Development of an optimisation framework for solving mixed-variables engineering design problems (LIUU16SF)

Real-world optimization problems are often complex and nonlinear. Here, the word ‘complex’ means two aspects: 1) the computational effort involved in the optimization loop is significantly expensive, for example, vehicle crashworthiness design; 2) the number of design variables in the optimization problems can be quite large (normally, we consider the number of design variables is less than 100) in many prominent application domains. The above features of complex models have to be considered in many optimization formulations and these keep a challenge to many existing optimization solvers.
Another complication of such optimization problems is that not all design variables are continuous, and some variables can only take certain discrete values. Mixed continuous/discrete optimization problems usually require special search techniques to find the optimum. However, there is no guarantee the optimum is the global best solution.

Taking the above situation into account, it is essential to develop a cost-effective and efficient platform to meet the above challenges from the complex structural optimization. This requires a common level of understanding of various optimization techniques, structural performances involved, especially high performance computing (HPC) applications.

The objectives of this PhD project are:
• To formulate the optimization problem and set up a library of algorithms which will efficiently accommodate different/specific optimization cases;
• To conclude a robust, effective, and efficient algorithm suitable to solve complex optimization problems by trade-off analysis between the efficiency of algorithms and the degree of accuracy.
• To implement the developed library in the existing optimizer to form a cost-effective and efficient framework applicable to the HPC system;
• To examine the robustness of the developed framework by various benchmark examples.