Hosseini, Seyed Hamed Seyed, Hajzargarbashi, Seyedhossein et Liu, Zhaoheng.
2025.
« Vibration analysis of robot manipulators and Gaussian process regression for estimating posture-dependent FRF ».
In Proceedings of the CSME-CFDSC-CSR 2025 International Congress (Montreal, QC, Canada, May 25-28, 2025)
Coll. « Progress in Canadian Mechanical Engineering », vol. 8.
Prévisualisation |
PDF
248 - Vibration analysis of robot manipu.pdf - Version publiée Licence d'utilisation : Tous les droits réservés aux détenteurs du droit d'auteur. Télécharger (1MB) | Prévisualisation |
Résumé
This research focuses on analyzing the vibration behavior of robot manipulators and predicting their frequency response functions (FRFs) based on the robot’s posture. A new experimental method was developed to understand how different joint angles affect vibration modes and to solve challenges such as sensor placement on nodal points, which can lead to inaccurate measurements. Hammer tests were performed at 254 different robot postures along a tool-tip path in the X, Y, and Z directions. The vibration data collected was carefully processed using a four-step peak selection method to identify natural frequencies, damping ratios, and modal stiffnesses with high accuracy.Analysis of the correlation between joint angles and vibration modes showed that joints 2 and 3 have a stronger effect on low-frequency vibrations than joint 1. This is because joints 2 and 3 absorb more energy in these modes, making them critical to the robot’s overall vibration behavior. In addition, parts of the robot that interact with external forces, like the end-effector, can introduce extra vibrations that influence specific modes. The energy absorbed or transferred by each component affects its role in different vibration modes through modal participation and shape. Also, flexible robot arms can face internal resonances between bending and twisting motions, which reduce precision. Predicting these complex vibration patterns is a challenge, and this study addresses it using a reliable machine learning model.A dataset was created containing joint angles and the related modal parameters from the experiments. This dataset was used to train a Gaussian Process Regression (GPR) model, which predicts how the robot will vibrate in different postures. The GPR model was able to predict the natural frequencies, damping ratios, and modal stiffnesses accurately, while also reducing the time and cost compared to traditional simulation methods. The predicted modal parameters were then used to build FRFs for any robot posture, and these matched well with the experimental results.The main contributions of this work are:1. A step-by-step approach for analyzing robot vibrations in different postures.2. New insights into how joint angles affect vibration behavior.3. A data-driven approach using GPR to predict modal parameters and estimate FRFs efficiently and accurately.This method can help improve robot performance in tasks like machining and is useful for vibration analysis in flexible robotic systems used in industry.
| Type de document: | Compte rendu de conférence |
|---|---|
| Éditeurs: | Éditeurs ORCID Hof, Lucas A. NON SPÉCIFIÉ Di Labbio, Giuseppe NON SPÉCIFIÉ Tahan, Antoine NON SPÉCIFIÉ Sanjosé, Marlène NON SPÉCIFIÉ Lalonde, Sébastien NON SPÉCIFIÉ Demarquette, Nicole R. NON SPÉCIFIÉ |
| Professeur: | Professeur Liu, Zhaoheng |
| Affiliation: | Génie mécanique |
| Date de dépôt: | 18 déc. 2025 15:11 |
| Dernière modification: | 18 déc. 2025 15:11 |
| URI: | https://espace2.etsmtl.ca/id/eprint/32415 |
Actions (Authentification requise)
 |
Dernière vérification avant le dépôt |