1. INTRODUCTION Operational Modal Analysis is a method for the identification of modal parameters, in monitoring structures which operate under various working conditions with a non-linear behavior. Thus, it is essential to monitor changes in the modal properties of the structures over time, in order to simulate the non-stationary vibrations with time-dependent properties. Vibrational behavior may be analyzed through non-parametric and time-frequency methods [1] such as the Short-Time Fourier Transform (STFT). In this paper, we present the first steps of an ongoing research on the vibration analysis of the flexible robot manipulator SCOMPI using the STAR software [2] for an automatically monitoring of the online variations of modal parameters in time. This method is based on the linear auto-regressive model, which is sequentially computed in a short-time scheme by using a sliding window [3]. The multivariate autoregressive model is used, and the modal parameters are estimated by least squares using the QR factorization. Rather than varying the parameters on a sample by sample basis, it uses a sliding window [4], and assumes the parameters constant inside each window. As a result, the parameters are adjusted window by window. The proposed method may be considered as the time domain counterpart of the Short-Time Fourier Transform method, and can be used with multi-channel measurements. This method has shown its efficiency and effectiveness in tracking the modal parameters and monitoring their evolution on both linear and non-linear structures, such as robotic machining which is the subject of our investigation. The method was experimentally applied on the real multi-channel data measured on the robot SCOMPI during grinding. We will show through this paper that the STAR method outperforms in terms of frequency identification and that it reflects adequately and accurately the changes in the system. Online modal analysis of a flexible robot during grinding Mokdad, Fatma École de Technologie Supérieure, Montréal (Qc), H3C 1K3, Canada Vu, V. Hung; Thomas, Marc; Rafieian Farzad; Liu, Zhaoheng École de Technologie Supérieure, Montréal (Qc), H3C 1K3, Canada marc.thomas@etsmtl.ca Hazel, Bruce Hydro-Québec’s research institute, Varennes, Qc, J3X 1S1, Canada ABSTRACT: Portable robotic systems are nowadays used as an effective and profitable solution for the automation of maintenance tasks on large hydroelectric equipments. Most utilities perform in-situ inspections and interventions to address issues such as cavitation and cracking. These repair tasks involve automated processes such as robotic welding and grinding. To improve the precision and the quality of grinding, it is necessary to reduce the vibrations caused by the flexibility of the portable robot used for carrying out these operations. In this paper, a method is developed for the operational modal analysis of vibrating structures whose properties may vary with time; it is based on auto-regressive model in a short-time scheme with a sliding window. This new method allows for the successful modal modeling and identification of output-only non-stationary systems, and allows for the tracking of modal parameter variations in time. The technique is experimentally applied on a flexible robot during grinding. The vibratory analysis for the robot working in various conditions is thus presented. The determination of modal parameters allows us to predict the dynamic model and to simulate the vibrations during the operations of grinding and consequently improves the quality of ground surfaces.