The scientific question is whether a surrogate (low-fidelity) machine model (Fig. 1) can be employed to accelerate a computationally expensive (high-fidelity) finite element machine simulation (Fig. 2).

The research hypothesis is that a well constructed surrogate may perform better than a pure algebraic or stochastic surrogate for standard machine types. It is expected that an established machine model can be trusted in a region which is substantially larger than a standard kriging surrogate of the high-fidelity model. In order to preserve accuracy, the surrogate model will be adapted algebraically such that it locally has at least a linear consistency with the finite-element model [1]. This will be achieved by additive or multiplication defect corrections. In particular, an additive correction with quasi-second-order consistency will be set up using Broyden-Fletcher-Goldfarb-Shanno updates for the Hessian of the high- and low-fidelity models.