Increasing the quality and reliable reproducibility of large-size composite structures molding using the vacuum infusion method, which is gaining popularity in various industries, is achieved in practice through numerous tests by try and errors that require significant costs and time. The purpose of these tests is to determine the layout of the ports for the resin injection and vacuum supply, as well as the temperature regime that ensures the absence of isolated non-impregnated zones, the minimum porosity and the required reinforcement volume fraction in the composite. The proposed approach removes the simplifying assumptions used in commercial software for modeling the process, which reduce the accuracy of reconstruction of its dynamics and the sensitivity to the formation of unrepairable defects such as dry spots. It involves multiphysics modeling of resin filling in a porous preform by describing the resin front dynamics by the phase field equation, pressure distribution in an unsaturated porous medium by the Richards equation, the evolution of the degree of cure by the convection / diffusion / thermokinetics equation, and thermal processes by the heat transfer equation using modified models of viscosity, the diffusion coefficient of the degree of cure, the boundary condition for the vacuum port. To reduce the finite element computation time of the investigated variants of the process, which is necessary for its computer optimization, the predictive partial sub-criteria were used, which give a reliable prediction before the beginning of the resin gel and solidification. Due to this, a gain in computation time is 30-50% with a significant prediction accuracy of quality objectives and the presence of possible defects.