Accurate Defect Detection in Thin-Wall Structures With Transducer Networks via Outlier Elimination

For the sake of structural health monitoring for defect detection in thin-wall structures, we present an outlier analysis approach by using robust principal component analysis (Robust PCA) for the Lamb wave signals generated by a transducer network. The signal matrix is decomposed into a low-rank matrix and a sparse matrix that contains outliers. By presenting a rank approximation function that matches the true rank more closely than does the nuclear norm and comparing a set of non-convex penalty functions to approximate the $\ell _{0}$ -norm, we derive an approximated non-convex Robust PCA (ANC-RPCA) algorithm to identify and even eliminate outliers in the data. This algorithm is able to find efficient defect detection methods under dimension-reduction conditions for high-dimensional measurement signals. The elimination of outliers is also evaluated by introducing the statistical Wasserstein distance, which provides a representation of the statistical distribution of the data. Experiments on wind turbine blade structures validate that the proposed ANC-RPCA is able to detect and eliminate outliers from three different situations that may occur in sensor systems. Therefore, the presented outlier analysis may perform as a promising preprocessing procedure to ensure that the data are in an acceptable statistical state for analysis.