Proc. SPIE. 9803, Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2016
KEYWORDS: Statistical analysis, Detection and tracking algorithms, Sensors, Feature extraction, Signal processing, Structural health monitoring, Damage detection, Signal detection, Data fusion, Data analysis
In this study we propose a regularized linear discriminant analysis approach for damage detection which does
not require an intermediate feature extraction step and therefore more efficient in handling data with high-dimensionality.
A robust discriminant model is obtained by shrinking of the covariance matrix to a diagonal matrix
and thresholding redundant predictors without hurting the predictive power of the model. The shrinking and
threshold parameters of the discriminant function (decision boundary) are estimated to minimize the classification
error. Furthermore, it is shown how the damage classification achieved by the proposed method can be extended
to multiple sensors by following a Bayesian decision-fusion formulation. The detection probability of each sensor
is used as a prior condition to estimate the posterior detection probability of the entire network and the posterior
detection probability is used as a quantitative basis to make the final decision about the damage.