Adaptive search and the preliminary design of gas turbine blade cooling systems
This research concerns the integration of Adaptive Search (AS) technique such as the Genetic Algorithms (GA) with knowledge based software to develop a research prototype of an Adaptive Search Manager (ASM). The developed approach allows to utilise both quantitative and qualitative information in engineering design decision making. A Fuzzy Expert System manipulates AS software within the design environment concerning the preliminary design of gas turbine blade cooling systems. Steady state cooling hole geometry models have been developed for the project in collaboration with Rolls Royce plc. The research prototype of ASM uses a hybrid of Adaptive Restricted Tournament Selection (ARTS) and Knowledge Based Hill Climbing (KBHC) to identify multiple "good" design solutions as potential design options. ARTS is a GA technique that is particularly suitable for real world problems having multiple sub-optima. KBHC uses information gathered during the ARTS search as well as information from the designer to perform a deterministic hill climbing. Finally, a local stochastic hill climbing fine tunes the "good" designs. Design solution sensitivity, design variable sensitivities and constraint sensitivities are calculated following Taguchi's methodology, which extracts sensitivity information with a very small number of model evaluations. Each potential design option is then qualitatively evaluated separately for manufacturability, choice of materials and some designer's special preferences using the knowledge of domain experts. In order to guarantee that the qualitative evaluation module can evaluate any design solution from the entire design space with a reasonably small number of rules, a novel knowledge representation technique is developed. The knowledge is first separated in three categories: inter-variable knowledge, intra-variable knowledge and heuristics. Inter-variable knowledge and intra-variable knowledge are then integrated using a concept of compromise. Information about the "good" design solutions is presented to the designer through a designer's interface for decision support.