UBC Theses and Dissertations
Design evolution of engineering systems using bond graphs and genetic programming Samarakoon Mudiyanselage, Buddhika
Engineering design is a complex task, which typically involves multiple physical domains. It can benefit from a modeling tool that can represent different domains in a unified manner. For complex designs, optimization by traditional techniques (such as gradient-based methods) may not be appropriate. Evolutionary methodologies may be used in design optimization of complex engineering systems. This research is based on a framework for evolutionary design system consisting of a machine health monitoring system, model-based evolutionary design optimization system, and a design expert system. The design weaknesses and faults of an existing engineering system are identified using a machine health monitoring system. Then under supervision of the design expert system, an optimal design is evolved using genetic programming. This thesis primarily addresses the modeling and evolutionary aspects and their integration. The thesis develops the integrated system consisting of bond-graph modeling and genetic programming. The performance of the developed system is studied using both experimentation and simulation. The drawbacks of the fitness calculation methodologies that are presented in literature are identified and improved fitness functions are developed in the present work. A methodology to automatically obtain the state-space model of a system represented using bond graphs is also developed. While previous researchers have investigated the integration of bond graphs and genetic programming in design, they have not applied the method in a real engineering system. The present work specifically addresses the application of the developed method for design improvement for an industrial machine. For this purpose a linear bond graph model of the industrial fish processing machine is developed and the parameter values are identified using genetic programming. The design of the actual system is modified according to the evolved bond graph model and the results are validated using the data from the actual engineering system. The proposed method is applicable particularly to actual systems, first because the initial model can be tested by comparing its simulated results with the corresponding results from the actual system, and second because the design improvements as suggested by the evolutionary design framework may be implemented and tested against the behaviour of the corresponding model.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International