the stack of fabrics influences the resin flow path and any variability in the local permeability from one part to the next could influence the filling dynamics resulting in voids within the composite. The local permeability variability of the fabric material is caused by the size and the distribution of the pinholes that exist in the fabric at the woven or stitched junctions. ☐ In this work, first a method is introduced, to characterize the full 3D permeability tensor of a fabric from one experiment. Next, the local variability (pinholes) of a woven fabric is experimentally measured, and a methodology to statistically model it is formulated. The statistical properties of the pinholes are used to generate the pinhole sizes and distribution based on Monte Carlo method to conduct 3D flow simulations to investigate the dynamics of resin flow and void formation due to the presence and distribution of the pinholes. The effect of this material variability on the manufacturing process is quantified by predicting the final void content in the part numerically and validating experimentally. ☐ Having characterized the pinholes of a woven fabric with its statistical properties,. unsupervised machine learning methods of dimensionality reduction (Principal Components Analysis (PCA) and Stochastic Neighbor Embedding (t-SNE)) are employed to extract the statistical properties of the pinholes from an image of a woven fabric. The extracted statistical properties are then used to create the pinhole sizes and distribution to conduct 3D flow simulations to predict void formation for varying properties of the pinhole distribution within the fabric and other material and process parameters.. Five parameters (pinhole properties, permeability of fabric, permeability of distribution media, and vent pressure) were identified as important variables that would influence the void fraction and size of the voids. The results were then organized using a Decision Tree method for more efficient analysis and classification. This study should prove useful in providing the guidelines to design a robust process which accounts for the input material, process and geometric variabilities to fabricate composites without voids.