Robust engineering design of electronic circuits with active components using genetic programming and bond Graphs

Abstract

Genetic programming has been used by Koza and manyothers to design electrical, mechanical, andmechatronic systems, including systems with both activeand passive components. This work has often requiredlarge population sizes (on the order of ten thousand)and millions of design evaluations to allow evolutionof both the topology and parameters of interestingsystems. For several years, the authors have studiedthe evolution of multi-domain engineering systemsrepresented as bond graphs, a form that provides aunified representation of mechanical, electrical,hydraulic, pneumatic, thermal, and other systems in aunified representation. Using this approach, called theGenetic Programming/Bond Graph (GPBG) approach, theyhave tried to evolve systems with perhaps tens ofcomponents, but looking at only 100,000 or fewer designcandidates. The GPBG system uses much smallerpopulation sizes, but seeks to maintain diverse searchby using sustained evolutionary search processes suchas the Hierarchical Fair Competition principle and itsderivatives. It uses stochastic setting of parametervalues (resistances, capacitances, etc.) as a means ofevolving more robust designs. However, in past work,the GPBG system was able to model and simulate onlypassive components and simple (voltage or current, inthe case of electrical systems) sources, which severelyrestricted the domain of problems it could address.Thus, this paper reports the first steps in enhancingthe system to include active components. To date, onlythree models of a transistor and one model of anoperational amplifier (op amp) are analysed andimplemented as two-port bond graph components. Theanalysis method and design strategy can be easilyextended to other models or other active components oreven multi-port components. This chapter describesdesign of an active analog low-pass filter withfifth-order Bessel characteristics. A passive filterwith the same characteristics is also evolved withGPBG. Then the best designs emerging from each of thesetwo procedures are compared. [The runs reported hereare intended only to document that the analysis toolsare working, and to begin study of the effects ofstochasticity, but not to determine the power of thedesign procedure. The initial runs did not use HFC orstructure fitness sharing, which will be included assoon as possible. Suitable problems will be tackled,and results with suitable numbers of replicates toallow drawing of statistically valid conclusions willbe reported in this paper, to determine whetherinteresting circuits can be evolved more efficiently inthis framework than using other GP approaches.]