Automatic system identification algorithm for processing ambient vibration data

When large quantities of data are acquired in long-term monitoring works, the use of automatic modal identification procedures is mandatory for the feasibility of real-time data interpretation, damage detection, model updating, or others. This paper presents an innovative algorithm for real-time remote processing the information recorded by ambient vibration tests. This algorithm aims at generating and interpreting the stabilization diagrams resultant from the application of parametric methods (such as the Stochastic Subspace Identification – SSI) to the collected time domain data. The proposed algorithm was validated in two stages: (i) considering numerical examples with artificially generated data and (ii) in a field test for tracking the stiffening process of concrete since early ages. The results of these two rounds of validation tests evidenced the high accuracy of the automatic estimations of this new algorithm and thus, the feasibility for its incorporation as a tool in future Structural Health Monitoring works. 2 IOMAC'11 – 4 International Operational Modal Analysis Conference In this paper a new algorithm is proposed for automatic system identification using parametric methods which mimics the decisions that an experienced modal analyst takes during the modal identification process. A general discussion of the issues related to the automatic system identification, details of the new algorithm as well as the results of the several rounds of validation tests will be further detailed. 2 PARADIGMS OF THE AUTOMATIC SYSTEM IDENTIFICATION As presented by Andersen et al. (2007), three problematic issues are related to the process of modal identification of civil engineering structures. The first one is the excitation source, as variations on the source throughout the experimental tests and variations on the excitation levels are commonly found throughout the experimental tests. The second aspect is the huge quantity of data recorded, encompassing many measurement channels, many setups and/or long sample periods. The third issue is related to the modal identification process itself, as there may be cases where manual extraction of modes might not be possible, or cases where the modes are weakly excited or highly damped. According to Magalhaes et al. (2009), the current research efforts to automate the identification process of structural modal parameters are focused on: a) conception of new identification algorithms using parametric methods in order to obtain clearer stabilization diagrams; b) definition of additional parameters or signal processing techniques that can result in more accurate estimations; and c) development of new methodologies for automatically interpreting the information obtained from the application of the data processing methods. Rainieri and Fabroccino (2009) presented the state of the art of the methodologies related to the automation of the data processing methodologies for Output-Only techniques. According to this, the existing proposals can be classified according to the domain in which they were developed. There are works using frequency domain processes such as the ones presented by Verboven et al. (2001), Verboven et al. (2002), Verboven et al. (2003), Guan et al. (2005), Brincker et al. (2007), Magalhaes et al. (2008), and Rainieri and Fabroccino (2009). On the other hand, there are works focused on time domain processes such as the ones presented by Peeters and De Roeck (2001), Scionti et al. (2003), Andersen et al. (2007), Deraemaeker et al. (2008), and Magalhaes et al. (2009). One issue that must be considered when developing automatic process algorithms is the software environment where they are developed. Currently, the most widespread data processing software is Matlab (Matlab, 2009) which offers fast and powerful tools for this purpose. However, Labview (Labview, 2006) should also be considered, as many of the acquisition processes of dynamic and static experimental tests are carried out using this tool. The use of Labview as data processing software allows combining data acquisition and data processing routines in the same environment which is really attractive for practical purposes. 3 PROPOSAL OF THE AUTOMATIC PROCESSING METHODOLOGY The proposed algorithm aims at automatic interpretation of results of the SSI-Data method (Peeters and De Roeck, 1999) using a combination of the cluster analysis and the rule-based approach for the feature extraction process. As the proposed tool is related to the analysis of stabilization diagrams, it can also be applied for interpreting the results of other parametric processing methods. The algorithm uses a simplified version of the K-means clustering algorithm (Macqueen, 1967) for building initial sets of clusters, taking into account previously defined groups with fixed central values. In the algorithm, the number of clusters of interest n is initially defined and corresponds to the number of frequencies of interest. The objects in each level of the cluster of interest are selected considering the similarity between the estimated frequencies and damping coefficients with respect to the central values. The similarities in the frequency estimations for the iteration level of the model order i and the referred control values CV are calculated according to their Euclidian distance d as stated in Equation 1, while an additional criterion is defined in Equation 2 for damping, namely: