IDENTIFICATION OF MODAL PARAMETERS FROM INCONSISTENT DATA Bart Cauberghe, Patrick Guillaume, Benoit Dierckx* and Peter Verboven Vrije Universiteit Brussel (VUB) Department of Mechanical Engineering (WERK) Pleinlaan 2, B-1050 Brussels Belgium *LMS International Interleuvenlaan 68, 3001 Leuven, bart.cauberghe@vub.ac.be ABSTRACT Most of the recent modal parameter estimators start from the fact that natural frequencies, damping ratios and modal participation factors are global parameters of the measured structure. Measuring the structure in different patches can however cause inconsistencies in the data due to e.g. the mass-loading effect of the accelerometers. When trying to fit a global model through these inconsistent data, errors can result by identifying multiple close-coupled poles. By means of simulated and experimental data, a strategy is proposed to solve this problem without increasing measurement time. First, each patch is processed individually. Next, the results are compared over the different patches and missing poles are re-estimated using adapted identification techniques. After compensating for the mass-loading effect, the obtained parameters are combined using clustering techniques and merged into one final global model. The proposed method can still be applied when the inconsistencies are not only due to mass-loading. The strategy has been tested on different modeling problems. Based on these results, an optimal measuring strategy has been developed to facilitate the automatic processing of the data. This optimal measuring strategy minimizes the occurrence of problems during the processing, and, consequently, optimizes the final quality of the estimated modal model. 1. INTRODUCTION When the number of available accelerometers or acquisition channels is smaller than the interested number of degrees of freedom, the structure is measured in different patches. Each patch consists out of one configuration of the accelerometers. These accelero- meters have a mass-loading influence on the structure. Because the accelerometers are placed in different locations for each patch the dynamic characteristics of the measured structure are also slightly differently from patch to patch. The data of the different patches are therefore inconsistent to each other. These mass-loading inconsistencies can be avoided by using laser measurements, but these systems are expensive and more difficult in use to measure three-dimensional structures with. 2. PROBLEM DESCRIPTION Traditionally all measurements of all the patches are taken together and the natural frequencies are considered as global parameters. In case of inconsistencies between different patches this is not applicable anymore. Because the natural frequencies are slightly different from patch to patch, multiple close- coupled poles are identified, if a high model order is used, instead of a single pole. The corresponding physical mode shape is scattered over these different close-coupled poles. In the best case the sum of these close-couples modes show some comparison with the physical mode. On the other hand if the model order is not very high, these closed coupled poles are not detected. Only the pole corresponding with the patch in which the mode is the best represented is estimated. Because this pole does not coincide with the same physical pole of a particular patch, the mode shape extraction is performed at the wrong frequency [1]. As a result the mode shape has no physical meaning. Different simulations and experiments show these phenomena [8]. To illustrate this a door of a car was measured perpendicular to the surface of a door in 79 DOF as shown in Figure 1. Each patch consists of 7 FRFs including the driving point FRF. Six accelerometers, each weighting 5.5 g, were moved from patch to patch. In total 13 patches were measured in a frequency band from 20 Hz to 200 Hz. The door was also measured with a 809