Non-intrusive optimal experimental design for large-scale nonlinear Bayesian inverse problems using a Bayesian approximation error approach | Synapse