The transport equations for momentum, enthalpy, and chemical species are solved to simulate the reactive flow of polyurethane foam in a refrigerator cavity. The chemical reactions are described by a mechanism with four reactions and eight species. The numerical findings are also supported by dimensional arguments, which lead to important design attributes. Results prove that the model can be used not only to predict the flow features during the expansion of a multi-component foam, but also to determine the locations and the size of the air vents to avoid air bubbles trapped during the solidification process. It appears that the distribution of the vent holes must be evenly balanced around the cabinet cavity, where the larger holes must be located at positions far from the injector. To prevent the formation of large air voids, which cannot reach the vent holes by the action of gravity, there must be additional holes located on the lower surfaces of the cavity to purge the air trapped. K E Y W O R D S computational fluid dynamics, injection molding, multi-component flow, polyurethane foams, reactive flow

Hence, the lift force varies nonlinearly. Furthermore, for blunted-nose bodies, a two-vortex system appears, first symmetric, but, as the angle of attack becomes large, the asymmetric vortices develop at the rear of the body without vortex shedding. As a result, the region, where the side force is effective, covers the entire afterbody. On stream-

Using the data obtained from the computational fluid dynamics simulations, a back-propagation neural network model was developed to predict the velocity magnitudes and the instantaneous wall shear stresses in two patient-specific aneurysms. The models were also used to determine the effect of the blood composition on the rapture risk of the aneurysms. Based on the possible combination, five back propagation models were developed. The architecture of five models is determined based on number of neurons in the hidden layer. All the models in each algorithm were trained and tested. The accuracy of the developed models was evaluated through statistical analysis of the network output in terms of mean absolute error, root mean squared error, mean squared error, and error deviation. According to the results obtained, all BPA effectively predicted velocity magnitude and instantaneous wall shear stress. Model 1 was, however, less accurate when compared to the other five models, as it had one neuron in its hidden layer. The analysis confirms that the neuron number in the hidden layer play a definitive role in predicting the respective outputs. The performance assessment all of the back-propagation models revealed that the error incurred was acceptable. The algorithms’ training and testing in this study were satisfactory, since the network output was in reasonably good conformity with the target computational fluid dynamics result.