Structural health monitoring (SHM) is currently a hot topic within the engineering disciplines due to the aging of civil infrastructures in North America and other regions. The scope of the application of SHM is increasingly broad, ranging from civil infrastructures to human health monitoring. In order to implement SHM, many advanced sensors have been developed, from contact-based sensors (such as microelectromechanical sensors) to noncontact sensors (such as air-coupled sensors, vision sensors using cameras, wireless rechargeable sensor networks, and radar sensor networks). With the advancements in these sensors and sensing system technologies, many methods of structural damage detection, localization, and quantification have also been developed. However, there are still a remarkable number of questions associated with both damage detection and the validation of detected damage using traditional sensors for civil infrastructures, mechanical systems, environmental changes, and human health.
One of the critical missions in SHM is discovering how to determine damage-sensitive features and how to extract information about them from the data measured by sensors, in order to detect damage or changes to systems. It is a challenging problem. A number of damage-sensitive features have been suggested, experimented upon, and tested, but there are as yet few successful cases of detecting structural damage from traditional contact sensors in civil infrastructure. Most of the methods, with their specific damage-sensitive features, were able to provide alerts about changes in the monitored systems. However, they were not able to detect specific damages (and their locations in large-scale infrastructure) because of the many associated uncertainties, including environmental changes and changes of boundary conditions. Nondestructive testing-based damage assessment for local damage has been quite successful by using impact-echo-based damage detection or piezoelectric sensors. The main problem with these approaches is that they require numerous sensors to cover the entire civil infrastructure. Even though these sensors can be applied to critical locations on the structures (based on engineering data), the location of the identified damage may be quite random.
To give an example from one of the papers in this special issue, a corrosion detection method using contact sensors was proposed. Structural deterioration due to aging is generally expressed as corrosion in steel structures, so corrosion detection and quantification are critical for evaluating the useful remaining life of structures. Many approaches have been proposed to detect steel corrosion or steel rebar corrosion in reinforced concrete beams, one of which is the use of contact sensors, but that approach uses a self-powered wireless sensor network for automated corrosion prediction. In order to monitor the progress of the corrosion of embedded steel reinforcement, a vibration-based energy harvester was developed for continuous corrosion data collection. Due to the nature of contact sensors, such a system cannot monitor locations where sensors are not installed. To cover the unmonitored areas, a spatial interpolation module was proposed to interpolate measured data from the sensors. The developed method provided for a period of five years an accurate prediction of corrosion by comparing the monitored data.
Corrosion of structures and/or wall thinning in pipes that could significantly reduce their mechanical strength can be detected by another structural health monitoring method, electromechanical impedance (EMI). A reattached EMI method has been developed to reduce the high cost of covering large areas, but, in spite of the efficiency of the reattached EMI method, there remains a problem: The impedance signatures are changed by reattaching the piezoelectric transducer. This issue can be solved by using a probabilistic neural network algorithm. An integrated electromechanical impedance method with probabilistic neural network algorithm was developed to measure the thickness of a metal layer stack. Repeatability with high accuracy can be achieved by integrating the probabilistic neural network algorithm with the electromechanical impedance method and selecting the result most frequently appearing among measurements. When the reattached EMI method is applied to monitoring large structures such as bridges and buildings, the small sensing area it is able to cover causes difficulty and problems related to cost. A promising potential solution to those problems may be the use of a drone with an embedded solar-powered battery.
Among the contact-sensor-based approaches in this special issue, the wireless sensor network (WSN) is widely used in a variety of areas including military industries, agriculture, and infrastructure monitoring. In particular, it is a promising method for effectively monitoring structures with historical significance. Many events and festivals, for which the use of pyrotechnics is popular and frequent, take place near historic buildings. The vibroacoustic impacts of fireworks, which can potentially weaken historic structures, can be monitored by a WSN. Experiments have been conducted to measure sound levels in dB and acceleration for a variety of events, including fireworks, cannon firing, and multiple rifle shots. It has been shown that the vibroacoustic impacts of pyrotechnics cannot be ignored in narrow areas such as alleys, while they are insignificant in a broad, open area. These experiments have been performed with realistic scenarios during actual festivals, resulting in practical findings. The effects of pyrotechnics on historical structures have been monitored and quantified in terms of both acoustics and vibration. The integration of WSN into various historic buildings and other structures will be required to develop a robust and reliable monitoring method for measuring the effects of pyrotechnics.
Many vibration-based damage detection approaches have been developed using traditional contact sensors, such as accelerometers and strain gauges. However, many of these traditional methods require a large number of contact sensors, and managing the number of sensors required to monitor a typical large-scale civil infrastructure is challenging. The postprocessing of the measured data from these sensors is a huge task with tremendous computational costs. Selecting and extracting damage-sensitive features from the measured data is the most critical, yet difficult, aspect of detecting potential future damage. In addition, confirming whether the collected data actually indicates structural damage or results from sensory system malfunction, noisy signals, or a combination of these things typically requires that the sensing systems and structures be checked in person. Thus, regular visual inspection by trained engineers is still a common method of SHM. However, human-based visual inspection is expensive, which leads to low inspection frequency and is restricted by the inaccessibility of areas such as the underside of bridge systems.
In order to overcome the drawbacks of traditional contact-sensor-based approaches, computer-vision-based methods have been proposed that use image processing techniques (IPTs) to detect damage and partially replace visual inspection. As an example from this special issue, advanced digital image correlation techniques were used for years to monitor erosion and sedimentation occurring repeatedly on South Korea’s east coast. A three-dimensional (3D) hydraulic model was developed to investigate the effects of waves induced by the current to transport sediment. Using very sensitive high-resolution video cameras, images were obtained to upgrade the existing dot-unit measuring method to a plane-unit measuring method. From this study, the generation routes of longshore currents and strong rip currents were calculated and the flow direction and flow velocity measured by the plane-unit method in field observations showed very similar tendencies to those of 3D hydraulic model tests.
Another example is the phase-sensitive optical time-domain reflectometer (Φ-OTDR), which is widely used in health monitoring to extract a distributed vibration signal in order to enhance the accuracy and efficiency of the monitoring. For two-dimensional (2D) use of Φ-OTDR, a time-encoded signal processing (TESP) algorithm can be used to effectively reduce redundancy and increase effectiveness, even for fiber points with poor signal-to-noise ratio (SNR). Moreover, in the case of three-dimensional data in Φ-OTDR, two methods—empirical mode decomposition (EMD) and nonnegative matrix factorization (NMF)—can be combined and optimized by the genetic algorithm (GA) method. Better performance on the length and time dimensions is shown in a Φ-OTDR signal represented by the TESP method and the combined EMD and NMF method optimized by the GA method (GAEMD-NMF method). The accuracy of the TESP method and the similarity between the GAEMD-NMF results and the sensor signal in frequency domain are both enhanced. Both 2D and 3D Φ-OTDR vibrant signals can be represented with high accuracy and improved similarity. The TESP method is used for 2D signals, while the GAEMD-NMF method is used for 3D signals. The effectiveness and practicability of the TESP and GAEMD-NMF methods are proved by experimental tests in a controlled anechoic chamber. A broad range of experimental tests is required for further validating the TESP and GAEMD-NMF methods.
SHM approaches are currently being used to monitor human health, including neck pain, one of the more common musculoskeletal disorders (MSDs). In order to understand the primary cause of an instance of neck pain, the head flexion posture during walking is measured by an inertial sensor (using information on gravitational forces) attached to the neck. The craniovertebral angle (CVA) is measured using an OptiTrack camera system with experimental verification. This study investigated the relationship between CVA and neck flexion angle (NFA) in both static and dynamic cases. The results indicated that NFA has a close relationship to CVA. Although the authors showed the good performance of the proposed method, there is clearly a need for more extensive experimental studies and a better algorithm for estimating a variety of angles.
In another study, physiological parameters were monitored using an electrocardiogram (ECG) to diagnose chronic cardiovascular diseases. This long-term monitoring requires wearable electrodes that are breathable, flexible, biocompatible, and friendly to the skin. Four conductive weave electrodes were developed for an extensive study of their performance as a sensing system. From these experiments, it was learned that conductive fabric for use on human skin must have lower skin-electrode impedance and greater comfort.
The technology of SHM is growing, and new technologies are introduced every year, even though many questions remain to be answered. In this special issue, we found potential solutions to address practical issues ranging from how to apply SHM methods to civil infrastructures and human bodies to improving on traditional contact-sensor-based approaches by using noncontact vision sensors. However, these vision-based approaches also require damage-sensitive features in order to pinpoint damage, and each approach can theoretically detect only a specific type of damage. Consequently, deep learning-based damage detection methods have recently been developed using a convolutional neural network [1, 2], although that development is not addressed in this special issue. The main advantage of deep learning-based approaches is that they automatically extract damage-sensitive features for multiple kinds of damage. Within a couple of years, we expect to see many new approaches and methods using these deep learning algorithms with noncontact vision sensors and autonomous, unmanned aerial vehicles to expand the scope of SHM.
Young-Jin Cha
Yeesock Kim
Taesun You