Outlier analysis for defect detection using sparse sampling in guided wave structural health monitoring

We propose an outlier detection-based statistical approach to identify and locate a defect in composite plates using far fewer number of sensing points compared to conventional imaging techniques. The key steps involved in this computationally inexpensive approach are the random sparse selection of the sensing points through Poisson disk sampling, followed by a two-step outlier detection process based on thresholding and computation of median absolute deviation. The robustness of the proposed technique is explored through extensive simulations involving different defect sizes, random locations on flat plate structures, and various values of signal to noise ratio (SNR). We experimentally demonstrate the feasibility of detection of delamination, whose size is comparable to the ultrasonic wavelength with probability of detection (PoD) better than 90% using <1% of the total number of samples required for conventional imaging, even under conditions wherein the SNR is as low as 5 dB.