EVALUATION OF SUBSPACE IDENTIFICATION TECHNIQUES FOR THE ANALYSIS OF FLIGHT TEST DATA

摘要

Aircraft in-flight test data are typically characterized by short time records due to safety reasons and the fact that the inputs may not be measurable and/or are poorly controllable. In case of quantifiable artificial inputs, the FRFs are contaminated by high environmental noise levels due to the presence of the turbulence. These in-flight data characteristics make that specific data processing techniques such as the Maximum Likelihood Estimator, the ARMA modeling and the Polyreference LSCE method using correlation functions need to be applied to extract the relevant parameters. Over the last years, it has been widely recognized that a new class of techniques, referred to as subspace identification, yields promising results for modal parameter extraction. In this approach, the reduced set of Kalman system states is first derived and then a state space model is identified in a least-squares sense. The class of stochastic subspace techniques allows to identify the modal parameters from output-only data. In this paper, a stochastic subspace method referred to as the Canonical Variate Analysis (CVA) is briefly discussed and subsequently applied to flight flutter test data. Practical aspects such as the model order selection and the distinction between the spurious modes and the physical modes are outlined. The results obtained with the stochastic subspace technique are compared with the estimates of the Polyreference LSCE method fed by correlation functions instead of impulse response functions. The Maximum Likelihood Estimator, a frequency domain input-output modeling technique is used to establish a reference model for the evaluation of the accuracy of the output-only techniques.