Manifold learning-assisted uncertainty quantification of system parameters in the fiber metal laminates hot forming process

The forming quality of Fiber metal laminates (FMLs) heavily depends on the material properties, fiber placing angles, blank holder force, and other process parameters. In some circumstances, the numerical perturbation of the key parameters has a potential impact on the mechanical properties of final products. To efficiently design a set of available system parameters to ensure the forming quality, a manifold learning-assisted approximate Bayesian computation (ABC) method is proposed to identify system parameters with uncertainties. In this study, the nonlinear manifold learning approach is employed to extract the feature vector of physical field information of sheet metal and composite core after hot forming. Furthermore, the mapping transformation of system parameters based on different modeling techniques is performed to shorten the time of obtaining feature vectors of new samples. The nested sampling method involving the wavelet mutation strategy is proposed to improve the sampling efficiency of the posterior distribution of system parameters while the tolerance criterion is guaranteed. Two hot stamp-forming cases are employed to validate the feasibility of the proposed approach. The numerical results show that the proposed method is effective in obtaining the system parameters necessary for achieving the high-quality forming of FMLs.