A method to aid in convergence of nonlinear equations by relocating the solver is introduced. Deactivating known variables and activating intelligently chosen unknown variables at predicted values efficiently formulates initial conditions. This repair strategy is integrated into an optimization procedure. The repair algorithm has the ability to (1) relocate the solver for non-converged models caused by poor initial conditions and (2) utilize points formulated in relocating non-converged models caused by infeasible sets of decision variables as a perturbation phase of a stochastic optimization algorithm. To show the effectiveness of the proposed repair strategy, a hybrid metaheuristic of a Variable Neighborhood Search (VNS) and Threshold Accepting (TA) is tested on the optimization of a large-scale industrial process, the primary units of a crude oil refinery. The performance of the hybrid VNS/TA Metaheuristic with repair is compared to the algorithm without the repair technique. (C) 2013 Elsevier Ltd. All rights reserved.