Robust defect detection under uncertainties using spatially adaptive capacitive imaging

Abstract

We develop a high-Q capacitive sensing based robust non-destructive evaluation (NDE) methodology that can be widely used in varied NDE applications. We show that the proposed method can detect defects in a host of robust regimes where uncertainties such as lift-off, probe tilt, fluctuations in sampling rates, and step sizes are inherent in the data collection process. We explicitly characterize the corruption in the capacitive sensing data due to various lift-off based uncertainties. We use a Bayesian decision theoretic approach to rigorously understand the impact of these corruptions on defect identification efficacy. Using an optimally tuned weighted classification loss, we prove that it is theoretically feasible to accurately detect defect location and sizes from capacitive sensing signals collected under the aforementioned uncertainties. The Bayesian decision theoretic study needs prior information for accurate detection that is not available in real NDE inspections. So, we develop a solely data driven algorithm that analyzes the capacitive sensing signals without any prior knowledge of defect or uncertainty types. The developed algorithm is non-parametric and uses spatially adaptive denoising to weed out uncertainty induced noises. By leveraging the spatial association in the capacitive sensing signals, our algorithm greatly improves on popular non-spatial approaches. Compared to popular thresholding methods and low-rank based denoising approaches, we demonstrate superior performance of the proposed method in terms of coverage and false positive metrics for defect identification. Using spatially adaptive denoising, we design a robust capacitive sensing method that can detect defects with high precision under various uncertainty regimes.