Kinematic Synthesis and Analysis for Soft Robots with Compliant Mechanisms Using Fuzzy Logic and Neural Networks

Abstract

This dissertation presents a novel framework for the kinematic synthesis and analysis of Compliant Mechanisms (CMs) that leverages fuzzy logic and neural networks to address inherent uncertainties in their design and behavior. Traditional deterministic and probabilistic methods often fail to capture the full spectrum of CM performance or are computationally prohibitive. The core contribution is the development of a Fuzzy-Kinematic Synthesis Framework that reformulates mechanism design using fuzzy arithmetic. The classic Freudenstein's equation is transformed into a parametric fuzzy form, treating input and output angles as Triangular Fuzzy Numbers (TFNs) to enable region-based synthesis. Solving these equations yields fuzzy link lengths—defined regions encompassing all viable mechanism configurations. This framework quantifies performance uncertainty through a "Function Spread" metric, derived from the fuzzy output. Building on this, an Envelope-Driven Control Methodology is developed. Forward and Inverse Kinematic Envelopes define the achievable workspace and input-output relationships. Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are applied to serve as efficient surrogates for kinematic problems, drastically reducing computational cost. A Mamdani-type Fuzzy Inference System enables generative design within the performance boundaries, creating adaptable systems from a single mechanism. The methodology demonstrates closed-loop control by synthesizing inputs to trace arbitrary paths within the positional envelope. The framework is validated through detailed case studies, including a compliant surgical grasper. Results show efficient handling of kinematic complexity, design optimization via performance envelopes, and robust prediction for mechanisms with complex sensitivity profiles.