For decades, monitoring and identification of the dynamic stress states of rock masses in reservoirs have been challenging tasks owing to the lack of effective observation and analysis methods. In this study, we used deep learning methods to identify whether acoustic emission (AE) signals of rock fractures are created by different loading conditions. We performed Brazilian split and uniaxial compressive experiments on six different engineering materials, and we obtained the AE waveforms. To take advantage of the powerful image processing capabilities of convolutional neural networks (CNNs), we transformed the AE waveforms into time–frequency images. We used five types of CNN to identify the time–frequency images of the AE signals created in the Brazilian split and uniaxial compressive experiments. As a result, we found that Xception model had the highest recognition accuracy. We analyzed the basis of the CNN models to recognize the AE signals using local interpretable model-agnostic explanations and found that the Xception model mainly used the patterns of the low-energy region of the time–frequency images to determine the loading modes of the rock fractures. Furthermore, the high-energy time–frequency region had little effect on recognition. The findings of this work can aid in automatically monitoring the dynamic stress states of fracture areas in reservoir formations and ensure the safety of oil and gas resource exploitation. 

Identifying the Types of Loading Mode for Rock Fracture via Convolutional Neural Networks

The acoustic activity in brittle specimens submitted to mechanical loading is explored in terms of two alternative approaches. The first one is based on the Euclidean three-dimensional distance between the sources of successive acoustic events while the second is based on the power of the acoustic signals recorded. In the direction of quantifying the acoustic activity according to these alternative approaches two functions are introduced, the time evolution of which is considered in juxtaposition to that of the externally applied mechanical load and the respective one of a previously introduced function, namely, the F -function. The latter is determined by taking advantage of the interevent time intervals between successive acoustic signals. Data gathered during an experimental protocol, with notched marble plate-shaped specimens subjected to uniaxial tension, are used to outline the procedure for calculating the newly introduced functions. Analysis of the time evolution of these functions exhibits series of common qualitative characteristics with the respective one of the F -function. Moreover, it is indicated that proper interpretation of the respective data in terms of the “time-to-failure” parameter provides interesting signs that could be considered as pre-failure indices. 

Exploring the acoustic activity in brittle materials in terms of the position of the acoustic sources and the power of the acoustic signals - Part I: Founding the approach

For decades, monitoring and identification of the dynamic stress states of rock masses in reservoirs have been challenging tasks owing to the lack of effective observation and analysis methods. In this study, we used deep learning methods to identify whether acoustic emission (AE) signals of rock fractures are created by different loading conditions. We performed Brazilian split and uniaxial compressive experiments on six different engineering materials, and we obtained the AE waveforms. To take advantage of the powerful image processing capabilities of convolutional neural networks (CNNs), we transformed the AE waveforms into time–frequency images. We used five types of CNN to identify the time–frequency images of the AE signals created in the Brazilian split and uniaxial compressive experiments. As a result, we found that Xception model had the highest recognition accuracy. We analyzed the basis of the CNN models to recognize the AE signals using local interpretable model‐agnostic explanations and found that the Xception model mainly used the patterns of the low‐energy region of the time–frequency images to determine the loading modes of the rock fractures. Furthermore, the high‐energy time–frequency region had little effect on recognition. The findings of this work can aid in automatically monitoring the dynamic stress states of fracture areas in reservoir formations and ensure the safety of oil and gas resource exploitation.

Of late, emerging algorithms such as machine learning have been increasingly used in shield tunneling construction management and control. This research article proposes a shield tunneling parameter matching model based on a support vector machine (SVM) and an improved particle swarm algorithm (PSO) to enhance the accuracy and reliability of shield tunneling parameter selection. First, the optimization performance of the algorithm is augmented by adjusting particle diversity. Simulation-based experiments are conducted to test the performance of the improved algorithm. The experimental results indicate that the improved algorithm has better accuracy. Meanwhile, the particle diversity adjustment strategy is given. Then, the SVM model to express the relationship between shield tunneling parameters and ground settlement is established and trained. Based on the obtained SVM model, the improved PSO is used to optimize the tunneling parameters. The results show that the shield tunneling parameter matching model based on SVM and improved PSO can obtain more accurate shield tunneling parameters and provides a more precise reference for selecting shield tunneling parameters.

/pdf/shield-tunneling-parameters-matching-based-on-support-vector-2w6wad9g.pdf

Shield Tunneling Parameters Matching Based on Support Vector Machine and Improved Particle Swarm Optimization

The acoustic activity in beam-shaped specimens made of cement is studied, assuming that the beams are loaded in three-point bending under a step-wise loading scheme. Attention is focused to the attenuation of the acoustic activity during the constant-load stage of each specific loading step. The experimental data are analyzed in terms of the interevent time intervals between any two successive acoustic hits (using the F-function concept) and, further, in terms of the power of the acoustic hits (in terms of the recently introduced P-function). It is indicated that while the mechanical load is kept constant, the acoustic activity attenuates steadily, and during the early steps of this attenuation phase, the temporal evolution of both the F- and P-functions is excellently described by an exponential law. Moreover, it is proven that for both the F- and P-functions, the relaxation exponents decrease monotonically with increasing load. This decrease becomes quite abrupt for loads exceeding about 80% of the fracture load, providing an interesting and promising pre-failure indicator, i.e., a warning signal that the specimen is entering into the stage of impending macroscopic fracture. The specific conclusions are in very satisfactory agreement, both qualitatively and quantitatively, with similar ones drawn by considering the temporal evolution of the respective b-value.

/pdf/attenuation-of-the-acoustic-activity-in-cement-beams-under-287tqlp5.pdf

Attenuation of the Acoustic Activity in Cement Beams under Constant Bending Load Closely Approaching the Fracture Load

Fracture mechanism of tight sandstone under high and complex 3-D stress compression: Insights from acoustic emission

Complex civil structures require the cooperation of many building materials. However, it is difficult to accurately monitor and evaluate the inner damage states of various material systems. Based on a convolutional neural network (CNN) and the acoustic emission (AE) time-frequency diagram, we used the transfer learning method for classifying the AE signals of different materials under external loads. The results show the CNN model can accurately classify cracks that come from different materials based on AE signals. The recognition accuracy can reach 90% just by re-training the full connection layer of the pre-trained model, and its accuracy can reach 97% after re-training the top 2 convolutional layers of this model. A realization of cracking source identification mainly depends on the differences in mineral particles in materials. This work highlights the great potential for real-time and quantitative monitoring of the health status of composite civil structures.

Xiaodong Wang

Papers

Fracture mechanism of tight sandstone under high and complex 3-D stress compression: Insights from acoustic emission

Classification of cracking sources of different engineering media via machine learning