Machine Learning Applications for the Synthesis of Planar Mechanisms — A Comprehensive Methodical Literature Review

Abstract Recent published works in mechanism synthesis have shown promising results by employing machine learning-based methods such as deep neural networks. As a foundation for further research, an extensive literature review on the application of machine learning algorithms for planar mechanism synthesis is provided. Following the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) guidelines for systematic review processes, the scientific landscape of the last two decades has been mapped. In total, 29 research articles have been identified and analyzed. The results were filtered, grouped by the machine learning algorithms used, and matched with the different tasks that occur in the mechanism synthesis process. Our analysis shows four major research aspects that allow improvement. First, including other requirements that occur in real-world motion tasks besides target paths, like velocities, transmission angles and other design requirements. Second, considering multiple kinematic chains and including different joint types, instead of learning on a few selected mechanism types, as in a real-world application the mechanism type is an output of the synthesis process, and not an input. Third, comparing the efficiency of different machine learning algorithms for synthesis problems, as well as applying algorithms that are, for example, widely used in similar design engineering tasks, and fourth, creating and utilizing standardized datasets of labeled mechanisms.