Monitoring rock damage subjected to cracks is an important stage in underground spaces such as radioactive waste disposal repository, civil tunnel, and mining industries. Acoustic emission (AE) technique is one of the methods for monitoring rock damage and has been used by many researchers. To increase the accuracy of the evaluation and prediction of rock damage, it is required to consider various AE parameters, but this work is a difficult problem due to the complexity of the relationship between several AE parameters and rock damage. The purpose of this study is to propose a machine learning (ML)-based prediction model of the quantitative rock damage taking into account of combined features between several AE parameters. To achieve the goal, 10 granite samples from KAERI (Korea Atomic Energy Research Institute) in Daejeon were prepared, and a uniaxial compression test was conducted. To construct a model, random forest (RF) was employed and compared with support vector regression (SVR). The result showed that the generalization performance of RF is higher than that of SVR_RBF. The R², RMSE, and MAPE of the RF for testing data are 0.989, 0.032, and 0.014, respectively, which are acceptable results for application in laboratory scale. As a complementary work, parameter analysis was conducted by means of the Shapley additive explanations (SHAP) for model interpretability. It was confirmed that the cumulative absolute energy and initiation frequency were selected as the main parameter in both high and low-level degrees of the damage. This study suggests the possibility of extension to in-situ application, as subsequent research. Additionally, it provides information that the RF algorithm is a suitable technique and which parameters should be considered for predicting the degree of damage. In future work, we will extend the research to the engineering scale and consider the attenuation characteristics of rocks for practical application.

Acoustic emission (AE) has garnered significant interest as a promising way to detect the early-stage development of internal cracks and damage in underground and geotechnical structures, associated with natural disasters. Meanwhile, AE source localization techniques that can identify the damage location in a piled-raft foundation (PRF) are premature because of its complex geometry, although the PRF is a widely used deep foundation type for high-rise buildings. In this study, we propose an integrated approach to localize AE sources in the PRF by using the modified Akaike information criterion (AIC) method and examine its accuracy to mark with pile zones. We performed a series of experiments on a scaled PRF model at a ratio of 1:50, composed of one raft and 25 piles. The results demonstrate that the combined approach with the modified AIC method and the Simplex method can localize the AE source zones with good accuracy, greater than 95% on average. The suggested two-stage AIC picker shows accurate onset time determination, and hence, it significantly improves the accuracy, particularly effective for the signals with low signal-to-noise ratios. The approach exploiting the two-stage AIC picker can be readily used for automated real-time AE monitoring to detect crack generation and its location in buried foundations that cannot be inspected visually.

This paper presents a new Acoustic Emission (AE) source location method in combination with a Wigner-Ville Distribution (WVD) and a theoretical velocity dispersion model in dispersive rock. This method does not require measuring the first arrival time of a signal, and is able to provide reliable and identical Time-of-Arrival-Differences (TOADs) regardless of the frequency components in the obtained signals. The basic idea behind this is that, if at least two dominant frequency components in the output signals can be determined from the WVD analysis of the test results, a full wave dispersion curve of the arriving signal is obtained in accordance with the theoretical wave model, where the time differences between dispersion curves are identical through the entire effective frequency range of the AE transducers. Biot’s model was applied as a velocity dispersion model, and the smoothed pseudo WVD was used to reduce the cross-term interference. It was found that the frequency-dependent arrival time traveling with the group velocity was easily determined with the help of the theoretical model, and the source location results based on the WVD algorithm and theoretical model agree better with the true locations than those from the Time-of-First-Arrival Differences (TOFADs).

Various sensors are used for structural health monitoring. An acoustic emission sensor detects an elastic wave propagating in the medium, so it can detect the possibility of defects occurring inside the structure. Using multiple sensors enables the estimation of the signal source through the differences in signals measured by each sensor. Among the information for signal analysis, the time difference of arrival is the most commonly used factor for estimating the location of the source. However, it is difficult to accurately determine the arrival time because the measured signal always contains ambient noise. Even though the arrival times of signals are determined, there is the following task to identify the source location, which is also complicated because the signal does not propagate with a constant velocity throughout the medium. To solve this problem, this study adopts a hybrid approach that applies artificial intelligence techniques step by step. In the first phase, the time-series data are classified as signal and non-signal by the long short-term memory network. The second phase is to identify the source location based on the naive Bayes classifier using the distribution of the arrival times of signals extracted from multiple sensors. Since this approach reduces complex computations in signal processing while minimizing the masking of physical meaning by black-box AI technology, it allows for versatile applications depending on the objectives. The proposed method was validated through an experimental test, and the results showed that the method had reliable performance.