Modal parameter estimation from inconsistent data sets

Most modern modal model estimation algorithms start from the observation that parameters such as resonance frequencies, damping ratios and modal participation factors are "global" for the structure under test. By means of a least squares procedure, a global model is forced on the available time or frequency domain data. In many practical cases however, these data are slightly to strongly inconsistent. Mass loading effects, temperature variations etc. make that the data measured consecutively in different "patches" can show slightly differing resonant frequencies. When trying to fit a global model through these data, errors can result by identifying multiple close poles instead of one single pole near single resonances. But most importantly, the mode shape extraction can be seriously affected as modal residues are extracted for "global" pole values that do not correspond to the actual value of the FRF or impulse response under consideration. This may lead to major problems in postprocessing the data by modal substructuring or modification analysis. Also, the columns of FRF matrices relating to multiple excitation tests are often inconsistent. Shaker or suspension constraints and small nonlinearities are often indicated as possible error sources. The relevance of these problems will be briefly reviewed by some practical case studies and pragmatic remedies are evaluated. This leads to the observation that such methods could also be useful as a smoothing pre-processor for FRF based impedance methods (substructuring, load analysis), where data inconsistencies may lead to major matrix inversion errors.