16.1.1 State of the practice

The advances made by the structural health monitoring (SHM) community over the past two decades will only benefit society if the value of the information provided by these systems is able to offset the costs of installation and operation of the SHM systems themselves. Presently, monitoring systems are considered cost-effective for only a select set of scenarios:¹

The common attribute among all of these motivations for monitoring is that the structure’s owner or other major stakeholders can see a low-risk, short-term return-on-investment for the monitoring system. That return can be realized as reduced liability, more cost-effective maintenance decisions, or reduced construction costs.⁹

For example, the 2011 Los Angles Amendment Building Code §91.1613.10¹⁰ requires the installation of a limited number of accelerometers to measure strong ground motions and structural responses in areas of high seismicity. The information collected by these relatively inexpensive systems has been used for post-disaster condition assessment and improvement of seismicity models.¹¹ In one case, inspection of accelerometer data showed that damage had occurred to a building during an earthquake. However, the location of the damage could not be identified and manual inspections of the structural connections were still necessary, costing a few hundred to a few thousand dollars for each connections.¹² A reduction in per-sensor installation costs, and the introduction of innetwork data processing, both made possible with a WSN, may bring about automated condition assessment with higher spatial resolution, thereby reducing the need for costly manual inspections.

Moss and Matthews¹ categorize the sensors used for SHM into three groups: those sensors that measure the application of loads to the structure (e.g., wind pressure sensors, anemometers, thermometers, and load cells); those that measure structural responses to applied load (e.g., linear variable differential transformers (LVDT), inclinometers, fiber optics, accelerometers, and strain gages); and those that directly measure structural degradation such as scour and corrosion. Transducers in these categories can vary greatly in price, from less than a dollar to over ten thousand dollars per sensor. The quintessential trade-off with cost is performance; a more expensive device will generally have a lower noise floor, better resistance to extreme environments, higher sensitivity, and a broader frequency response. Variations in these performance measures all affect the capabilities of the SHM system as a whole. The actual cost of monitoring systems has been noted to be up to $4000 per channel for strong ground motion accelerometers in buildings.⁶ At these prices, the broad application of SHM systems is in doubt, especially when higher sensor densities are required for improved spatial resolution of damage detection. Clearly, lower cost sensors are direly needed.

Toward achieving this goal, academia and industry alike have focused effort on reducing the cost of SHM by exploiting recent wireless communication breakthroughs. The main challenges associated with the design and deployment of a successful structural monitoring program are cost-effective data transmission, data management, and system identification (ID).⁹ Each of these challenges is lessened when switching to a wireless monitoring system, but new surmountable challenges arise (e.g., supplying wireless nodes with power and maintaining data quality on par with wired systems¹³). The development of a cost-effective, high-performance SHM architecture with wireless telemetry has applicability beyond civil structures (e.g., aircraft¹⁴ and naval vessels^15,16).

16.1.2 State of the art

The idea of structural monitoring using wireless data transmission was first proposed by Straser and Kiremidgjian¹⁷ in the mid 1990s as a method to reduce the costs associated with using cables. Since then, researchers have developed a wide variety of wireless sensor prototypes aimed at reducing power requirements, extending communication ranges, and improving the efficient use of available bandwidth. Lynch and Loh¹⁸ present an extensive summary review of the developments in wireless monitoring for SHM. As the research has developed over the years, it became apparent that switching to wireless telemetry would achieve more than a cost reduction associated with the absence of extensive cabling. The embedded computing inherent in each wireless node can be used to process data in a decentralized manner, thereby reducing the computational load on the centralized server and simultaneously improving system scalability to larger nodal counts. With improved scalability comes an increased sensor density, leading to more accurate system ID and improved damage detection capabilities.

WSN for SHM can come in many flavors. Most of the research and methods presented herein consider wireless SHM as the practice of replacing wired communication between transducers and a central data aggregator with a wireless link. The wireless links most often use the industrial, scientific, and medical (ISM) radio frequency (RF) spectrum, as defined by the International Telecommunications Union (ITU) and adopted by the US Federal Communications Commission (FCC). The key benefit of using this spectrum is the ability to transmit without a license, provided local regulations are met. Within the ISM spectrum, many WSNs operate around 2.4 GHz due to high information throughput; however, 900 MHz has been used when an increase in range was required.¹⁸ Cellular data networks, i.e. the networks typically used for mobile phones, have also been used in SHM to transmit data from the small on-site data repository to an offsite server managed by the SHM system operator.¹⁹ Brownjohn et al.²⁰ pioneered a unique form of SHM that eliminates the cabling of traditional systems, but does not transmit any response data wirelessly. Instead, a global positioning system (GPS) signal is used to synchronize the clocks of independent data acquisition systems, from which the data are retrieved periodically by manual means.

Beyond moving to a new type of telemetry, SHM using WSN has introduced other exciting opportunities, such as ad hoc communication, mobile sensors, and in-network computing, all of which increase the viability of SHM. Ad hoc communication allows the communication network to self-heal in case a wireless link fails. Additionally, sensors can be upgraded, augmented, or replaced without having to significantly restructure the network.²¹ In all wired SHM systems, the sensors are (semi-) permanently installed and cannot autonomously move to return better SHM results. Taking advantage of the mobile freedom associated with wireless telemetry, Zhu et al. proposed and developed an autonomous wireless robot that augments the static WSN on a bridge with mobile excitation and response data.^22,23 With this technology, the special resolution of the transmissibility functions used for system ID is improved without significantly increasing the number of nodes in the network. Although wireless networks have limited communication bandwidth, compared to those that are wired, each wireless node has embedded computing (e.g., a microcontroller (MCU)) that can preprocess the data, thereby reducing the amounts of data that need to be transmitted. Not only does this in-network data processing alleviate bandwidth limitations, it can also reduce data inundation of the site’s main server. Traditional system ID algorithms, such as the eigen-system realization algorithm (ERA), stochastic subspace identification (SSI), and frequency domain decomposition (FDD), have been successfully embedded as applications in a WSN, and can be used to lessen the computational load on the server. Zimmerman et al.^24–26 have embedded simulated-annealing algorithms for identifying model and state-space structural parameters using only the network’s distributed computing capability, and utilized a market-based approach for dynamic computational load balancing across the network. The key to these bandwidth optimizations is the transformation, at the wireless node, of raw time-history data to a sparser domain (e.g., the frequency domain) that is then transmitted wirelessly.

With these aforementioned advantages over wired SHM, it is no wonder that researchers and commercial interests have begun successfully deploying WSN. The large majority of the deployments have been on bridge structures, because of their accountable and transparent governmental owners. In 2005, Pakzad et al.²⁷ measured acceleration on the Golden Gate Bridge in California using 64 MicaZ wireless sensors. A multi-hop pipelining approach was used to wirelessly deliver the data from one end of the bridge to the other, a distance that a single wireless transmission could not cover. A large-scale long-term deployment of a WSN with power harvesting was installed on the Jindo Bridge in South Korea.²⁸ Similarly, a long-term deployment of a WSN on the New Carquinez Bridge in California was integrated with advanced data management tools implemented upon an extensive cyber-infrastructure to aid the operators in analyzing the extensive amount of data collected.¹⁹ A more complete review of successful wireless SHM installations can be found in Chapter 17.

16.3.1 Anatomy of a wireless sensor

Modern wireless sensors (Plate XIVe through g) convert a physical measure into a voltage that is then converted to a digital value, processed, and wirelessly transmitted. The sensing transducer converts the physical measure into an electrical signal that is then typically passed through a signal conditioner and into an ADC. In the node’s core (Plate XIVe), the MCU, essentially an extremely small computer, processes measurements, temporarily stores data in external memory, and then packetizes the processed data into a packet, with a destination address and other information, for transmission by a wireless transceiver. The wireless transceiver modulates the packetized data onto a carrier radio wave and transmits it over an antenna to the receiving unit or server. Optionally, wireless nodes may contain a digital-to-analog converter (DAC) that can then be used to excite the structure for active input–output system ID or for structural control.⁴⁰ The whole system must be powered by either a wired power source or rechargeable batteries that can be charged using power harvesting techniques.

Low-cost, low-power MCUs have been one of the key enabling technologies for WSN. The MCU, and its embedded firmware (Plate XIVf), handles the node’s computation load but also contains a combination of other features including:

The selection of a MCU for a wireless sensor node is a complex task that depends upon the wireless sensor’s application. When choosing a commercial wireless sensor, it is sometimes not possible to know which MCU is used, but the node’s datasheet will typically contain abstracted specifications (e.g., speed, memory, and power consumption). Therefore, it is important for the designer of a WSN to understand the repercussion of the MCU specifications on overall WSN performance. Since wireless sensors typically run off of batteries, it is important to choose a low-power MCU with sleep modes yielding lower power consumption when a peripheral or computation is not necessary. The most common supply voltages for MCUs are 5 V and 3.3 V, but recently, lower supply voltages of 1.8 V have been growing more common since lower operating voltages generally correlate to lower power consumption. The trade-off most often associated with power consumption is computational power and speed. In other words, higher MCU speeds and computational capabilities require higher power consumption.

The lowest-power MCUs make computations using only 8-bits (i.e., numbers between 0 and 255, or − 128 to 127) at a time in a fixed-point manner. Since measurement data are stored in 16-, 32-, or 64-bit variables, many 8-bit operations are required to process each data point at the node, even for simple processing such as scaling and offset correction. Newer 16-bit MCUs are becoming more power efficient, as are 32-bit and 64-bit MCUs, yet greater power is required to operate the wider data busses. The MCU’s clock rate determines how long each primitive operation takes and can be anywhere from less than 1 MHz to greater than 500 MHz at the cost of significant power consumption. It is easiest for the user to program data processing algorithms using floating point operations. However, for efficient execution of floating point operations, the MCU should have native hardware floating point capabilities. Otherwise floating point operations must be converted to multiple fixed-point calculations. Floating point numbers stored in 32-bits are sufficient for most SHM applications, but 64-bit calculations may sometimes be necessary.

More and more MCU manufacturers are integrating additional peripherals such as ADCs, larger amounts of memory, and even wireless radio circuitry into their MCU product lines. Each of these can drive down circuit component counts and thereby reduce node costs in addition to making node design simpler. However, off-chip peripherals may still be desired if higher performance is designed (e.g., higher ADC resolution or memory size). Quality documentation, an integrated development environment (IDE), and in-system programming capabilities provided by the manufacturer or design community can further ease the job of a wireless node designer.

Important selection criteria for off-chip ADCs include resolution, speed, and power consumption. Resolutions can range from 8- to 24-bits with most structural monitoring applications using 12- or 16-bit ADCs. High sampling rates are less of an issue for structural applications with low-natural frequencies, but ultrasonic monitoring can require sampling in the MHz range. High sampling rates and resolution require greater amounts of power, and therefore should be chosen as low as possible. Other desirable features that may be available include low-power sleep modes, high data-transfer rates, and a low supply-power voltage. Additionally, a wireless node designer may want to consider ADCs with inputs capable of negative and positive voltages, integrated signal conditioning, and multiple sample-and-hold circuits with multiplexing.

Wireless radio modules (i.e., mountable transceivers) that are easily interfaced with the node’s MCU reduce the need for an RF engineer to aid in the wireless node design. These modules most often operate on the 900 MHz or 2.4 GHz ISM license-free radio frequencies using the IEEE 802.15.4 standard for the physical and data link layers. However, radios operating with other standards including IEEE 802.10 (the foundation for Bluetooth) and IEEE 802.11 (the foundation for WiFi) are available. Once again, power is a key selection criterion, which is in a trade-off with transmission range and receiver sensitivity. In fact, the power requirements of the radio are typically the greatest for the entire node. Proper antenna selection can increase communication reliability and range without increased power consumption. ‘Smart antenna’ designs that can actively control output power and signal direction are an active area of research. Well-timed sleep modes in which the receiver is shut off can save considerable power. A node designer’s job can be eased by modules with embedded firmware that automatically implement the lowest levels of the RF stack, relieving the MCU of some computation load.

16.3.2 Wireless sensor families

Application considerations will determine which type of wireless sensor should be used in a wireless structural monitoring system. Power consumption will be less of an issue for short-term deployments of a few days while low power consumption and power harvesting will almost certainly be required for long-term installations. The physical quantity to be measured will also affect sensor selection. While the node can be designed from scratch, it would be most advantageous to use ready-to-deploy, all-in-one commercial units if and only if available devices meet project requirements. More flexible wireless sensor modules, such as the ‘mote’ family, enable the designer to choose the desired transducer, design the power management circuitry, modify the embedded firmware, and fabricate an assembly enclosure for the specific application. If the greatest amount of flexibility is required, yet the designer wishes to not design a unit from the ground up, academic prototypes may be purchased that bring additional features such as advanced sensing (e.g., piezoelectrics^41,42), feedback control,⁴³ and the greatest amount of extensibility. In 2006, Lynch and Loh¹⁸ provided a review of available wireless sensors for structural monitoring at the time, but since technology advances so quickly, newer and more advanced wireless sensor models now exist. Regardless of the family of wireless sensors the user decides to choose from, it will be important that the datasheet of each device being considered is carefully studied. The following datasheet specifications are commonly misread and require intense scrutiny: active versus sleep power consumption along with the time anticipated to be spent active and sleep; maximum theoretical wireless data rate versus the realistic rate at which measurements can be reliably streamed; ADC resolution versus effective resolution including effects by circuit noise; transducer signal bounds, sensitivity, and signal conditioning circuitry; and wireless communication range in environments such as line-of-sight versus lightly or heavily constructed facilities. The remainder of Section 16.3.2 will compare and contrast three main families of wireless sensors currently in use (Fig. 16.3) in an effort to show readers important traits that should be broadly applicable to future generations of devices.

16.3.3 Wireless sensor peripherals

Besides the node core (Plate XIVf, see colour section between pages 294 and 295), the designer or user of a wireless sensing node needs to consider attached peripherals such as transducers, signal conditioners, and energy harvesters.

16.4.1 Operating systems (OS)

The software that makes up the OS on a wireless sensor provides system management features and hardware interfaces to the upper middleware and application layers. It should be noted that the OS on a wireless node is significantly different from the OS on a consumer PC (e.g., Windows, MacOS, and Linux) due to the limited memory and computational capability on each node. The design or selection of the OS task scheduler, the software that determines when tasks are executed based on priority and timing, requires careful consideration. Since only one task can run on the processor at any given instant, the OS must be able to preempt tasks (i.e., interrupt a lower priority task to allow a higher priority task to complete) as quickly as possible by moving the current state of the preempted task into temporary memory and allocate resources for the preempting task. This feature allows an MCU to execute higher priority applications quicker in the middle of longer lower priority applications instead of having to wait for the first application to complete. However, task preemption can lead to system failures such as deadlock and task priority inversion. The former causes the processor to ‘freeze’ and can be mitigated by the strategies out-lined in Reference 66. The latter famously led to a failure of the NASA Mars Pathfinder spacecraft; fortunately, the error was discovered and fixed through a remote system upgrade.⁶⁷

A robust task scheduler and the ability to perform remote system upgrades are just two of many features an embedded OS should contain. The OS should be as memory efficient as possible, since MCUs typically used in the design of wireless sensor nodes have limited memory compared to larger computing systems, such as PCs. Just as in hardware design, the software design must always consider power consumption. The processor should go into a low-power idle or sleep state after all tasks in the queue have completed and wake again when an interrupt occurs. The OS is often selected, and not designed, by the designer of the WSN. As such, the OS should be well documented and easy for the WSN designer to implement and/or develop the middleware and application layers. Besides a well-documented application programming interface (API), firmware development time will be shortened if the OS has hardware abstraction layers (HAL) for a wide variety of target processors, a strong community user base, reliable professional support, and the ability to write applications in a popular programming language (e.g., C/C++).

The ubiquity of embedded systems has led to the development of a wide range of embedded OSs. Since nearly instantaneous response to external events such as measurement sampling and radio transmissions are critical for wireless sensors, real-time operating systems (RTOS) have become popular for wireless sensors. An RTOS can guarantee certain capabilities within specified time constraints and should autonomously handle common system failures. FreeRTOS⁶⁸ and the Micrium RTOS⁶⁹ are open source RTOS that are available for free, allow the user to modify the source code, and have become popular for WSN. A commercial RTOS, such as Keil RTX⁷⁰ for ARM processors and Wind River VxWorks⁷¹ developed by Intel, is extensively used in industry and provides professional support and a richer and more robust feature set. All of these embedded RTOSs are multi-threaded. While not considered an RTOS, TinyOS⁷² was developed specifically for WSN and is uniquely programmed using the nesC programming language, a derivative of C. Extra care must be taken when using a non-RTOS to ensure data collection is synchronized across the network and time-critical tasks (e.g., interrupts associated with feedback control for structural excitation) are executed with minimum jitter (i.e., small amounts of delay from when the task should have been executed to when it is actually executed). These issues have been addressed with a well-defined middleware⁷³ in the iMote 2 sensors running TinyOS.^28,50,74 Unfortunately, TinyOS is not multi-threaded, which is another limitation. The complexity, expense, and memory requirements associated with an OS are not always required, in which case a simple interrupt-based state machine along with a HAL can be custom developed. This strategy has been successfully employed on the Narada wireless sensor with a low-power 8-bit MCU and only 128 kB of external memory⁷⁵ compared to the 32-bit processor on the iMote 2.0 with 64 MB of memory running TinyOS.⁷⁶ A state-machine strategy can also be properly designed with interrupts so that the embedded firmware is effectively multi-threaded, yet still capable of real-time operation.

16.4.2 Middleware

Middleware is a software link between the OS and the application layer. It is intended to simplify the job of the application designer and to provide a wide variety of functionality. Wireless boot loaders are desirable because they allow remote firmware upgrades to the WSN, and can even allow software ‘agents’ to be dynamically distributed throughout.^77,78 Since the IEEE 802.15.4 wireless transmission protocol (see Section 16.2) has no prescribed network layer for handling packet failure, middleware is required to correct these failures in the most efficient way, based on the WSN topology. As has been common to all aspects of WSN discussed in this chapter, middleware can improve energy efficiency, by scheduling the use of lowpower ‘sleep modes’ and duty cycling the radio receiver as discussed in Section 16.2 for TDMA wireless protocols. Additional energy can be saved by reducing the number of bits wirelessly transmitted by compressing the data to be sent using a lossless compression algorithm.⁷⁹ Although structural monitoring WSN are relatively static, a method should still exist for new units to be added to the network, or existing units upgraded. Novel techniques for resource allocation include: adaptive fidelity algorithms, in which nodes are near an ‘important event’ sample with greater resolution than those far away;⁸⁰ and dynamic allocation of computation resources using a buyer/seller framework.⁸¹

Message routing is an important middleware feature because it determines the reliability and efficiency of communications. Message routing in single-hop and multi-hop network architectures is shown in Plate XII. A multi-hop strategy, i.e., one in which the radio output power is reduced so the signal reaches only the closest node in a path to the destination, can be employed to minimize each node’s transmission power. The area under the curves in Figure 16.4 are analogous to the amount of power required to transmit over the distance shown for the single-hop configuration (dashed) and multi-hop (dotted). If n transmissions are used to cover a range R, then the reduction in total transmission power is proportional to R²/n. However, in practice, choosing the correct transmission power, if selectable at all, is a difficult task to accomplish reliably. Additionally, the power saving advantage needs to balance the increased latency associated with each ‘hop’ the message makes and the exponential decrease in reliability with respect to the number of ‘hops,’ which would increase power consumption due to many re-transmissions.

Time synchronization of the data collected by all the nodes in the WSN is an inherent challenge with wireless sensors, and is best achieved with effective middleware. In effect, each node maintains its own clock. This is in stark contrast to wired monitoring systems, where the data sampling for all sensors is triggered by the same clock channel. The presence of multiple clocks brings a unique challenge to WSN: a means to synchronize all of the nodal clocks such that an agreement on a common time basis is necessary. De-synchronization can occur on the network level due to network latency and on the hardware level due to gradual drifts of a node’s clock commonly provided by a low-cost crystal. Clock drift can be reduced by using a more expensive high quality crystal or a thermally corrected crystal that draws additional power. If no ‘hops’ are required to transmit from the network coordinator to all the nodes, a simple beacon can be used to synchronize the network to within 30 μs.⁸² Although synchronized, all the nodes in the network are delayed by an unknown amount from the coordinator’s timing due to radio latency on the order of 10 ms. Multi-hop mesh networks require a more complex strategy such as the flooding time synchronization protocol⁸³ that quantifies the stochastic delay in each link of the WSN.

The most extensive middle package for SHM was developed by Nagayama, et al.⁷³ for the iMote2 running TinyOS, and includes features such as reliable data transfer, network data aggregation, and sensor synchronization. All of the middleware solutions deployed should share a common application interface so that the application design can effectively utilize the provided features.

16.4.3 Application software

The application software embedded in the WSN is significantly different from the applications that run on a typical PC. While the available computational and memory resources might be equivalent to a PC, they are distributed across all the nodes in the network, as schematically shown in Fig. 16.5; therefore, the applications must also be distributed. Embedded applications must be efficient in terms of processor cycles, memory, and energy. Often, only a single application runs on each node due to limited local resources, while a multitude of different applications run simultaneously across the WSN. The ultimate goal of WSN applications is similar to those for wired structural monitoring systems (e.g., data management, damage detection, system ID, and even structural control). The local data processing at the node is a major paradigm shift from wired sensor networks. Important considerations for application designers include accuracy, speed, power efficiency, wireless bandwidth usage, and end-user usability.

In many structural monitoring systems, large amounts of data are collected, but only key characteristics are needed by the user. This disparity necessitates automated methods for data processing and management. For wireless impedance-based SHM the time-history data are converted to the more sparse frequency domain, which is then transferred to the site repository for future processing and analysis.⁵³ Since the ability to compress data by domain transformations is not always possible, the data repository can direct the method by which the WSN relays the data in real-time depending on whether raw or pre-processed data is required.⁸⁴

When the goal of a wireless SHM installation is damage detection, the analysis can take place autonomously, relaying only the network’s prognosis (damage versus undamaged). Automated wireless structural damage detection was first proposed by Straser and Kiremidjian, who used the Arias intensity computed at each node using local acceleration data as an indirect method to detect the energy dissipated as a component in the structure is damaged during an earthquake.¹⁷ Developed for automated damage detection application to situations other than earthquakes, ARX (autoregressive with exogenous input) models have been embedded in a WSN and the magnitude of the residual of the coefficients for nominal can be used to identify the presence of damage.⁸⁵ Instead of using global vibration characteristics to detect damage in concrete structures, displacement measurements can be autonomously compared to embedded damage index models to provide estimates of local structural damage.⁸⁶

An entire model of the system can be computed from vibration data using a combination of in-network computing and server-side computing. One realization use a node’s local response data to compute Markov parameters, which are then sent to the base station using significantly less bandwidth than would be required to send raw data. The Markov parameters are then input into an ERA.⁸⁷ The algorithm successfully identified system parameters of an auditorium balcony using less wireless bandwidth than other strategies. These examples show that many traditional damage detection and system ID algorithms can be distributed across the computational capabilities of the nodes in the wireless network, thus increasing scalability compared to wired systems with central data processing centers.

The lessons learned from developing wireless structural monitoring systems have been extended to wireless feedback control systems. Structural control was first proposed in 1972, but has seen only limited use due to installation costs and reliability concerns.⁸⁸ Augmenting these initial control systems with wireless telecommunications has been shown to lead to improved performance over completely decentralized systems, without the added cost of cabling associated with centralized controllers. Due to bandwidth limitations and latency associated with wireless communication, novel distributed control algorithms were developed that utilize the embedded computing available at each node and communicate with other nodes as little as possible.

Wireless control networks are well suited for large systems with many actuators and sensors such as semi-active control, i.e. actuation involves changing stiffness and damping properties of structural elements instead of actively applying a force. Initial research on wireless semi-active control systems considered full measurement feedback at slow sampling rates or fully decentralized control at high sampling rates.⁴⁰ Building on existing optimal control theory, the sub-optimal clip-linear quadratic regulator has been an extensive topic of research, along with decentralization of the associated state estimator. One method to strategically utilize wireless bandwidth and achieve performance near that of full state feedback is to use the error between two parallel estimators on each node and request asynchronous measurement updates from other nodes in the network when the error reaches a threshold.⁸⁹ The robust H-∞ control law can be applicable by homotopically transforming the feedback matrix to yield sparsity aligned with the sensor sub-networks.⁹⁰ In this way, the advantage of feedback is achieved by communicating feedback data on different, non-interfering, wireless channels for each sub-network. Less traditional control techniques, but well-fitting to agent-based wireless networks, are the class of game-theory inspired actuator resource allocation algorithms. These have been experimental validated with large-scale structural-control experiments⁹¹ and can account for the inherent non-linear characteristics of semi-active control devices, a feature not present in clipped-optimal controllers.⁹² Most recently, a wireless control and monitoring system has been developed that unifies structural monitoring with feedback semi-active vibration control with a graphical interface for end users⁹³ for maintenance and management.

Researchers are continuing to develop distributed algorithms for WSN that are inspired by, but different from, those found in parallel supercomputers due to the significant communication latency in wireless networks. Well-developed applications interfaced with the appropriate middleware, OS, and hardware yield a WSN that can perform as well as a comparable wired monitoring system with new features, decreased costs, and reduced installation time.