Pattern Matching Synthesis as an Automated Approach to Mechanism Design

Abstract

A new knowledge-based approach for the synthesis of mechanisms, referred to as Pattern Matching Synthesis, has been developed based on a combination of committee machine and Hopfield neutral network models of pattern classification and matching applied to coupler curves. Computational tests performed on a dimensionally-parameterized four bar mechanism have yielded 15 distinct coupler curve groups (patterns) from a total of 356 generated coupler curves. This innovative approach represents a first step toward the automation of mapping structure-to-function in mechanism design based on the application of artificial intelligence programing techniques. Demonstrative examples of its application to “real-world” mechanism synthesis problems, including the design and evaluation of a two-stroke pump mechanism and the redesign of a variable-stroke engine mechanism have been included, establishing its viability for creative mechanism synthesis.