A Large-scale Bi-objective Optimization of Solid Rocket Motors Using Innovization.

Many design optimization problems from practice involve a large number of variables. In handling such problems, optimization algorithms, in general, suffer from the well-known ”curse of dimensionality” issue. One of the ways to alleviate the issue somewhat is to use problem information to update the optimization algorithm so that more meaningful solutions are evolved quickly. In this paper, we consider a solid rocket motor design problem involving hundreds of integer variables and two conflicting objectives – minimize the error in matching developed thrust with a desired time-dependent thrust profile and simultaneously minimize the unburnt residue of propellant at the end of the burning process. The evaluation of both objectives involve a detailed burn simulation from the core to the shell of the rocket. After finding a set of trade-off solutions using an evolutionary multi-objective optimization algorithm, we use two learning-based optimization methods (akin to the concept of innovization) to find similar set of solutions using a fraction of the overall solution evaluations. The proposed methods are applied to seven different thrust profiles. Besides solving the large-scale problem quicker, a by-product of our approach is that learnt innovized principles stay as new and innovative knowledge for solving the solid rocket design problem, a matter which is extremely useful to the practitioners.