Physics-informed machine learning model of a parallel manipulator with flexible links for control

• Grant number: 24/07967-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): August 01, 2024
• Effective date (End): July 31, 2025
• Field of knowledge: Engineering - Mechanical Engineering - Mechanical Engineering Design
• Principal Investigator: Maira Martins da Silva
• Grantee: Tony Jun Tanaka
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

Some ways to obtain dynamic models for the control of flexible and parallel manipulators are the Lagrange multipliers and finite element methods; however, they are not suitable for every type of system. In some cases, these methods end up being too slow for control implementation, as the generated model is large-scale. Thus, a different method of obtaining dynamic models should be used, and one alternative is the physics-informed machine learning model. This method requires fewer data than deep neural networks to function due to the imposition of physical laws, which generates additional information. Moreover, it is capable of integrating noisy data and mathematical models to generate simple and fast results to control. Therefore, the objectives of this research are: studying a manipulator to form a model in MSC Adams software, validating this model, exploring the literature on physics-informed machine learning, deciding the metamodel structure, data collection, model training, validation, and identification of limitations. The planar robotic manipulator with flexible links under investigation was developed at EESC-USP (FAPESP2018/21336-0). The results of this research will contribute to understanding the operation of this new type of machine learning and systems with parallel joints and flexible links.