Dynamic identification of industrial robots

In the field of industrial robotics, manipulator robots are articulated electromechanical systems designed to interact with their environment via a tool attached to their end effector. These systems are inherently closed-loop: their mechanical structure consists of rigid solids connected by joints, forming a closed system whose dynamic behavior must be precisely controlled to ensure performance, safety, and reliability.

Controlling these robots requires knowledge of a model of the system. This mechanical model, which is central to many robotic applications, enables not only movement control, but also simulation, design, diagnosis, and fault detection. It can be broken down into several levels of complexity: geometric models that locate the tool in the workspace, kinematic models that establish links between joint speeds and effector speeds, and dynamic models that describe the interactions between forces and accelerations.

It is the dynamic models that interest us in the context of this article. Their use in advanced control algorithms, such as those designed to compensate for forces or generate optimal trajectories, requires detailed knowledge of their physical parameters, such as masses, moments of inertia, and first moments. These parameters are not generally provided with sufficient accuracy by manufacturers or design tools, and must therefore be identified experimentally.

Dynamic identification is based on the use of measurements recorded as the robot moves along defined trajectories. It is therefore an offline identification of parameters. It is a fundamental process for ensuring that the robot's actual behavior matches its mathematical model. This process is delicate because it involves complex systems, often subject to safety and accessibility constraints, and operates on noisy data from onboard sensors.

This article focuses on operational methodologies for identifying dynamic models of industrial robots. It aims to present the main approaches for estimating inertial parameters from recorded measurements of positions and motor torques, within the framework of closed-loop identification. We discuss validity assumptions, excitation trajectory requirements, constraints related to real systems, and the advantages and limitations of the various methods proposed. This contribution is part of a context in which model accuracy is a crucial issue for advanced automation, predictive maintenance, and the integration of collaborative robots in modern industrial environments.