Structural damage diagnosis based on stochastic subspace identification, Kalman model, and principal component analysis

This paper deals with the application of statistical process control techniques for damage diagnosis based on vibration measurements. The first approach considered in this work is based the Stochastic Subspace Identification (SSI) algorithm, from which a Kalman model is constructed to fit the measured response histories of the undamaged (reference) structure. This model may be used to make a prediction of the newly measured responses. The residual error between the model predictions and the actual measurements is defined as a damage-sensitive feature. Outlier statistics provides then a quantitative indicator of damage. The advantage of the method is that model extraction has to be performed only once using the reference data and that no further modal identification is needed. Thus on-line structural health monitoring may easily be realized. In the second approach, principal component analysis (PCA) of the sensor time-responses is used to extract principal directions (i.e. features), which define a subspace that is representative of the dynamics of the instrumented structure. Any change in the response of a single sensor affects the subspace spanned by the complete sensor response set. It follows that the subspace corresponding to the current state of the structure can be compared to the subspace of the initial state of the structure, assumed to be healthy, in order to diagnose possible damage. Principal component analysis may also be performed for every potential subset of damaged sensors in order to identify the involved sensor, and, therefore, the damaged substructure. © AFM, EDP Sciences 2007.