Geometry-based modeling of the mold filling process using neural networks

Composite materials have gained increasing attention in the past several years due to their superior mechanical properties and improved strength-to-weight ratio over traditional materials. Resin Transfer Molding (RTM) is an attractive composite processing method due to its potential for providing consistently superior parts at a lower cost than other composite manufacturing techniques. The resin transfer molding process involves a large number of variables that are linked to the design of the component, the selection and formulation of the constituent materials, such as resin and fiber, and the design of the mold and molding process. These variables are strongly coupled to the system performance, for example mold filling time and RTM part quality.; In this study, a geometry-based methodology is developed for process modeling of RTM using neural network techniques. The proposed process modeling approach is applicable to other manufacturing processes such as injection molding and casting. In this RTM model, the preforms are assumed to be thin and flat with isotropic, orthotropic or anisotropic permeabilities. The position of the weld lines formed by the merging of multiple flow fronts originated from specified inlet ports are predicted using a neural network based back-propagation algorithm. The neural network was trained with data obtained from simulation. The network was trained over a wide range of parameters and models and was applicable for a wide range of systems. This methodology is based on decomposition of part geometry into the subdomains containing only one inlet port and bounded by the part geometry and positions of the weld lines predicted using the neural network program. In addition, the neural network technique was also applied to predict the position of the weld lines formed by the recombination of a single flow front around the inserts in each subdomain. Once the mold was decomposed into subdomains containing only one inlet port, and the perimeter of the subdomains were identified, geometry-based solutions were applied to find the location of the vents required to avoid trapping air bubbles. Finally, the time required to fill the subdomains as well as the total mold filling time was found by analytical methods.; A variety of preforms with different shapes and with or without inserts were used to verify the approach. The location of the weld lines as well as the location of the vents predicted by the model were in a good agreement with the location of the weld lines and vents that were found by the simulation. Furthermore, the model was applied to predict the flow front advancement within the part, during the mold filling process. It was found that such flow front prediction is independent of the grid structure created within the part. The method is also applicable in modeling the edge-effect and race-tracking effect in a mold containing non-uniform fiber preform. The models developed in this study can be effectively utilized in iterative optimization methods where use of numerical simulation models is cumbersome. The savings in computational times and automated model evaluation resulting from the use of neural networks and domain decomposition approach for process simulations were the key advantages of this approach.