On Model Order Reduction in Simulations of Laser Shock Peening

Laser shock peening (LSP) is a modern but already established approach to improve the strength and fatigue resistance of components by hardening their surface.LSP is based on the use of a high-energy laser to generate mechanical loading of the component surface.Compared to standard peening techniques, e.g., shot peening, LSP allows for precision surface treatment since the individual laser spots can be positioned with an accuracy of the order of 0.1 mm.Furthermore, the laser energy ( 1 J) as well as the spot size ( 1 mm) are tunable parameters.This process tunability allows for component-and material-based optimization that can be accelerated through simulations.However, simulations of LSP treatment of real-life components are time consuming in such a way that it disallows their direct usage in process optimization.In this contribution, we present an approach to mitigate the computational costs of simulation-based LSP process optimization.The approach is based on sampling the optimization parameter space by computing several simulations while varying the optimized parameters.Next, the precomputed solutions are used to construct a reduced order model (ROM) usable in optimization.ROM in question leverages proper orthogonal decomposition (POD) and artificial neural networks (ANNs).Compared to the original full order model (FOM), ROM evaluation is by orders of magnitude faster.Furthermore, the loss of information between FOM and ROM can be controlled during the ROM construction.