Evolutionary computing techniques for handling variable interaction in engineering design optimisation

The ever-increasing market demands to produce better products, with reduced costs and lead times, has prompted the industry to look for rigorous ways of optimising its designs. However, the lack of flexibility and adequacy of existing optimisation techniques in dealing with the challenges of engineering design optimisation, has prevented the industry from using optimisation algorithms. The aim of this research is to explore the field of evolutionary computation for developing techniques that are capable of dealing with three features of engineering design optimisation problems: multiple objectives, constraints and variable interaction. An industry survey grounds the research within the industrial context. A literature survey of EC techniques for handling multiple objectives, constraints and variable interaction highlights a lack of techniques to handle variable interaction. This research, therefore, focuses on the development of techniques for handling variable interaction in the presence of multiple objectives and constraints. It attempts to fill this gap in research by formally defining and classifying variable interaction as inseparable function interaction and variable dependence. The research then proposes two new algorithms, GRGA and GAVD, that are respectively capable of handling these types of variable interaction. Since it is difficult to find a variety of real-life cases with required complexities, this research develops two test beds (RETB and RETB-II) that have the required features (multiple objectives, constraints and variable interaction), and enable controlled testing of optimisation algorithms. The performance of GRGA and GAVD is analysed and compared to the current state-of-the-art optimisation algorithm (NSGAII) using RETB, RETB-II and other ‘popular’ test problems. Finally, a set of real-life optimisation problems from literature are analysed from the point of variable interaction. The performance of GRGA and GAVD is finally validated using three appropriately chosen problems from this set. In this way, this research proposes a fully tested and validated methodology for dealing with engineering design optimisation problems with variable interaction.