This paper presents some pragmatic approaches and results to engineering optimization problems or issues in solving them in an industrial environment. The paper uses a mechanical linkage of a High Voltage Circuit Breaker as an example of multi-objective optimization problem and focuses on two typical issues that can be encountered with highly non-linear problems. First, while the input variables are independent or can be made such in certain ranges, these ranges are dependent on the variable values themselves. This introduces a non-trivial search or selection problem for the input variables. Second, with very non-linear problems the underlying numerical simulation model might have problems in converging, resulting in a high number of failed simulations. Particularly in optimization problems where input variables have large ranges and the simulation engineer might not have had the time or possibility to find solver- or other parameter settings to ensure numerical convergence for all combinations of input parameters. This makes finding optimal solutions even more difficult for the optimization algorithms since it is difficult for it to distinguish between a far from optimal (bad) result and a failed simulation. The mechanical model was built in a multibody simulation software and the thermodynamic model in lumped parameter or reduced dimensional modeling platform using the Modelica programming language. The simulation was run as a co-simulation using an in-house co-simulation interface between the mechanical and thermodynamic solvers. The optimization workflow and the simulation models were built in a commercially available simulation software. The paper discusses and compares different approaches and algorithms in solving the specific example. The algorithms used are MOGA-II, NSGA-II, pilOPT, last of which is a proprietary multi-strategy algorithm combining local and global search methods. Additionally, a manual heuristic (engineering) approach based on a DOE exploration of the design space and manual selection of the promising designs is compared to optimization algorithms in terms of efficiency of finding pareto optimal solutions. The different approaches include implementation of a two-level workflow to check input variable validity, penalizing violation of constraints in different ways and implementing logical fails instead of constraints on the objectives. The paper makes no attempt to generalize, and all the results presented are particular to the example problem. However, the authors believe they can be relevant to a wide range of engineering optimization problems.