DESIGN AND DEVELOPMENT OF A SIT-TO-STAND DEVICE USING A VARIATIONAL AUTOENCODER-BASED DEEP NEURAL NETWORK

This work brings together rigid body kinematics with machine learning to present a mechanism synthesis pipeline for design and development of a Sit-to-Stand (STS) device. Practical device design problems require multiple constraints to be satisfied simultaneously. Most of the focus in the past has been on satisfying the key functional requirements presented as a path or motion generation problem and being content with a handful of solutions obtained. We present a new design pipeline, which begins with effective and compact data generation, to leveraging a deep neural network for representation of coupler curves and mechanism parameters, and finally ending with new metrics for quantitative evaluation of design constraints and rank ordering design concepts. This framework is capable of generating a large number of plausible solutions while meeting design constraints. As an example, we present many single-degree-of-freedom six-bar mechanisms that satisfy the given constraints and are ranked-ordered on the basis of the metric. While the focus of this paper is on the design of STS motion for integration in a multi-functional mobility assist device, this approach is broadly applicable to device design problems in other areas as well.