Development of advanced strategies for modeling, design and virtual sensing of multimaterial metallic-composite robots

Abstract

Robotic manipulators with lightweight design are becoming increasingly popular given their superior mobility, ease of setup and reduced dangers in case of collisions with humans. In contrast with what traditionally done for bulky manipulators, design of lightweight manipulators requires to take compliance of the components into account. Compliance should in fact be minimized in order to ensure accuracy, while keeping energy usage as low as possible to ensure efficient actuation. The requirements for energy efficiency and stiffness are often contrasting and difficult to address with traditional materials and design methodologies. Great performance improvements can be achieved employing advanced materials such as composites, thanks to their low specific weight and high stiffness. Lightweight manipulators made of composite materials would allow for better dynamic performances, improved safety and reduction of actuation power. The use of composites in robotic applications has been particularly limited both in research and industry, mainly due to the high cost and complex design choices required to avoid damage during machining. Hybrid multimaterial designs have been proposed in literature to solve these issues, with few components in composite materials and the most complex ones in traditional materials. The design of multimaterial mechanical systems results in a complex challenge. Given a set of requirements, their dependency on design parameters is often highly nonlinear and complicated to extract. Furthermore, the solutions require high quality while satisfying numerous and often contrasting requirements. Simulation based on numerical models can predict the performances and in turn enable automated design methods based on optimization algorithms. The numerical models employed in the design phase can then be updated based on experimental data to closely match any physical instance of the systems, constituting a Digital Twin (DT). The availability of a DT enables in turn applications in the operation phases such as Virtual Sensing (VS), which allows to indirectly estimate quantities that can result difficult, expensive or even impossible to directly measure. Models employed for optimization and VS need to satisfy two often contrasting requirements: high accuracy and computational efficiency. Common industrial-level numerical models require large sizes in order to guarantee accuracy, resulting in inefficient simulation and model updating procedures. This work aims at the development of accurate and efficient numerical modeling strategies for multimaterial mechanical systems that enable design optimization and VS applications. The systems on which the methodologies are applied are multimaterial robotic manipulators in which composite materials are employed to achieve high performance and energy savings. The work initially focuses on modeling strategies based on the Finite Element (FE) and Flexible MultiBody (FMB) methodologies. The use of component-level parametric Model Order Reduction (pMOR) techniques allows to define component-level and system-level models with high accuracy and efficiency both in terms of performance evaluation and design update. The proposed methodology is validated showing good results, with particular focus on the model of a 5-DOF robotic manipulator. A multimaterial version of the robotic manipulator is then designed through a MultiObjective Optimization (MOO) technique. The use of the developed efficient system-level models proves fundamental in this phase to grant efficiency and accuracy. The results demonstrate the gains in terms of maneuver accuracy and energy consumption resulting from the use of composite materials. The final part of the thesis employs the models for VS applications based on the Kalman Filter (KF) framework for the joint estimation of states, inputs and material parameters. The methodology is developed both for component-level and system-level applications, allowing to track the evolution of the system in time through a limited set of output-only measurements. A wide set of numerical and experimental validations shows good results.