A machine learning-based probabilistic computational framework for uncertainty quantification of actuation of...

Area of Science:

• Robotics and Mechanical Engineering
• Computational Science
• Data Science

Background:

• Clustered tensegrity structures offer lightweight, foldable, and deployable solutions for flexible manipulators and soft robots.
• The actuation of these soft structures exhibits high probabilistic sensitivity, necessitating accurate uncertainty quantification and deformation control.

Purpose of the Study:

• To develop a comprehensive data-driven computational framework for uncertainty quantification (UQ) and probability propagation in clustered tensegrity structures.
• To create a surrogate optimization model for precise control of flexible structure deformation.

Main Methods:

• Utilized machine learning methods, including Gaussian Process Regression (GPR) and Neural Networks (NN), to overcome nonlinear Finite Element Analysis (FEA) convergence challenges.
• Developed a surrogate model for real-time prediction of uncertainty propagation.
• Implemented Sequence Quadratic Programming (SQP) and Bayesian optimization for controlling actuated deformation.

Main Results:

• Demonstrated the validity of the data-driven approach using a clustered tensegrity beam subjected to clustered actuation.
• Achieved fast, real-time predictions for uncertainty propagation.
• Successfully optimized the actuated deformation of the flexible structure.

Conclusions:

• The proposed data-driven computational approach is effective for UQ and deformation control in clustered tensegrity structures.
• The framework shows potential for extension to other UQ models and optimization objectives in soft robotics and flexible manipulators.