Change detection using the generalized likelihood ratio method to improve the sensitivity of guided wave structural health monitoring systems

Abstract

The transition from one-off ultrasound–based non-destructive testing systems to permanently installed monitoring techniques has the potential to significantly improve the defect detection sensitivity, since frequent measurements can be obtained and tracked with time. However, the measurements must be compensated for changing environmental and operational conditions, such as temperature, and careful analysis of measurements by highly skilled operators quickly becomes unfeasible as a large number of sensors acquiring signals frequently is installed on a plant. Recently, the authors have developed a location-specific temperature compensation method that uses a set of baseline measurements to remove temperature effects from the signals, thus producing “residual” signals on an unchanged structure that are essentially normally distributed with zero-mean and with standard deviation related to instrumentation noise. This enables the application of change detection techniques such as the generalized likelihood ratio method that can process sequences of residual signals searching for changes caused by damage. The defect detection performance offered by the generalized likelihood ratio when applied to guided wave signals adjusted either via the newly developed location-specific temperature compensation method or the widely used optimal baseline selection technique is investigated on a set of simulated measurements based on a set of experimental signals acquired by a permanently installed pipe monitoring system designed to monitor tens of meters of pipe from one location using the torsional, T(0,1), guided wave mode. The results presented here indicate that damage on the order of 0.1% cross section loss can reliably be detected with virtually zero false calls if the assumptions of the study are met, notably the absence of sensor drift with time. This represents a factor of 20–50 improvement over that typically achieved in one-off inspection and makes such monitoring systems very attractive. The method will also be applicable to bulk wave ultrasound signals.

Causal dilated convolutional neural networks for automatic inspection of ultrasonic signals in non-destructive evaluation and structural health monitoring

“…More advanced methods of temperature compensation that combine the stretching procedure with capabilities of phase compensation, such as [6], are expected to yield better alignment throughout, and hence lower residuals after subtraction. Incoherent noise was superposed to the FE simulated measurements to mimic the effects of instrumentation noise, using a procedure similar to that employed in [15], which is described briefly in the following. In typical ultrasonic measurements, frequency components outside a band around the center frequency of the excited signal can be easily filtered out via hardware or software filters, while noise at and around the frequency of excitation is much more problematic.…”

Long-term guided wave structural health monitoring in an uncontrolled environment through long short-term principal component analysis

“…Much of this work has focused on addressing temperature variations. Several methods, such as optimal baseline selection, 8,15 baseline signal stretch (BSS), 1,7,16 physicsbased modeling, 17 and other data-driven models [18][19][20] have been created to compensate for temperature variations. Other researchers have used principal component analysis (PCA), 5 genetic algorithms, 21 ensemble classification, 9 Gaussian mixture models, 22 neural network, 23 and other machine learning approaches to detect damage directly under temperature variations.…”

A Development Strategy for Structural Health Monitoring Applications

“…5(b), then the probabilities of detection and false alarm are a function of the ratio of the defect signal amplitude to the standard deviation of the signal from an undamaged structure; this is discussed in more detail in Ref. [45].…”

“…Although the GLR is rather computationally intensive, it is particularly attractive as it does not require the extent or time of change to be specified in advance. Mariani and Cawley[45] tested it on guided wave pipe monitoring data of the type shown in Figs.5 and 6and showed that it can enable the reliable detection of defects equivalent to as little as 0.1% cross section loss in cases similar to those of Fig.6without a significant number of false calls. This assumes that the sensor is stable with time and that the residual signal in a given structure state remains normally distributed.…”

“…Case of guided wave pipe monitoring system using LSTC temperature compensation with defects corresponding to 0.12% cross section loss (blacklowest solid curve), 0.24% loss (bluemiddle solid curve), and 0.36% loss (redtop solid curve). Data are used to compute Fig.13(a) of Ref [45]…”