Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
MODE OBSERVER BASED ON MOVING HORIZON ESTIMATION APPROXIMATION THROUGH NEURAL NETWORKS
Submission Author:
Lara Candido Alvim , RJ
Co-Authors:
Lara Candido Alvim, Leonardo Dias Pereira, Helon Vicente Hultmann Ayala, Elias Dias Rossi Lopes
Presenter: Lara Candido Alvim
doi://10.26678/ABCM.COBEM2023.COB2023-2100
Abstract
The advances in robotics allow a growing range of applications for robotic manipulators, such as the execution of tasks in which the human and robot are sharing the same environment or that require the detection ability of the End-effector for safety or performance issues, which encourages the application of control methods for contact detection. In this paper, for the contact detection estimation, we divide the system states into two groups the free mode when the robot is not in contact and the contact mode when it is. We consider contact as a control mode of the robotic manipulator. Mode detection is achieved based on identifying the active states of the nonlinear switching system. Artificial Neural Networks (ANNs) approach is applied to detect the estimated system states and then the system modes. This method estimates an approximative function that describes a dataset by minimizing the error between predicted and expected outputs. The dataset used to train the ANNs results of the Moving-Horizon State Estimation (MHSE) implementation for contact detection of a robotic manipulator. The main idea behind the MHSE approach is to estimate the system states using past measurements and a moving horizon window to solve at each instant of time a constrained nonlinear optimization problem. The implemented ANNs method can estimate the states and classify the mode effectively, presenting low RMSE values, high values of R² above 0.9, and a reduction in the processing time of the estimation algorithm when compared with the MHSE method.
Keywords
Switching System, Contact Detection, Moving-Horizon Estimation, Artificial Neural Network (ANNs)
