Neural network-integrated multiobjective optimization of the 3D-printed conformal cooling channels

The present study proposes a neural network-supported multiobjective optimization procedure for two different conformal cooling channel designs in plastic injection technology. After completing the 3D design of a plastic model, the computational heat transfer & fluid dynamics simulations find the cooling time, temperature uniformity, and pressure drop data, which are the main objectives of the minimization-based multiobjective optimization problem. The performance maps of the objectives are created via i) a classic MATLAB-based interpolation and ii) MATLAB-based neural networks. Then, multiobjective optimization is conducted. The results show that the neural networks can create the performance maps of the cooling time trends with less than 1% error whereas its best trade-off point calculation has nearly 1.1 % deviation. Also, the use of neural networks can decrease the computational time in both simulation and optimization processes.