Deep convolutional neural network aided optimization for cold spray 3D simulation based on molecular dynamics

This study proposed a deep convolutional neural network (DCNN) aided optimization (DCNNAO) method to improve the quality of deposition during the cold spray process which was simulated by molecular dynamics (MD). The idea of the DCNNAO is to extract the value of the objective function from the MD simulation snapshots directly by DCNN aided image process technique. Considering the huge memory requirement for MD result files, the main superiority of DCNNAO is to reduce the memory requirement and improve the efficiency of the optimization process by using a contour image (several hundred kilobytes) as the input instead of an MD result file (several hundred gigabytes). To complete this strategy, a Python script is written to generate required snapshots from result files automatically. Moreover, three boosted decision trees based optimization methods including surrogate optimization and heuristic algorithms are also implemented for comparison study. A detailed optimization result demonstrates that all the above methods can obtain an acceptable solution. The comparison is also given for an informed selection of them based on the trade-off between efficiency and accuracy.