Section: New Results
Participants : Michael Döhler, Maurice Goursat, Laurent Mevel.
See modules 3.2 , 3.5 , 4.2 , 7.1 , 7.6 and 7.7 .
Improving subspace identification : slight modifications of the algorithm
There are numerous ways to implement the covariance subspace identification method. The variants relate to the choice of parameters, which have to be tuned by the user. Such tuning is only reliably done by the mechanical engineer, when the parameter has a physical meaning. From the data to the results there are 4 main steps in the covariance subspace identification method. In this paper  , we are concerned with the computation of the decomposition of the Hankel matrix and the mean square estimation of the transition matrix. We compare different cases for the algorithms related to these steps: how to choose the parameters and what are the corresponding perturbations on the results for the different basic numerical algorithms. The conclusion is that with some modifications we can easily obtain more precise and reliable results. The idea is also to minimize the number of parameters to be tuned by the user. With a matrix normalization and some approximation scheme we get an implementation being a very efficient and stable identification procedure. All these algorithms are theoretically equivalent, but in practice some give much better results. These new schemes have been validated on different concrete cases (civil structure, aircraft, space launcher). Improvements in damping stability is the main output of this work. Implementation of this work in COSMAD (see module 5.1 ) and for transfer to SNECMA and SVIBS (see modules 7.6 and 7.7 ) has been done in 2009.
Monitoring of aerospace rockets
We revisit the problem of the modal analysis of space launchers. We consider the Ariane 5 launcher with its usual equipment during a commercial flight under the natural unknown excitation. The case of space launchers is a typical example of a complex structure with sub-structures strongly and quickly varying in time. This issue becomes especially important in e.g. estimation of damping of aerospace vehicles. The eigenfrequencies are also sliding during the flight but the modeshapes are more stable. Recently, a new implementation of the subspace identification method has been proposed, leading to cleaner and more stable stabilization diagrams. We monitor the behavior of estimated modal parameters by applying this crystal clear implementation of the data driven and the covariance driven Stochastic Subspace Identification algorithms. We show the importance of crystal clear to monitor successfully frequencies and damping estimates over time in such a non stationary case. This work will be presented in  .