Yujie Ying

Toward Data-Driven Structural Health Monitoring: Application of Machine Learning and Signal Processing to Damage Detection

AbstractA multilayer data-driven framework for robust structural health monitoring based on a comprehensive application of machine learning and signal processing techniques is introduced. This paper focuses on demonstrating the effectiveness of the framework for damage detection in a steel pipe under environmental and operational variations. The pipe was instrumented with piezoelectric wafers that can generate and sense ultrasonic waves. Damage was simulated physically by a mass scatterer grease-coupled to the surface of the pipe. Benign variations included variable internal air pressure and ambient temperature over time. Ultrasonic measurements were taken on three different days with the scatterer placed at different locations on the pipe. The wave patterns are complex and difficult to interpret, and it is even more difficult to differentiate the changes produced by the scatterer from the changes produced by benign variations. The sensed data were characterized by 365 features extracted from a variety of...

Application of Mellin transform features for robust ultrasonic guided wave structural health monitoring

Guided wave based structural health monitoring systems are sensitive to environmental and operational conditions. This leads to false-positive results for most conventional detection methods. In this paper, we investigate the capabilities of the Mellin transform for detecting damage under variable environmental conditions. The Mellin transform is chosen due to its invariance to scaling operations and robustness to wave velocity. From experimental results, we demonstrate that the Mellin transform features can detect a mass on a steel pipe under variable internal pressure with an overall average accuracy of 94.00% while equivalent Fourier transform features detect the mass with only a 67.00% accuracy.

Ultrasonic monitoring of a pipe under operating conditions

The paper presents experimental results of applying an ultrasonic monitoring system to a real-world operating hot-water supply system. The purpose of these experiments is to investigate the feasibility of continuous ultrasonic damage detection on pipes with permanently mounted piezoelectric transducers under environmental and operational variations. Ultrasonic guided wave is shown to be an efficient damage detector in laboratory experiments. However, environmental and operational variations produce dramatic changes in those signals, and therefore a useful signal processing approach must distinguish change caused by a scatterer from change caused by ongoing variations. We study pressurized pipe segments (10-in diameter) in a working hot-water supply system that experiences ongoing variations in pressure, temperature, and flow rate; the system is located in an environment that is mechanically and electrically noisy. We conduct pitch-catch tests, with a duration of 10 ms, between transducers located roughly 12 diameters apart. We applied different signal processing techniques to the collected data in order to investigate the ongoing environmental and operational variations and the stationarity of the signal. We present our analysis of these signals and preliminary detection results.

Time reversal for damage detection in pipes

Monitoring the structural integrity of vast natural gas pipeline networks requires continuous and economical inspection technology. Current approaches for inspecting buried pipelines require periodic excavation of sections of pipe to assess only a couple of hundred meters at a time. These inspection systems for pipelines are temporary and expensive. We propose to use guided-wave ultrasonics with Time Reversal techniques to develop an active sensing and continuous monitoring system. Pipe environments are complex due to the presence of multiple modes and high dispersion. These are treated as adverse effects by most conventional ultrasonic techniques. However, Time Reversal takes advantage of the multi-modal and dispersive behaviors to improve the spatial and temporal wave focusing. In this paper, Time Reversal process is mathematically described and experimentally demonstrated through six laboratory experiments, providing comprehensive and promising results on guided wave focusing in a pipe with/without welded joint, with/without internal pressure, and detection of three defects: lateral, longitudinal and corrosion-like. The experimental results show that Time Reversal can effectively compensate for multiple modes and dispersion in pipes, resulting in an enhanced signal-to-noise ratio and effective damage detection ability. As a consequence, Time Reversal shows benefits in long-distance and lowpower pipeline monitoring, as well as potential for applications in other infrastructures.

Damage Detection in Pipes under Changing Environmental Conditions Using Embedded Piezoelectric Transducers and Pattern Recognition Techniques

This paper presents the preliminary results of a research project that investigates the feasibility of continuous monitoring techniques using piezoelectric transducers (PZTs) permanently installed on steel pipes. The ultrasonic waves generated by PZTs are multimodal and dispersive. Therefore, it is difficult to detect changes created by the presence of damage, and it is even more difficult to differentiate changes produced by damage from benign changes produced by variation in environmental and operational conditions. In this paper, the results are reported of applying pattern recognition techniques to detect a mass scatterer (a proxy for damage) under ambient variations primarily due to varying internal pressure of a pipe. Using wavelet methods, 303 features are extracted, and adaptive boosting, modified adaptive boosting, and support vector machines for damage detection are employed. The performances of the three classifiers are evaluated over 41 trials with different combinations of training and testing data, resulting in the average accuracies of 85, 89, and 94%, respectively. Finally, the effectiveness of wavelet processing and features selected are discussed.

Time reversal data communications on pipes using guided elastic waves: Part II. Experimental studies

Embedded sensors in large civil structures for structural health monitoring applications require data communication capabilities between sensor nodes. Conventional communication modalities include electromagnetic waves or acoustical waves. However, ultrasonic guided elastic waves that can propagate on solid structures such as pipes for a great distance have rarely been studied for data communication purposes. The multi-modal and dispersive characteristics of guided waves make it difficult to interpret the channel responses and to transfer useful information along pipes. Time reversal is an adaptive transmission method that can improve the spatial and temporal wave focusing. Based on the focusing effect of time reversal, we have developed a data communication technique using guided waves in a highly dispersive pipe environment. In this paper, we experimentally demonstrate the data communication using time reversal pulse position modulation (TR-PPM). Three-step laboratory tests have been performed using piezoelectric transducers in a pitch-catch mode. We first measure the channel responses between the transmitter and the receiver on a pipe. We then carry out the time reversal transmission by reversing the sounding signal and feeding it back to the same channel. Finally, we perform the time reversal communication experiment by sending the modulated time reversal signals as a stream of binary bits at a given data rate. A series of experiments are conducted on steel pipes. Experimental results demonstrate that time reversal pulse position modulation for data communications can be achieved successfully using guided elastic waves.

Applications of Machine Learning in Pipeline Monitoring

In the field of structural health monitoring, researchers focus on the design of systems and techniques capable of detecting damage in structures. However, most traditional detection methods fail under environmental and operational variations that tend to distort the signals and masquerade as damage. In this paper, we investigate the applications of machine learning techniques to developing a damage detection system robust to changes in the internal air pressure of a pipe. From each of the 240 experimental datasets, we extract 167 features and implement three classification algorithms for detecting damage: adaptive boosting, support vector machines, and a method combining the two. The performances of the three classifiers are evaluated over 30 detection trials with different combinations of training and testing data, resulting in the average accuracies of 87.7%, 92.5% and 93.5%, respectively. The combined method is a promising classifier for damage detection. Through feature selection, we also demonstrate the effectiveness of features related to the curve length, the shift-invariant correlation coefficient and the peak amplitude of the signal.

Time reversal data communications on pipes using guided elastic waves: Part I. Basic principles

Piezoelectric sensors that are embedded in large structures and are inter-connected as a sensor network can provide critical information regarding the integrity of the structures being monitored. A viable data communication scheme for sensor networks is needed to ensure effective transmission of messages regarding the structural heath conditions from sensor nodes to the central processing unit. In this paper we develop a time reversal based data communication scheme that utilizes guided elastic waves for structural health monitoring applications. Unlike conventional data communication technologies that use electromagnetic radio waves or acoustical waves, the proposed method utilize elastic waves as message carriers and steel pipes as transmission channels. However, the multi-modal and dispersive characteristics of guided waves make it difficult to interpret the channel responses or to transfer correctly the structural information data along pipes. In this paper, we present the basic principles of the proposed time reversal based pulse position modulation and demonstrate by simulation that this method can effectively overcome channel dispersion, achieve synchronization, and delivery information bits through steels pipes or pipe-like structures correctly.

Ultrasonic Monitoring of a Pressurized Pipe in Operation

Pipes carrying pressurized fluids are an important part of the civil infrastructure, and structural health monitoring (SHM) could ensure structural integrity by predicting and preventing structural failures. Guided wave ultrasonics is a good candidate for use in pipe SHM because guided waves can propagate long distances and are sensitive to structural damage such as cracks and corrosion losses. However, the multi-modal and dispersive characteristics of guided waves make it difficult to interpret their arrival records. Moreover, guided waves are also sensitive to environmental and operational variations, limiting the effectiveness of ultrasonic methods to detect pipe damage in a real environment. We introduce a damage detector based on singular value decomposition (SVD) that can identify a change of interest, caused by a mass scatterer that simulates subtle damage, under realistic environmental variations. We show the effectiveness and robustness of this method on experimental data collected on a pipe segment under realistic environmental and operational variations over a time period of several months.

Robust change detection in highly dynamic guided wave signals with singular value decomposition

Ultrasonic guided waves are sensitive to small scatterers and can, in principle, be used to detect damage in pipe structures. However, pipes are often subjected to varying environmental and operational conditions (EOC), which can produce false positives or mask the change of interest. We apply singular value decomposition as a robust change detection method in ultrasonic signals. We test the methods on experimental data collected in a realistic highly dynamic environment, and show successful detection of a mass scatterer as a physical simulation of damage. We also compare our method to two other change detection methods that are robust to EOC.