Optimal sensor placement for damage identification - An efficient forward-backward selection algorithm

Structural damage causes characteristic local changes in stiffness, damping and/or mass. Corresponding to these changes, shifts of the dynamic characteristics like eigenfrequencies, damping and mode shapes occur. The deviations between the actual properties and the undamaged state can be used to diagnose the locations and the extents of damage. This contribution deals with the increase of the information content of measurement data. This can be achieved by maximizing the determinant of the Fisher-information matrix which is based on the eigenvector sensitivities of a numerical model. After the formulation of the optimization problem, a forward-backward selection algorithm for sensor placement is presented to find out optimized positions for a certain number of sensors. Moreover an information-entropy concept is introduced that can be used to compare sensor-setups with a different number of sensors and mode shapes. The diagnosis of damage is carried out using an optimization procedure. It is shown with simulated and real measurement data that the proposed sensor placement algorithm leads to a higher amount of information and to a more successful diagnosis of damage.