Integrated coevolutionary synthesis of mechatronic systems using bond graphs

Mechatronics is a natural stage in the evolution of modern products, many containing components from different engineering domains, such as mechanical, electrical, and software control systems. As part of concurrent engineering practice, mechatronics is a synergistic system design philosophy to optimize the system as a whole simultaneously. Yet there is still lack of support of this design principle in practice. To date, conventional design tools have been limited to single domain problems and require a trial-and-error synthesis process. In order to support the concurrent synthesis process of mechatronic products, theoretical modeling of multi-domain engineering systems, with a formal unified representation and a well-defined algorithmic and flexible synthesis procedure, is needed. These are essential to accommodate the complexity of such systems and support the design automation process. In this work, multi-domain mechatronic system design is treated as a network synthesis problem, extending from single domain electrical network synthesis. Desired design performance is specified in an impedance matrix that captures the dynamic relations of effort and flow variables at input-output interaction ports. An extended multi-port bond graph representation is developed to unify power and signal flows at a high-level abstraction across engineering domains, which also integrates active control system design. The unified representation of both physical systems and their control systems in bond graphs is achieved by applying “controller design in the physical domain” philosophy, to design and synthesize the whole system simultaneously at the conceptual design stage. The graphical structure of bond graphs being close to reality also gives intuitive physical insight of the interactions among physical components for detailed level design realization and simulation in different domains to verify the entire system. This approach makes full use of computational power to automatically explore the design space for both design configuration and parameterization utilizing biology-inspired optimization techniques: genetic algorithms, genetic programming, and coevolution. Bond graph elements are encoded as genetic programming functional and terminal primitives, to evolve low-level building blocks to high-level functionality by applying genetic operations based on population-based natural evolution. It aids design exploration of a wider range of possible creative design options and achieves synergy in coevolving different subsystems, including both active control strategies and physical system design configurations, for overall system optimality. Two mechatronic design case studies are provided: a one-axis robotic manipulator system and a quarter-car suspension system. The computational results are compared with the design solutions obtained by human designers from trial-and-error synthesis and theoretical analysis. The coevolutionary synthesis approach is capable of discovering design options with better performance, more creativity and flexibility than those perceived by human designers. It is our belief that the establishment of such mechanism will enhance the capability of computers to automatically generate and evaluate innovative and alternative solutions to multi-domain dynamic systems and enable intelligent assistance to engineering designers at the early stages of system modeling and development for concurrent engineering practices.