Sudden Event Monitoring of Civil Infrastructure Using Demand ... - MDPI

sensors Article

Sudden Event Monitoring of Civil Infrastructure Using Demand-Based Wireless Smart Sensors Yuguang Fu 1, *, Tu Hoang 1 , Kirill Mechitov 2 , Jong R. Kim 3 , Dichuan Zhang 3 and Billie F. Spencer, Jr. 1 1 2 3

*

Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA; [email protected] (T.H.); [email protected] (B.F.S.J.) Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA; [email protected] Department of Civil and Environmental Engineering, Nazarbayev University, Astana 010000, Kazakhstan; [email protected] (J.R.K.); [email protected] (D.Z.) Correspondence: [email protected]

Received: 18 November 2018; Accepted: 14 December 2018; Published: 18 December 2018







Abstract: Wireless smart sensors (WSS) have been proposed as an effective means to reduce the high cost of wired structural health monitoring systems. However, many damage scenarios for civil infrastructure involve sudden events, such as strong earthquakes, which can result in damage or even failure in a matter of seconds. Wireless monitoring systems typically employ duty cycling to reduce power consumption; hence, they will miss such events if they are in power-saving sleep mode when the events occur. This paper develops a demand-based WSS to meet the requirements of sudden event monitoring with minimal power budget and low response latency, without sacrificing high-fidelity measurements or risking a loss of critical information. In the proposed WSS, a programmable event-based switch is implemented utilizing a low-power trigger accelerometer; the switch is integrated in a high-fidelity sensor platform. Particularly, the approach can rapidly turn on the WSS upon the occurrence of a sudden event and seamlessly transition from low-power acceleration measurement to high-fidelity data acquisition. The capabilities of the proposed WSS are validated through laboratory and field experiments. The results show that the proposed approach is able to capture the occurrence of sudden events and provide high-fidelity data for structural condition assessment in an efficient manner. Keywords: sudden event monitoring; wireless smart sensors; demand-based nodes; event-triggered sensing; data fusion

1. Introduction Many civil infrastructure damage scenarios involve sudden events, such as natural disasters (e.g., earthquakes) and human-induced hazards (e.g., collisions, explosions, acts of terrorism). The occurrence of these events is generally unpredictable, and the consequences can be catastrophic. A typical example of catastrophic sudden event is found in the accidental collision between barges and a piling of railroad bridge in Mobile, Alabama, in 1993 [1]. As a result, the railroad bridge was damaged and gave way 20 min later when an Amtrak train crossed, killing 47 people. If this collision had been detected immediately and timely structural assessment of the bridge made, then the deaths of these individuals may have been prevented. To mitigate the consequences of sudden events, the development of monitoring systems is of great importance. Traditional monitoring systems use wired sensors [2–5]. These monitoring systems not only enable sudden event detection but can also facilitate rapid condition assessment of civil Sensors 2018, 18, 4480; doi:10.3390/s18124480

www.mdpi.com/journal/sensors

Sensors 2018, 18, 4480

2 of 17

infrastructure. Such wired monitoring systems require line power to operate and can be expensive for many large-scale structures, due primarily to high installation costs [6], often ranging from $5 K to $20 K per channel (e.g., the Bill Emerson Memorial Bridge monitoring system cost a total of $1.3 M for 86 sensors [4,7]). High-fidelity wireless sensors offer tremendous opportunities to reduce costs and realize the promise of pervasive sensing for structural condition assessment. However, sudden event detection using wireless sensors remains elusive. For example, the monitoring system installed on the Golden Gate Bridge was unable to detect the three earthquakes that occurred during the three-month monitoring deployment [8]. Two main challenges to detect sudden events are apparent: (i)

(ii)

Limited power. Most wireless sensors are duty-cycled to preserve limited battery power; as a result, wireless sensors will miss the occurrence of sudden events when they are in power-saving sleep mode. Because the duty cycle is typically below 5% [9], this scenario is quite likely to occur. Response latency. Response of wireless smart sensors (WSS) from sleep mode to data acquisition may take over a second, resulting in the loss of critical information in short-duration events (e.g., earthquakes and collisions). Moreover, even if awake, sensors may be busy with other tasks (e.g., data transmission); therefore, they will be unable to respond immediately to the occurrence of sudden events, and hence miss the short-duration events.

Addressing these challenges is critical to realizing a WSS for sudden event detection. One intuitive strategy is to provide sustainable power for WSS to enable continuous monitoring of structures subjected to sudden events, emulating traditional wired monitoring systems. For example, in 2011, Potenza et al. [10] installed a wireless structural health monitoring (SHM) system consisting of 17 WSS on a historical church, which was damaged during the 2009 L’Aquila earthquake. The nodes were powered by the existing electrical lines, which guaranteed the continuity of operation and successfully detected several earthquakes over a 3-year monitoring period. Their strategy of using electrical lines to power WSS does not retain the inherent advantages of cable removal, and thus is not practical for other sudden event monitoring applications using WSS. Energy harvesting and wireless power transfer technologies also do not provide an efficient solution. Although technologies such as solar and wind energy harvesting have been developed and validated to power WSS for periodic monitoring [11–13], the challenge is that energy harvesting from the ambient environment is intermittent and time-varying, which is not reliable to support continuous monitoring of structures. Radio frequency (RF) energy transfer and harvesting is another wireless power technique in which WSS convert the received RF signals into electricity. The energy can be transferred reliably over a distance from a dedicated energy source to each node, or dynamically exchanged between different nodes [14]. However, the energy harvesting rate is on the order of micro-watts with low efficiency [15] and is insufficient for high-power high-fidelity monitoring of sudden events. On the other hand, power consumption can be reduced by employing various energy-saving mechanisms, which help to mitigate, but do not fully address the challenge of limited power for WSS. For example, Jalsan et al. [16] proposed layout optimization strategies for wireless sensor networks to prolong the network lifetime by optimizing communication schemes without compromising information quality. Other examples of energy-saving mechanisms include data reduction, radio optimization, and energy-efficient routing. More detailed discussion can be found in Reference [17]. Most of these strategies are designed to reduce power consumption for wireless transmission, which does not help energy conservation for continuous sensing, because most of the power draw comes from the always-on sensor. Recent developments in event-triggered sensing present both opportunities and challenges to realize sudden event monitoring using WSS. In event-triggered sensing, WSS only initiates measurement in response to signaling of events, which helps to save both energy and memory resources, and thus prolong the lifetime of WSS. Research has been conducted to implement low-power

Sensors 2018, 18, 4480

3 of 17

components (sensors and radios) that enable continuous operation and triggering mechanisms inside each sensor node. Lu et al. [18] designed the TelosW platform, which is an upgrade of the TelosB platform [19], by adding ultra-low power wake-on sensors and wake-on radios. The wake-on sensor is able to wake up the microcontroller (MCU) on occurrence of events with a predetermined threshold. Additionally, the wake-on radio can wake up the MCU when a triggering radio message is received. Similarly, Sutton et al. [20] presented a heterogeneous system architecture which included a low-power event detector circuit and low-power wake-up receivers. Although these two technologies achieve low power consumption, they do not satisfy the high-fidelity requirement of sudden event monitoring for civil infrastructure. For example, the TelosW’s analog to digital converter has only 12-bit resolution. Event-triggered sensing is also developed and implemented to facilitate railway bridge monitoring, because strain cycles and vibrations induced by trains are the most important data for bridge condition assessment (e.g., fatigue), but the arrival time of trains is generally unpredictable. Bischoff et al. [21] deployed a wireless monitoring system which provided strain measurement and fatigue assessment of the Keraesjokk Railway Bridge. Each node was triggered independently by a low-power microelectromechanical systems (MEMS) accelerometer which operated continuously and detected an approaching train. Bias due to the transient start-up nature of the strain gage was removed by a post-processing technique. Liu et al. [22] developed an on-demand sensing system, named ECOVIBE, to monitor train-induced bridge vibrations. In each wireless node, a passive event detection circuit was designed to monitor bridge vibration with no power consumption, and another adaptive logical control circuit powered off the node once the designated tasks were finished. While effective for some applications, all the aforementioned approaches will lose critical data between the occurrence of the event and the time that data begins to be collected. Conversely, response times of wireless sensors from a cold boot to data acquisition are typically well over a second, making this problem particularly acute for short-duration sudden events (e.g., impacts can last only fractions of a second). Moving the triggering mechanism to outside the sensor nodes provides a solution to address the challenge of data loss. In general, a separate trigger node or system is used to monitor the events continuously and notify of events to sensor nodes which are in power-saving mode most of the time. The trigger node/system is required to send notifications with a certain amount of time before the arrival of events at the structure, compensating the response latency of other sensor nodes. For example, an event-driven wireless strain monitoring system was implemented on a riveted steel railway bridge near Wila, Switzerland [23]. Two trigger nodes, referred to as sentinel nodes, were placed at 50 and 85 m away from the bridge, detecting approaching trains and sending alarm messages using a reliable flooding protocol to wake up sensor nodes on the bridge, before the train arrived. In a 47-day deployment, the system successfully detected 99.7% of train-crossing events. Likewise, in order to detect earthquakes and initiate seismic structural monitoring, Hung et al. [24] developed an intelligent wireless sensor network embedded with an earthquake early warning (EEW) system which was able to detect P-waves before earthquakes arrived. In addition, each sensor node was implemented with a wake-on radio which supported ultralow-power periodic listening of wake-up commands, while the main sensor node was in deep sleep mode. Once the P-wave was detected, the gateway node, integrated with the EEW system, sent wake-up commands to sensor nodes approximately 2 s ahead of earthquakes. Subsequently, sensor nodes started measurement with a latency time of only 229 ms. Despite successful detection of train-crossing and seismic events, the aforementioned methods do not provide a universal solution to address the challenge of data loss for many other sudden events, e.g., bridge impact by over-height trucks and ships which can hardly be detected ahead of impacts. In addition, some progress has been made in addressing the challenge of response latency to sudden events, when WSS are awake but not in sensing mode. Cheng & Pakzad [8] proposed a pulse-based media access control protocol. When an earthquake occurs, a trigger message with high priority is propagated from an observation site across the WSS network to preempt current tasks; sensors will be forced to conduct measurement to capture the structural response under the earthquake. Dorvash et al. [25] developed the Sandwich node to reduce the response latency for unexpected events.

Sensors 2018, 18, 4480

4 of 17

A smart trigger node continuously measures the structural response; it will broadcast a proper message across a network of Sandwich nodes in the case of occurrence of events. Sandwich nodes keep listening to the trigger message; they will preempt current tasks once the trigger message is received. Response delay of Sandwich nodes is around 8 ms to the occurrence of events. Although response delay is reduced in these two strategies, the wireless sensor’s radio must always be on to listen for messages from a trigger node. Unless employing an ultralow-power wake-up radio, these strategies will result in a significant power draw. This paper proposes a new approach for monitoring civil infrastructure subjected to sudden events, aimed at detecting sudden events of any duration and capturing complete transient response of any length. A demand-based wireless smart sensor (WSS) is developed that can capture data during the sudden event that is suitable for rapid condition assessment of civil infrastructure. As opposed to periodic monitoring, the demand-based WSS only wakes up and initiates sensing in response to specific conditions, such as sudden events. The results of laboratory experiments and a field experiment show that our proposed approach can capture the occurrence of sudden events and provide high-fidelity data for structural condition assessment in a timely and power-efficient manner. 2. Demand-Based WSS As discussed in the previous section, the primary issues that must be overcome to use wireless sensors to monitor civil infrastructure subjected to sudden events are: (i) the sensor must operate on battery power, (ii) high-fidelity data appropriate for SHM application must be obtained, (iii) data surrounding the occurrence of sudden events must not be lost, and (iv) the WSS node must have sufficient computational power to translate the data collected into actionable information. This section describes a demand-based WSS system that can address these issues. 2.1. Ultralow-Power Trigger Accelerometer for Continuous Monitoring To ensure that the occurrence of sudden events is not missed, the monitoring system must be continuously in an on state. A wireless node that is always on would quickly deplete its battery. Therefore, the solution proposed herein is to use an ultralow-power trigger accelerometer that can continuously monitor the vibration of structures; the data from the accelerometer is stored in a First-In-First-Out (FIFO) buffer. When an event occurs, the data in the FIFO buffer will be frozen, and the sensor triggers an interrupt signal to wake up the main sensor platform and start sensing. Such a trigger accelerometer should have low power consumption to enable continuous monitoring for several years, good sensing characteristics, including a high sampling rate and adequate resolution, and a large FIFO buffer to ensure data is not lost after the triggering event. Trigger accelerometers in the market today were compared and the candidates that satisfied the basic needs of sudden event monitoring are listed in Table 1. The power consumption reported in the table correspond to the ultralow-noise mode of each sensor. More specifically, the ADXL362, developed by Analog Devices, consumes much less power than the other trigger accelerometers. The ADXL372, an updated high-g version of ADXL362, has a larger sampling rate and measurement range, but with a sensing resolution of only 100 mg. The LIS3DSH from STMicromechanics has high resolution of 0.06 mg, but it has a high-power draw and an inadequate FIFO buffer. Finally, the MPU6050 developed by InvenSense features a large FIFO buffer and high resolution, but it consumes substantial power. In sum, based on application needs, the ADXL362 has been selected for this study (Figure 1); it integrates a three-axis microelectromechanical systems (MEMS) accelerometer with a temperature sensor, an analog-to-digital converter, and a Serial Peripheral Interface (SPI) digital interface. The ADXL362 consumes only 13 uA in ultralow-noise mode at 3.3 V, which theoretically could work continuously for over two years on a single coin-cell battery. A sampling rate up to 400 Hz and a resolution of 1 mg is supported, satisfying many SHM applications. The large FIFO buffer allows the sensor to save up to 512 samples, which corresponds to 1.7 s for all three axes sampled at 100 Hz.

Sensors 2018, 18, 4480

5 of 17

Moreover, has PEER built-in logic Sensors 2018, 18,it x FOR REVIEW

for acceleration threshold detection; a detected event can be 5used of 17as a trigger to wake up the primary sensor node. In sum, based on application needs, the ADXL362 has been selected for this study (Figure 1); it Table 1. Comparison of trigger accelerometers in the market [26–29]. integrates a three-axis microelectromechanical systems (MEMS) accelerometer with a temperature sensor, an analog-to-digital converter, and a Serial Peripheral Interface (SPI) digitalMPU6050 interface. The ADXL362 ADXL372 LIS3DSH ADXL362 consumes only 13 uA in ultralow-noise mode at 3.3 V, which theoretically could work Manufactures Analog devices Analog devices STMicroelectronics InvenSense continuously for over two years on a single coin-cell battery. A sampling rate up to 400 Hz and a Supply voltage (V) 1.6–3.5 1.6–3.5 1.7–3.6 2.4–3.5 Power consumption (uA) 13 33 225 500 resolution of 1 mg is supported, satisfying many SHM applications. The large FIFO buffer allows the Sampling rate (Hz) 12.5~400 400~6400 3.125~1600 4~1000 sensor toMeasurement save up torange 512 samples, to 1.7 s for all±4,three (g) ±which 2, ±4, ±corresponds 8 ±200 ±2, ±8, ±axes 16 sampled ±2, ±4, ±at 8, 100 ±16 Hz. Resolution (mg) 1 mg 100 mg 0.06 mg 0.06 mg Moreover, it has built-in√logic for acceleration threshold detection; a detected event can be used as a (µg/ Hz) 175–350 5300 150 400 trigger toSpectral wakenoise up the primary sensor node. Buffer size (samples)

512

512

32

(a)

512

(b) Figure 1. 1. ADXL362: (a)(a) sensor chip; (b)(b) Functional block diagram [26]. Figure ADXL362: sensor chip; Functional block diagram [26].

2.2. High-Fidelity Sensor Platform Sudden Event Monitoring 2.2. High-Fidelity Sensor Platform forfor Sudden Event Monitoring provide high-quality measurement data and enable rapid condition assessment structures ToTo provide high-quality measurement data and enable rapid condition assessment ofof structures subjected to sudden events, the sensor platform should have following features: (i) sensors and a data subjected to sudden events, the sensor platform should have following features: (i) sensors and a acquisition system that can obtain high-quality data at a high sampling rate for the event; (ii) powerful data acquisition system that can obtain high-quality data at a high sampling rate for the event; (ii) microcontroller to acquiretoand analyze sensor datasensor in neardata real time. Other features include: powerful microcontroller acquire and analyze in near realimportant time. Other important reliable communication, open-source software, and efficient data and power management. features include: reliable communication, open-source software, and efficient data and power A summary of the most advanced wireless sensor platforms available in the market currently management. is given in Table Xnode, developed bysensor Embedor Technology, hasina the 24-bit Analog-to-Digital A summary of2. theThe most advanced wireless platforms available market currently is Converter (ADC), which is the best in its class. The microprocessor unit (MCU) of Waspmote (Libelium, given in Table 2. The Xnode, developed by Embedor Technology, has a 24-bit Analog-to-Digital Zaragoza,(ADC), Spain) iswhich not able processing large amount of data. The MCU information Converter is to thesupport best inrapid its class. The of microprocessor unit (MCU) of Waspmote for the AX-3D (BeanAir, Berlin, Germany) and the G-Link-200 (LORD Sensing, Williston, VT, The USA) (Libelium, Zaragoza, Spain) is not able to support rapid processing of large amount of data. is unavailable, operating systemsBerlin, of these platforms Note that several MCU informationand for the the AX-3D (BeanAir, Germany) andare theproprietary. G-Link-200 (LORD Sensing, high-performance wireless sensor platforms (e.g., Imote2) are no longer commercially available, Williston, VT, USA) is unavailable, and the operating systems of these platforms are proprietary. and hence not compared herein. Note that several high-performance wireless sensor platforms (e.g., Imote2) are no longer

commercially available, and hence not compared herein.

Table 2. Comparison of most advanced wireless sensor platforms in the market [30–34].

Table 2. Comparison platforms in the market [30–34]*. Martletof most advanced AX-3D wireless sensor Waspmote G-Link-200 ADC resolution (bits) Sampling rate (Hz)

12

Martlet

up to 3 M

ADC resolution 12 MCU Frequency (bits) 80 M (max. Hz) Sampling rate (Hz) up to 3 M LOS range (m) 500 MCU Frequency 80 M Operating system State-machine (max. Hz) Data memory 32 GB LOS range (m) 500 Internal power source dry cell battery Operating system State-machine External power source

NA

16

16

20

AX-3D

Waspmote

G-Link-200

16

16

20

3.5 k

0.5–1 k

0–4 k

NA Proprietary

32 k Proprietary

NA Proprietary

16 GB

8 million data points

lithium-ion battery

dry cell battery

3.5 k NA 650

1 million data points

650 Proprietary

lithium-ion battery primary cell/8–28 V DC

0.5–1 k 32 k

1.6 k

1.6 k Proprietary 7 V DC

0–4 k NA 2k

2k Proprietary NA 8 million data points

Xnode 24

Xnode 0–1.6 k

24

204 M

0–1.6 k 1k

204 M FreeRTOS 4 GB

1k FreeRTOS

lthium-ion battery 5 V DC

Data memory

32 GB

1 million data points

16 GB

4 GB

Internal power source External power source

dry cell battery

lithium-ion battery

lithium-ion battery

dry cell battery

lthium-ion battery

NA

primary cell/8–28 V DC

7 V DC

NA

5 V DC

Sensors 2018, 18, 4480

6 of 17

Sensors 2018, 18, x FOR PEER REVIEW

6 of 17

Because of its high sensing resolution, high sampling rate, powerful microprocessor, Because of its high sensing resolution, high sampling rate, powerful microprocessor, and and open-source software, the Xnode Smart Sensor [34] has been selected as the host wireless open-source software, the Xnode Smart Sensor [34] has been selected as the host wireless sensor sensor platform in this study. The standard Xnode consists of three modular printed circuit boards platform in this study. The standard Xnode consists of three modular printed circuit boards (PCB): (PCB): (i) the processor board, (ii) the radio/power board, and (iii) the sensor board (Figure 2). (i) the processor board, (ii) the radio/power board, and (iii) the sensor board (Figure 2). In particular, In particular, it employs an 8-channel, 24-bit ADC (Texas Instruments ADS131E8), a maximum it employs an 8-channel, 24-bit ADC (Texas Instruments ADS131E8), allowing aallowing maximum sampling sampling rate to 16and kHz NXP LPC4357 microcontroller at frequencies up to rate up to up 16 kHz anand NXPanLPC4357 microcontroller operatingoperating at frequencies up to 204 MHz, 204 MHz, which can be used to execute data-intensive on-board computation. Moreover, it implements which can be used to execute data-intensive on-board computation. Moreover, it implements open-source middleware services [35], which custom application development. In addition, open-source middleware services [35], facilitates which facilitates custom application development. In it possesses SPI controllers, it possible communicate with thewith selected trigger addition,two it possesses two SPI making controllers, making ittopossible to communicate the selected accelerometer, the ADXL362. trigger accelerometer, the ADXL362.

(a)

(b)

Figure 2. Xnode: (a) stacked modular boards; weather-proofenclosure enclosure[34]. [34]. Figure 2. Xnode: (a) stacked modular boards; (b)(b) weather-proof

2.3. Integration of Wake-up Sensor andand High-Fidelity Sensor Platform 2.3. Integration of Wake-up Sensor High-Fidelity Sensor Platform To capture the entire event without lossloss of critical information, To capture the entire event without of critical information,the theADXL362 ADXL362accelerometer accelerometer and the Xnode mustmust be carefully integrated to build a demand-based WSS. the Xnode be carefully integrated to build a demand-based WSS.This Thisintegration integrationisisdiscussed discussed in the remainder of this section, in terms of hardware, software, and digital signal the remainder of this section, in terms of hardware, software, and digital signalprocessing. processing. 2.3.1.2.3.1. Hardware Consideration Hardware Consideration To address the challenge of physical integration the ADXL362accelerometer accelerometer into into the Xnode, To address the challenge of physical integration of of the ADXL362 Xnode, a programmable event-based switch was designed and implemented on radio/power the radio/power board of a programmable event-based switch was designed and implemented on the board of the theinXnode in the demand-based Xnode the demand-based WSS. WSS. When a sudden occurs, andvibration the vibration exceeds a user-defined threshold, an interrupt When a sudden eventevent occurs, and the exceeds a user-defined threshold, an interrupt pin pin in the ADXL362 generates a triggering signal. This signal is connected to a MOSFET its in the ADXL362 generates a triggering signal. This signal is connected to a MOSFET to fliptoitsflip state, state, on and the Xnode andhigh-fidelity initiating high-fidelity sensing. Whenends the (lack eventofends (lack of turning on turning the Xnode initiating sensing. When the event acceleration acceleration above a threshold), the other interrupt pin in the ADXL362 generates a signal to notify above a threshold), the other interrupt pin in the ADXL362 generates a signal to notify the Xnode to the Xnode to stop high-fidelity sensing. After data acquisition is completed, the triggering signal is stop high-fidelity sensing. After data acquisition is completed, the triggering signal is cleared, and the cleared, and the MOSFET turns off the Xnode. The communication of control messages between the MOSFET turns off the Xnode. The communication of control messages between the ADXL362 and the ADXL362 and the Xnode is carried out via SPI bus through a four-wire connection. In addition to Xnode is carried out via SPI bus through a four-wire connection. In addition to enable event-triggered enable event-triggered sensing, the proposed switch should be designed to retain traditional sensing, the proposed switch should be designed to retain traditional functionality for periodic functionality for periodic monitoring. Specifically, sensor nodes are operated on low duty cycles, monitoring. Specifically, sensor nodes are operated on low duty cycles, and the base station can access and the base station can access the network of nodes at random to initiate operations or the network of nodes at random initiate operations the network. To achieve measurements in the network.toTo achieve this goal,or a measurements real time clock, in DS3231, is employed in the this goal, a real time clock, DS3231, is employed in the proposed switch. When a user-defined period proposed switch. When a user-defined period passes or at a specific time of day, the DS3231 sends a passes or at a specific time of day, the DS3231 sends a triggering signal which flips the MOSFET triggering signal which flips the MOSFET switch and turn on the Xnode. Then the node remains switch and turn theperiod Xnode. nodefor remains awake a short period Once of time to listen for awake for a on short ofThen time the to listen messages fromfor the base station. a command is messages fromthe thenode basecarries station. a command obtained, nodecheck). carriesThe outcommunication the required task obtained, outOnce the required task is (e.g., sensing,the battery of (e.g.,control sensing, battery between check). The communication of control messages between the bus. DS3231 and the messages the DS3231 and the Xnode is conducted through the I2C Xnode is conducted through the bus. concept of the proposed switch. Five major components are Figure 3a illustrates theI2C design implemented, including ADXL362 (U1), a realswitch. time clock DS3231 an AND gate Figure 3a illustrates thea trigger design sensor concept of the proposed Five major(U2), components are (U3), a latch (U4), and a MOSFET Interrupt from ADXL362 theanDS3231 are implemented, including a trigger sensor(U5). ADXL362 (U1),pins a real timethe clock DS3231and (U2), AND gate to and the MOSFET through the ANDpins gate.from Thisthe circuit enablesand thethe MOSFET beconnected triggered (U3), connected a latch (U4), a MOSFET (U5). Interrupt ADXL362 DS3231toare either the real time orgate. the trigger sensor. In addition, a latch component is added between to thebyMOSFET through theclock AND This circuit enables the MOSFET to be triggered by either the

Sensors 2018, 18, 4480

7 of 17

Sensors 2018, 18, x FOR PEER REVIEW

7 of 17

real time clock or the trigger sensor. In addition, a latch component is added between the AND gate and the MOSFET, to keep the power showsFigure the realized PCB the proposed the AND gate and the MOSFET, to supply keep thestable. powerFigure supply3bstable. 3b shows thefor realized PCB switch. five major components, as well as companion and capacitors, arecapacitors, all soldered for theThe proposed switch. The five major components, as wellresistors as companion resistors and onare theall edge of topon side. some useside. cases downloading to the board debugging, soldered the In edge of top Insuch someasuse cases such ascode downloading codeand to the board theand sensor platform should always be on and therefore the designed switch needs to be bypassed. debugging, the sensor platform should always be on and therefore the designed switch needs to To be achieve this To goal, a 2-pin is added. When the two pins thepins jumper are not connected, bypassed. achieve thisjumper goal, a 2-pin jumper is added. When theon two on the jumper are not proposed works as designed, itotherwise, it is bypassed. theconnected, proposed the switch worksswitch as designed, otherwise, is bypassed.

(a)

(b)

Figure3.3.PCB PCBdesign designfor forthe the event-based event-based switch: realized PCB . Figure switch:(a) (a)design designconcept concept(b)(b) realized PCB.

2.3.2. Software Consideration 2.3.2. Software Consideration In In addition an effective effectiveapplication applicationframework framework additiontotohardware hardwaredevelopment development of of the the prototype, prototype, an is is required to to control the realize event-triggered event-triggeredsensing. sensing. required control thebehavior behaviorofofdemand-based demand-based WSS WSS to realize Figure4 4 shows shows aa flowchart application framework for thefordemand-based WSS. More Figure flowchartofofthethe application framework the demand-based WSS. specifically, when usersusers turn turn on the switchswitch of a main sensorsensor platform, the Xnode first More specifically, when onphysical the physical of a main platform, the Xnode initializes itself and sends which contain configuration parameters (e.g., threshold, first initializes itself and sendscommands commands which contain configuration parameters (e.g., threshold, timers, and data buffer size) to the event-based switch discussed in the previous section. Once thethe timers, and data buffer size) to the event-based switch discussed in the previous section. Once commands have received, the switch completes configuration of the device settings. commands have beenbeen received, the switch completes configuration of the device settings. Subsequently, Subsequently, the ADXL362 starts measurement in ultralow-noise mode, and the Xnode is the ADXL362 starts measurement in ultralow-noise mode, and the Xnode is turned off. If aturned sudden off. If a sudden occurs and the acceleration obtained in theexceeds ADXL362 exceeds the user-defined event occurs andevent the acceleration obtained in the ADXL362 the user-defined threshold, threshold, an interrupt pin, INT1 on the ADXL362 sends a trigger signal to turn on the Xnode. an interrupt pin, INT1 on the ADXL362 sends a trigger signal to turn on the Xnode. Concurrently, Concurrently, the ADXL362 saves 512 data samples into its FIFO buffer surrounding the onset of the the ADXL362 saves 512 data samples into its FIFO buffer surrounding the onset of the event and event and waits for the Xnode to retrieve the data. The Xnode starts high-fidelity data acquisition waits for the Xnode to retrieve the data. The Xnode starts high-fidelity data acquisition using its using its built-in high-power high-accuracy MEMS accelerometer. When the event stops and the built-in high-power high-accuracy MEMS accelerometer. When the event stops and the acceleration acceleration obtained in the ADXL362 is lower than a user-defined threshold for a certain period of obtained in the ADXL362 is lower than a user-defined threshold for a certain period of time, the other time, the other interrupt pin, INT2, in the ADXL362 is triggered. Subsequently, the Xnode stops interrupt pin, INT2, in the ADXL362 the data Xnode stops sensing. high-fidelity sensing. After sensing is is triggered. completed,Subsequently, the Xnode reads from the high-fidelity FIFO buffer of the After sensing is completed, the Xnode reads data from the FIFO buffer of the ADXL362 and fuses ADXL362 and fuses it with the Xnode data. In addition, when the Xnode is busy with other tasks it with thedata Xnode data. In addition, when the Xnode is busy with other (e.g., (e.g., transmission of a previous event), but another sudden event tasks occurs, the data INT2 transmission pin can be of configured a previous event), but another sudden event occurs, pintocan be configured interrupt to interrupt undergoing tasks and force the theINT2 Xnode start high-fidelitytosensing undergoing tasks force the Xnode to startresults high-fidelity sensing In addition, immediately. Inand addition, timing analysis for each stage immediately. of a demand-based WSStiming are analysis results for each stage of a demand-based WSS are presented in the left of Figure 4. presented in the left of Figure 4. For sudden the thresholds thresholdscan canbebedetermined determined based For suddenevents eventsthat thatare arerare, rare,e.g., e.g., earthquakes, earthquakes, the based onon priori informationabout aboutthe thesudden sudden events events that that are monitored. monitored. The cancan be be priori information Thea apriori prioriinformation information estimated numericalanalysis, analysis, the the data data in the history, a preliminary estimated bybynumerical history, or or the themeasurement measurementdata datainin a preliminary test. For someevents eventsthat thatoccur occur frequently, frequently, e.g., e.g., railway vehicles, test. For some railway bridge bridgeimpacts impactsfrom fromover-height over-height vehicles, thresholds canbebedetermined determinedadaptively, adaptively, starting starting from a “training thethe thresholds can from aa relatively relativelylow lowvalue valueduring during a “training phase” and then adjusteduntil untilthe thedetection detection errors errors are research phase” and then adjusted are minimized. minimized.AAmore morecomprehensive comprehensive research regarding the triggering mechanism is the subject of future work to be addressed in the near regarding the triggering mechanism is the subject of future work to be addressed in the nearfuture. future.

Sensors 2018, 18, 4480

8 of 17

Sensors 2018, 18, x FOR PEER REVIEW

8 of 17

Figure 4. Flowchart of demand-based wireless smart sensors (WSS) for event-triggered sensing. Figure 4. Flowchart of demand-based wireless smart sensors (WSS) for event-triggered sensing.

2.3.3. Data Fusion of Trigger Sensor and Xnode Acceleration Record 2.3.3. Data Fusion of Trigger Sensor and Xnode Acceleration Record The objective of the demand-based WSS is to obtain the data from before the trigger event occurs The objective of the demand-based WSS is to obtain the data from before the trigger event occurs until the structural accelerations stop. Specifically, the ADXL362 can record structural measurements until the structural accelerations stop. Specifically, the ADXL362 can record structural surrounding the onset of a sudden event, whilst the Xnode starts sampling the data approximately measurements surrounding the onset of a sudden event, whilst the Xnode starts sampling the data 0.9 s after being triggered. Therefore, the ADXL362 data and the Xnode data must be synchronized approximately 0.9 s after being triggered. Therefore, the ADXL362 data and the Xnode data must be and fused to produce a complete representation of the acceleration record. The following paragraphs synchronized and fused to produce a complete representation of the acceleration record. The describe the challenges encountered in this process, along with the associated resolutions. following paragraphs describe the challenges encountered in this process, along with the associated To fuse the two overlapping data streams, two main challenges should be addressed, including resolutions. (i) differences ratedata between thetwo ADXL362 and the Xnode, and (ii) synchronization To fuse in thethe twosampling overlapping streams, main challenges should be addressed, including error between theinADXL362 datarate andbetween the Xnode Moreand precisely, the first results from (i) differences the sampling the data. ADXL362 the Xnode, andchallenge (ii) synchronization theerror differences the clock rates ofXnode the ADXL362 and the Xnode. The internalresults clock from rate in betweenbetween the ADXL362 data and the data. More precisely, the first challenge thethe ADXL362 has a standard deviation of approximately 3%. One approach might be to calibrate differences between the clock rates of the ADXL362 and the Xnode. The internal clock rate in the a standard deviation of approximately 3%.approach One approach be toascalibrate theADXL362 ADXL362has incorporated in each Xnode; however, this is not might practical, the clockthe rate ADXL362 incorporated in each Xnode; however, this approach is not as the clock will change with temperature, invalidating the initial calibration. Thepractical, second challenge is rate due will to the changeinwith temperature, invalidating the initial seconda challenge is duewill to the variance start-up time of the sensing task on thecalibration. Xnode. AsThe a result, random offset exist variance in start-up time of the sensing task on the Xnode. As a result, a random offset will exist between the two data streams. between the the twochallenges data streams. To tackle identified in the previous paragraph, the beginning of the Xnode data, To tackle the challenges identifieddata, in the previous paragraph, thethe beginning of the Xnode which is overlapped with the ADXL362 was utilized to calibrate entire ADXL362 data data, stream. which is overlapped with the ADXL362 data, was utilized to calibrate the entire ADXL362 data Figure 5 shows a flowchart of this approach. More specifically, the ADXL362, acc adxl0 , with a nominal acc stream. rate Figure 5 shows a flowchart of this approach. MoreHz). specifically, with 0 ,acc sampling of 100 Hz, were first up-sampled to f (1000 The lastthe 400ADXL362, data points ofadxl the s

adxl0

a nominal sampling of 100 Hz, were first up-sampled to f s overlapped (1000 Hz). The data points were chosen as acc adxl1rate , which was assumed to be approximately withlast the400 beginning of the Xnode data. In the meantime, the Xnode data, acc , was sent through an 8-pole elliptic low-pass xnode0 of the accadxl 0 were chosen as accadxl1 , which was assumed to be approximately overlapped with filter with a cutoff frequency of 50 Hz, to have the same bandwidth with acc adxl0 . The first 400 data the beginning of the Xnode data. In the meantime, the Xnode data, acc 0 , was sent through an points of acc xnode0 were considered as acc xnode1 . Based on the datasheetxnode of ADXL362 [26], the clock 8-pole elliptic low-pass filter with a cutoff frequency 50 Hz, have the same bandwidth withthe frequency deviation from the ideal value was within the of range of −to10% and 10%. Therefore, to find accadxl 0 . The first 400 data points of accxnode 0 were considered as accxnode1 . Based on the datasheet

Sensors 2018, 18, x FOR PEER REVIEW

9 of 17

of ADXL362 [26], the clock frequency deviation from the ideal value was within the range of −10% Sensors 2018, 18, 4480

9 of 17

and 10%. Therefore, to find the actual sampling frequency of accadxl1 , exhaustive search was

applied from 900 Hz to 1100 Hz. For Step i, the estimated sampling frequency ( f e ) of the ADXL362 actual frequency of acc adxl1 , exhaustive search was applied from 900 Hz to 1100 Hz. For Step datasampling was set as, i, the estimated sampling frequency ( f e ) of the ADXL362 data was set as, f e  1000  df [i ] (1) f e = 1000 − d f [i ] (1) where, df [i ]  i  100, i  [0, 200] . accadxl1 was resampled from f e to f s using

where, d f [i ] = i − 100, i ∈ [0, 200Then, ]. accto was resampled from f e toerror f s using resampling-based resampling-based approach [36]. estimate the synchronization between accadxl1 and adxl1 approach [36]. Then, to estimate the synchronization error between acc and acc adxl1optimal xnode1 , offset, accxnode1 , the cross-correlation between the two data segments was calculated. The the cross-correlation between the two data segments was calculated. The optimal offset, SE[i ], SE[i ] , was obtained, for which the cross-correlation reaches its maximum value. Afterwards, was obtained, for which the cross-correlation reaches its maximum value. Afterwards, acc adxl1 was was shifted by SE[i ] , and then the data fusion error (Err) was calculated as, accadxl shifted by1 SE [i ], and then the data fusion error (Err) was calculated as,

Err[i ]  acc

 acc

adxl1− acc xnode1 k2 Err [i ] = k acc adxl1 xnode1 2

(2)

(2)

where, represents Euclidean norm. After completing these steps, the best estimations of 2 where, k k2 represents Euclidean norm. After completing these steps, the best estimations of sampling samplingf afrequency errorobtained obtained in the step that achievesErr. f a and synchronization SEa were frequency and synchronization error SEa were in the step that achieves minimal Subsequently, f and SE were applied to calibrate the original data set, acc . Finally, acc a a adxl0 data set, acc adxl0 and minimal Err. Subsequently, f a and SEa were applied to calibrate the original adxl 0 . acc xnode0 were combined and down-sampled to 100 Hz for ensuing analysis. Finally, accadxl 0 and accxnode 0 were combined and down-sampled to 100 Hz for ensuing analysis.

(a)

(b) Figure 5. Post-sensing data fusion: (a) illustration of data two data sources, (b) flowchart data fusion Figure 5. Post-sensing data fusion: (a) illustration of two sources, (b) flowchart of dataoffusion strategy. strategy.

3. Validation of the Demand-Based WSS Performance To validate the performance of the demand-based WSS, laboratory tests were carried out for data fusion and earthquake monitoring. The detailed test setup and results are presented in this section.

Sensors 2018, 18, x FOR PEER REVIEW

10 of 17

3. Validation of the Demand-Based WSS Performance SensorsTo 2018, 18, 4480the performance of validate

10data of 17 the demand-based WSS, laboratory tests were carried out for fusion and earthquake monitoring. The detailed test setup and results are presented in this section. The performance performance of of the the demand-based demand-based WSS WSS is is discussed, discussed, in in terms terms of of power power consumption, consumption, sensing sensing The characteristics, and data quality for sudden event monitoring. characteristics, and data quality for sudden event monitoring.

3.1. 3.1. Validation Validation of of Data Data Fusion Fusion A lab test test was was conducted conducted to to illustrate illustrate the the challenge challenge of of data data fusion fusion between between the the ADXL362 ADXL362 data data A lab and theXnode Xnodedata. data.Specifically, Specifically, a demand-based was located 10thoffloor of an building 18-story and the a demand-based WSSWSS was located at 10thatfloor an 18-story building model shown in Figure 6. The ADXL362 was configured to capture samples at Hz,s model shown in Figure 6. The ADXL362 was configured to capture samples at 100 Hz, starting100 at 0.2 starting 0.2 s before ands after continuing until s after the event. The before theattriggered event the and triggered continuingevent until 1.5 the event. The1.5 event-triggering threshold event-triggering threshold was set to 150 mg, at which time, the Xnode was turned on and 1000 Hz was set to 150 mg, at which time, the Xnode was turned on and 1000 Hz high-fidelity measurement high-fidelity measurement was started. To reduce false positives, two consecutive data points was started. To reduce false positives, two consecutive data points exceeding the threshold were exceeding threshold wereInrequired cause piezoelectric triggering. Inaccelerometer, addition, a wired required tothe cause triggering. addition,toa wired modelpiezoelectric PCB353B33, accelerometer, model PCB353B33, was installed on the same floor and sampled at a frequency 128 was installed on the same floor and sampled at a frequency of 128 Hz. The acceleration fromof these Hz. The acceleration from these sensors served as reference data. A sudden event was simulated by sensors served as reference data. A sudden event was simulated by manually shaking the building manually shaking the building model in horizontal direction. model in horizontal direction.

Figure 6. Experiment setup setup for for a a demand-based demand-based WSS. Figure 6. Experiment WSS.

To make a direct comparison in the time domain, the Xnode data and wired sensor data were sent To make a direct comparison in the time domain, the Xnode data and wired sensor data were through an 8-pole elliptic low-pass filter with a cutoff frequency of 50 Hz, as displayed in Figure 7. sent through an 8-pole elliptic low-pass filter with a cutoff frequency of 50 Hz, as displayed in Figure The direction of acceleration measurement was the same with the vibration direction specified in 7. The direction of acceleration measurement was the same with the vibration direction specified in Figure 6. The vibration exceeded the threshold at 0.7 s, and the event-based switch turned on the Xnode. Figure 6. The vibration exceeded the threshold at 0.7 s, and the event-based switch turned on the The Xnode required 0.92 s for initialization. Fortunately, as shown in Figure 7b, acceleration data Xnode. The Xnode required 0.92 s for initialization. Fortunately, as shown in Figure 7b, acceleration stored in the FIFO buffer of the ADXL362 was recorded during this period. Specifically, the ADXL362 data stored in the FIFO buffer of the ADXL362 was recorded during this period. Specifically, the data can be divided into three parts: (i) Part 1 is the pre-triggered data which is around 0.28 s in ADXL362 data can be divided into three parts: (i) Part 1 is the pre-triggered data which is around length; (ii) Part 2 is the data that cover the time where the Xnode is initializing; (iii) Part 3 is where the 0.28 s in length; (ii) Part 2 is the data that cover the time where the Xnode is initializing; (iii) Part 3 is ADXL362 data overlaps with the Xnode data. The length of the data in Part 3 is approximately 0.6 s. where the ADXL362 data overlaps with the Xnode data. The length of the data in Part 3 is As shown in Figure 7b, the data obtained from the ADXL362 does not match well with the reference approximately 0.6 s. As shown in Figure 7b, the data obtained from the ADXL362 does not match data from the wired sensors, because the sampling frequency of the ADXL362 was slightly smaller well with the reference data from the wired sensors, because the sampling frequency of the than 100 Hz, which illustrates the first challenge mentioned in the Section 2.3.3. In addition, the time ADXL362 was slightly smaller than 100 Hz, which illustrates the first challenge mentioned in the offset between the Xnode data (blue line in Figure 7a) and the ADXL362 data (red line in Figure 7b) Section 2.3.3. In addition, the time offset between the Xnode data (blue line in Figure 7a) and the must be accurately estimated to fuse these two data streams, which illustrates the second challenge of ADXL362 data (red line in Figure 7b) must be accurately estimated to fuse these two data streams, data fusion. which illustrates the second challenge of data fusion.

Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18, 4480

(a) (a)

11 of 17 11 of 17 11 of 17

(b) (b)

Figure 7. Time history data comparison: (a) Xnode measurement, (b) ADXL362 data buffer. Figure Figure7.7.Time Timehistory historydata datacomparison: comparison:(a) (a)Xnode Xnodemeasurement, measurement,(b) (b)ADXL362 ADXL362data databuffer. buffer.

The datafusion fusion strategydiscussed discussed in theprevious previous sectionisisapplied applied to thetest test data.Figure Figure 8 The Thedata data fusionstrategy strategy discussedininthe the previoussection section is appliedtotothe the testdata. data. Figure8 8 shows a comparisonofoftime time historydata data betweensensor sensor datafrom from wiredsensors sensors andthe the fuseddata data shows showsa acomparison comparison of timehistory history databetween between sensordata data fromwired wired sensorsand and thefused fused data from the demand-based WSS. The excellent agreement demonstrates the ability of the proposed from demand-based WSS. The The excellent agreement demonstrates the ability the proposed strategy fromthethe demand-based WSS. excellent agreement demonstrates the of ability of the proposed strategy to seamlessly capture the structural response to subjected toevent. a sudden event. tostrategy seamlessly capture the structural response subjected a sudden to seamlessly capture the structural response subjected to a sudden event.

(a) (a)

(b) (b)

Figure8.8.Results Resultsofofdata datafusion: fusion:(a) (a)time timehistory historydata, data,(b) (b)zoomed zoomedview view(ADXL362 (ADXL362data). data). Figure Figure 8. Results of data fusion: (a) time history data, (b) zoomed view (ADXL362 data).

3.2. 3.2.Earthquake EarthquakeMonitoring Monitoring 3.2. Earthquake Monitoring As Asa atypical typicalsudden suddenevent, event,ananearthquake earthquakeisistransient transientand andunpredictable, unpredictable,and andthe theconsequences consequences As a typical sudden event,efforts an earthquake is transient andcost-effective unpredictable, and the consequences can be catastrophic. Continuous are required to develop earthquake can be catastrophic. Continuous efforts are required to develop cost-effective earthquakemonitoring monitoring can be catastrophic. Continuous efforts are required to develop cost-effective earthquake monitoring systems mitigate the effect of earthquakes. Demand-based WSS have significant potentialpotential to enableto systemstoto mitigate the effect of earthquakes. Demand-based WSS ahave a significant systems to mitigate the effect of earthquakes. Demand-based WSS have a significant potential to earthquake detectiondetection and rapidand condition assessmentassessment of civil infrastructure, which was which validated enable earthquake rapid condition of civil infrastructure, was enable earthquake detection and rapid condition assessment of civil infrastructure, which was through a lab test in athis validated through labsection. test in this section. validated through a lab in this section. The test setup was thetest same with thatthat in Section 3.1, as in Figure 6. The6.structure modelmodel was The test setup was the same with in Section 3.1,shown as shown in Figure The structure The test setup was the same with that in Section 3.1, as shown in Figure 6. The structure model mounted on a uniaxial shaking shaking table. This shaking can simulate in one horizontal was mounted on a uniaxial table. Thistable shaking table canearthquakes simulate earthquakes in one was mounted on a kg uniaxial shaking table. This shaking table cancm. simulate earthquakes in one direction, driving a 15 mass at 2.5 g with a maximum stroke of ± 7.5 The El Centro earthquake horizontal direction, driving a 15 kg mass at 2.5 g with a maximum stroke of ±7.5 cm. The El Centro horizontal direction, driving a 15 kg mass at 2.5 g with a maximum stroke The of ±7.5 cm. Thethreshold El Centro excitation was generated bygenerated the shaking to represent a sudden event. earthquake excitation was by table the shaking table to represent a suddendetection event. The detection earthquake excitation was generated by the shaking table to represent a sudden event. The detection for the event-based switch in demand-based WSS was configured follows: the onset of the event wasof threshold for the event-based switch in demand-based WSS was as configured as follows: onset threshold for the event-based switch in demand-based WSS was configured as follows: the onset of detected when the acceleration was above 80 mg over 0.02 s, and the end of event was detected when event was detected when the acceleration was above 80 mg over 0.02 s, and the end of event was event was detected when40 themg acceleration was configuration above 80 mg parameters over 0.02 s, are andthe thesame end with of event was the acceleration overbelow 5 s. Other the detected whenwas the below acceleration was 40 mg over 5 s. Other configuration parameters aretest the detected when the acceleration was below 40 mg over 5 s. Other configuration parameters are the insame Section 3.1, such frequencies and filterfrequencies parameters. with the testas insampling Section 3.1, such as sampling and filter parameters. same the test insSection 3.1,time suchhistory as sampling frequencies and parameters. A with segment of 90 recorded is shown in Figure 9a.filter As can be seen in the zoomed view of the time history data (Figure 9b,c), the ground motion started at 10.7 s, but the vibration in the

Sensors 2018, 18, x FOR PEER REVIEW

12 of 17

A segment of 90 s recorded time history is shown in Figure 9a. As can be seen in the zoomed 12 of 17 view of the time history data (Figure 9b,c), the ground motion started at 10.7 s, but the vibration in the beginning was very small. From 11.79 s to 11.80 s, two consecutive acceleration samples obtained by the trigger accelerometer 80 mg. As consecutive a result, theacceleration event-based switchobtained turned by on the the beginning was very small. Fromexceeded 11.79 s to 11.80 s, two samples demand-based WSS immediately started high-fidelity measurement. Thedemand-based acceleration trigger accelerometer exceeded 80 and mg. the As aWSS result, the event-based switch turned on the became smaller than mg after 54.50 s. Approximately 5 s later, the event-based switch stopped the WSS immediately and40 the WSS started high-fidelity measurement. The acceleration became smaller high-fidelity sensing. Figure 9d shows the powerswitch spectral density in the than 40 mg after 54.50 s.Furthermore, Approximately 5 s later, the event-based stopped the (PSD) high-fidelity frequency domain. The excellent agreement between results of inwired sensors domain. and the sensing. Furthermore, Figure 9d shows the power spectral the density (PSD) the frequency demand-based WSS in thebetween both time frequency domain demonstrates the abilityWSS of the The excellent agreement theand results of wired sensors and the demand-based in proposed the both WSSand to detect the earthquake and capturethe theability accurate structural response in a time frequency domain demonstrates of the proposed WSS toduring detectearthquakes the earthquake timely and efficient manner. and capture the accurate structural response during earthquakes in a timely and efficient manner. Sensors 2018, 18, 4480

(a)

(b)

(c)

(d)

Figure9.9.Test Test results of earthquake monitoring: (a) time history data, (b) zoomed of time Figure results of earthquake monitoring: (a) time history data, (b) zoomed view ofview time history history data for event (c) view zoomed viewhistory of timedata history data for event end, (d) power density spectral data for event start, (c) start, zoomed of time for event end, (d) power spectral density (PSD) data. (PSD) data.

3.3. 3.3.Evaluation Evaluationand andDiscussion Discussion In Inthe thelab labtests testsdescribed describedininprevious previoussections, sections,the theattractive attractiveperformance performanceofofthe thedemand-based demand-based WSS was successfully validated to detect sudden events and provide high-quality sensing WSS was successfully validated to detect sudden events and provide high-quality sensingdata datafor for SHM analysis. SHM analysis.

(1) Power Power consumption consumption tests tests showed showed that that the the proposed proposed WSS WSS has has aa current current draw draw of of only only 365 365 µA µA (1) when no no sudden sudden event event occurred, occurred, but but the the power power consumption consumption of of the the original original Xnode Xnode sensor sensor when platformisisapproximately approximately170 170mA mAduring duringsensing. sensing.Considering Consideringthat thatsudden suddenevents eventsare arerare rareand and platform short-duration,most mostof of the demand-based WSS deployed on a isstructure is in short-duration, the the timetime the demand-based WSS deployed on a structure in low-power low-power measurement mode. Therefore, using a 3.7 V DC, 10,000 mAh, rechargeable measurement mode. Therefore, if using a 3.7 Vif DC, 10,000 mAh, rechargeable lithium polymer lithium employing polymer battery, employing WSS can extend the lifetime of always-on battery, the proposed WSSthe canproposed extend the lifetime of always-on monitoring from three days to over three years using a single lithium battery. This feature helps to successfully

Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18, 4480 Sensors 2018, 18, x FOR PEER REVIEW

13 of 17 13 of 17 13 of 17

monitoring from three days to over three years using a single lithium battery. This feature helps to successfully detect occurrence of sudden events minimal power budget in long-term monitoring from threethe days to over three yearsminimal using awith single This feature helps detect the occurrence of sudden events with powerlithium budgetbattery. in long-term monitoring. monitoring. In addition, the current draw in each operation associated with duration for a to thedraw occurrence ofoperation sudden events with minimal powerfor budget in long-term In successfully addition, thedetect current in each associated with duration a demand-based demand-based WSS was determined experimentally and shown in Figure 10, in which thea monitoring. In addition, the current and draw in each operation with duration for WSS was determined experimentally shown in Figure 10, inassociated which the majority component majority component of power consumption is sensing. demand-based WSS was determined experimentally and shown in Figure 10, in which the of power consumption is sensing. (2) The data component obtained from the demand-based WSS high-quality, matching well with the data from majority of power consumption is is sensing. (2) The data obtained from the demand-based WSS is high-quality, matching well with the data from wired piezoelectric accelerometers. In particular, high-fidelity sensing well enables 24-bit sensing (2) The data obtained from the demand-based WSS is high-quality, matching with24-bit the data from wired piezoelectric accelerometers. In particular, high-fidelity sensing enables sensing resolution and over 1 kHz sampling rates. This feature helps to conduct structural condition wired piezoelectric particular, high-fidelity enables 24-bit sensing resolution and overaccelerometers. 1 kHz samplingInrates. This feature helps tosensing conduct structural condition assessments accurately under sudden events. resolution and over 1 kHz sampling rates. This feature helps to conduct structural condition assessments accurately under sudden events. (3) The test results show that, when an event a seamless transition from the low-power assessments accurately under sudden events.occurs, (3) sensing The testtoresults show that, when an event occurs, a seamless transition from thethe low-power high-fidelity measurement is carried out, without losing any data about event. (3) The test results show that, when an event occurs, a seamless transition from the low-power sensing to high-fidelity measurement is carried out, without losing any data about the event. sensing to high-fidelity measurement out, WSS without losing any data about the event. These three features demonstrate that is thecarried proposed satisfies the demands of sudden event These three features demonstrate that the proposed WSS satisfies the demands of sudden monitoring. These three features demonstrate that the proposed WSS satisfies the demands of sudden event event monitoring. monitoring.

Figure 10. Flowchart of event-triggered sensing regarding current draw and duration for each operation. Figure of of event-triggered sensing regarding currentcurrent draw and duration for each operation. Figure 10. 10.Flowchart Flowchart event-triggered sensing regarding draw and duration for each operation. 4. Field Application Application 4. Field

To validate 4. Field Application To further further validate system system performance, performance, aa field field test test was was conducted conducted on on aa steel steel railroad railroad bridge bridge north of Champaign, Illinois. Having vibration data while in-service trains traverse the bridge is northTo offurther Champaign, Illinois. Having vibration data while in-service trains traverse the bridge is validate system performance, a field test was conducted on a steel railroad bridge useful to assess the bridge condition [37]. Train events havehave similar features to sudden naturalnatural events, useful to assess the bridge condition [37]. Train events similar features to sudden north of Champaign, Illinois. Having vibration data while in-service trains traverse the bridge is e.g., unpredictability due to uncertain schedules, butschedules, occur morebut frequently and therefore provide events, unpredictability due totrain uncertain occur more frequently and useful toe.g., assess the bridge condition [37]. Traintrain events have similar features to sudden natural atherefore convenient test platform. A demand-based WSS was deployed on the bottom side of a bridge girder. provide a convenient test platform. A demand-based WSS was deployed on the bottom side events, e.g., unpredictability due to uncertain train schedules, but occur more frequently and Simultaneously, wired sensors, modelwired PCB353B33, were selected as reference and deployed of a bridge girder. Simultaneously, model PCB353B33, weresensors selected reference therefore provide a convenient test platform.sensors, A demand-based WSS was deployed on theas bottom side close to the WSS (see Figure 11). A detection threshold was configured to be the same as the test in sensors and girder. deployed close to the WSS (see sensors, Figure 11). A detection threshold configured to be of a bridge Simultaneously, wired model PCB353B33, were was selected as reference Section avoid saturation, the measurement rangethe of measurement the trigger accelerometer set the same3.2. as To the test insignal Section avoid range of thewas trigger sensors and deployed close to 3.2. the To WSS (see signal Figuresaturation, 11). A detection threshold was configured to be to the maximum value of 8 g. At 10:52:06 a.m. on 7 May 2019, an Amtrak passenger train passed by accelerometer to the maximum value of 8saturation, g. At 10:52:06 a.m. on 7 May 2019, the same as thewas testset in Section 3.2. To avoid signal the measurement range of an theAmtrak trigger the bridge. passenger trainwas passed by the the maximum bridge. accelerometer set to value of 8 g. At 10:52:06 a.m. on 7 May 2019, an Amtrak passenger train passed by the bridge.

Figure 11. Field application of the demand-based WSS.

Figure 11. Field application of the demand-based WSS.

Figure 12a–d shows the raw acceleration data of the bridge in vertical direction. The train came to the bridge at 121 s and left at 128 s. The event had a short duration of 7 s, and it was successfully Figure 12a–d shows the raw acceleration data of the bridge in vertical direction. The train came

to the bridge at 121 s and left at 128 s. The event had a short duration of 7 s, and it was successfully

Sensors 2018, 18, 4480

14 of 17

Sensors 2018, 18, x FOR PEER REVIEW

14 of 17

detected by the demand-based WSS. Figure 12e shows the PSD data. Some slight discrepancies detected by the demand-based WSS. Figure 12e shows the PSD data. Some slight discrepancies between the data from two sensors are possibly due to the different locations of the sensors. In sum, between the data from two sensors are possibly due to the different locations of the sensors. In sum, good agreement can be observed between the wired sensor and the demand-based WSS both in the time good agreement can be observed between the wired sensor and the demand-based WSS both in the and frequency domain, demonstrating that the new WSS can capture the sudden event and obtain time and frequency domain, demonstrating that the new WSS can capture the sudden event and high-fidelity measurement in real applications. obtain high-fidelity measurement in real applications.

(a) 200

300 PCB353B33 Demand-based WSS

200

PCB353B33 Demand-based WSS

100

100 0 0 -100

-100 -200 120.5

121 time, s

121.5

-200 125

125.2

125.4 125.6 time, s

125.8

126

(c)

(b) 40

100 PCB353B33 Demand-based WSS

PCB353B33 Demand-based WSS

20

50

0 0

-20 -50

-40 -60 128

129

130 131 time, s

(d)

132

133

-100

0

10

20

30

40

50

frequency,Hz

(e)

Figure Test results a railroadbridge: bridge:(a) (a)time time history history data, (b) Figure 12. 12. Test results onon a railroad (b) zoomed zoomedview viewofofevent eventstarts starts (ADXL362 data), (c) zoomed view of event data (Xnode data), (d) zoomed view of event ends (ADXL362 data), (c) zoomed view of event data (Xnode data), (d) zoomed view of event ends(Xnode (Xnode data), PSD data. data), (e) (e) PSD data.

Sensors 2018, 18, 4480

15 of 17

5. Conclusions This paper presented the design of demand-based WSS to meet the application requirements of sudden event monitoring. The proposed WSS mainly consist of a unique programmable event-based switch and a powerful high-fidelity WSS platform. In particular, the event-based switch is built on a trigger accelerometer which allows the new WSS to measure the structural response in ultralow-power in long term, so as not to miss sudden events. In addition, the software of event-triggered sensing and data fusion is implemented. The performance of the proposed WSS is evaluated through the lab tests of earthquake monitoring and a field application on a railroad bridge. The test results show that the proposed WSS can continuously monitor structural response with minimal power budget, and hence detect the occurrence of the sudden event with the smallest delay. Besides detecting sudden events, the proposed WSS have the excellent features of high sampling rates and sensing resolution, which finally helps to provide high-quality data in sudden events for rapid condition assessment of civil infrastructure. Moreover, the proposed WSS are powerful and versatile not only for sudden events (e.g., earthquakes), but also for autonomous monitoring of other general events (e.g., bridge/highway overloads). For large-scale structures, one demand-based WSS is not sufficient, and a network of nodes are needed for a meaningful characterization of the structural response. When subjected to a sudden event, each node may be triggered independently to initiate measurement at slightly different times due to varying response levels in the structure. Future work will address the challenges encountered for a network of demand-based WSS under sudden events. For example, one critical issue is to synchronize data from different sensor nodes without introducing delay of event-triggered sensing. Author Contributions: Conceptualization, Y.F.; methodology, Y.F. and T.H.; software, Y.F. and T.H.; validation, Y.F., T.H., K.M. and B.F.S.J.; formal analysis, Y.F.; investigation, Y.F. and T.H.; resources, K.M. and B.F.S.J.; data curation, Y.F.; writing—original draft preparation, Y.F.; writing—review and editing, T.H., K.M., J.R.K., D.Z., B.F.S.J.; visualization, Y.F.; supervision, K.M. and B.F.S.J.; project administration, B.F.S.J. and J.R.K.; funding acquisition, J.R.K. and D.Z. Funding: This research was funded in part by the Nazarbayev University Research Fund under Grant #SOE2017003, ZJU-UIUC Institute Research under Grant #ZJU083650, Federal Railroad Administration under Grant #DTFR53-17-C00007, as well as China Scholarship Council scholarship. Acknowledgments: The authors gratefully acknowledge the support of this research by the Federal Railroad Administration, Canadian National Railway, and Sandro Scola, as well as the China Scholarship Council. Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2.

3.

4. 5. 6.

7.

Garner, A.C.; Huff, W.A. The wreck of Amtrak’s Sunset Limited: News coverage of a mass transport disaster. Disasters 1997, 21, 4–19. [CrossRef] [PubMed] Yamazaki, F.; Katayama, T.; Yoshikawa, Y. On-line damage assessment of city gas networks based on dense earthquake monitoring. In Proceedings of the 5th U.S. National Conference on Earthquake Engineering, Chicago, IL, USA, 10–14 July 1994; Volume 4, pp. 829–837. Yamazaki, F.; Motomura, H.; Hamada, T. Damage assessment of expressway networks in Japan based on seismic monitoring. In Proceedings of the 12th World Conference on Earthquake Engineering, Auckland, New Zealand, 30 January–4 February 2000. Celebi, M. Real-time seismic monitoring of the New Cape Girardeau Bridge and preliminary analyses of recorded data: An overview. Earthq. Spectra 2006, 22, 609–630. [CrossRef] Okada, K.; Nakamura, Y.; Saruta, M. Application of earthquake early warning system to seismic-isolated buildings. J. Disaster Res. 2009, 4, 242–250. [CrossRef] Rice, J.A.; Mechitov, K.; Sim, S.H.; Nagayama, T.; Jang, S.; Kim, R.; Spencer, B.F.; Agha, G.; Fujino, Y. Flexible smart sensor framework for autonomous structural health monitoring. Smart Struct. Syst. 2010, 6, 423–438. [CrossRef] Caicedo, J.M.; Clayton, E.; Dyke, S.J.; Abe, M.; Tokyo, J. Structural health monitoring for large structures using ambient vibrations. In Proceedings of the ICANCEER Conference, Hong Kong, China, 15–20 August 2002.

Sensors 2018, 18, 4480

8.

9.

10.

11.

12.

13. 14. 15. 16. 17. 18.

19.

20.

21.

22. 23. 24.

25. 26. 27. 28.

16 of 17

Cheng, L.; Pakzad, S.N. Agility of wireless sensor networks for earthquake monitoring of bridges. In Proceedings of the 2009 IEEE Sixth International Conference on Networked Sensing Systems (INSS), Pittsburgh, PA, USA, 17–19 June 2009. Guo, S.; Kim, S.M.; Zhu, T.; Gu, Y.; He, T. Correlated flooding in low-duty-cycle wireless sensor networks. In Proceedings of the 2011 19th IEEE International Conference on Network Protocols, Vancouver, BC, Canada, 17–20 October 2011; pp. 383–392. Potenza, F.; Federici, F.; Lepidi, M.; Gattulli, V.; Graziosi, F.; Colarieti, A. Long-term structural monitoring of the damaged Basilica S. Maria di Collemaggio through a low-cost wireless sensor network. J. Civ. Struct. Health Monit. 2015, 5, 655–676. [CrossRef] Miller, T.I.; Spencer, B.F., Jr.; Li, J.; Jo, H. Solar Energy Harvesting and Software Enhancements for Autonomous Wireless Smart Sensor Networks; NSEL Report No. 022; University of Illinois at Urbana-Champaign: Urbana, IL, USA, 2010. Jang, S.; Jo, H.; Cho, S.; Mechitov, K.A.; Rice, J.A.; Sim, S.H.; Jung, H.-J.; Yun, C.-B.; Spencer, B.F., Jr.; Agha, G. Structural health monitoring of a cable-stayed bridge using smart sensor technology: Deployment and evaluation. Smart Struct. Syst. 2010, 6, 439–459. [CrossRef] Park, J.W.; Jung, H.J.; Jo, H.; Spencer, B.F., Jr. Feasibility study of micro wind turbines for powering wireless sensors in a cable-stayed bridge. Energies 2012, 5, 3450–3464. [CrossRef] Lu, X.; Wang, P.; Niyato, D.; Han, Z. Resource allocation in wireless networks with RF energy harvesting and transfer. IEEE Netw. 2015, 29, 68–75. [CrossRef] Lu, X.; Wang, P.; Niyato, D.; Kim, D.I.; Han, Z. Wireless networks with RF energy harvesting: A contemporary survey. IEEE Commun. Surv. Tutor. 2015, 17, 757–789. [CrossRef] Jalsan, K.E.; Soman, R.N.; Flouri, K.; Kyriakides, M.A.; Feltrin, G.; Onoufriou, T. Layout optimization of wireless sensor networks for structural health monitoring. Smart Struct. Syst. 2014, 14, 39–54. [CrossRef] Rault, T.; Bouabdallah, A.; Challal, Y. Energy efficiency in wireless sensor networks: A top-down survey. Comput. Netw. 2014, 67, 104–122. [CrossRef] Lu, G.; De, D.; Xu, M.; Song, W.Z.; Cao, J. TelosW: Enabling ultra-low power wake-on sensor network. In Proceedings of the 2010 IEEE Seventh International Conference on Networked Sensing Systems (INSS), Kassel, Germany, 15–18 June 2010. Polastre, J.; Szewczyk, R.; Culler, D. Telos: Enabling ultra-low power wireless research. In Proceedings of the 4th International Symposium on Information Processing in Sensor Networks, Boise, ID, USA, 15 April 2005; p. 48. Sutton, F.; Da Forno, R.; Gschwend, D.; Lim, R.; Gsell, T.; Beutel, J.; Thiele, L. A Heterogeneous System Architecture for Event-triggered Wireless Sensing. In Proceedings of the 2016 15th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN), Vienna, Austria, 11–14 April 2016. Bischoff, R.; Meyer, J.; Enochsson, O.; Feltrin, G.; Elfgren, L. Event-based strain monitoring on a railway bridge with a wireless sensor network. In Proceedings of the 4th International Conference on Structural Health Monitoring of Intelligent Infrastructure, Zurich, Switzerland, 22–24 July 2009; Volume 2224, p. 7482. Liu, Y.; Voigt, T.; Wirström, N.; Höglund, J. ECOVIBE: On-Demand Sensing for Railway Bridge Structural Health Monitoring. IEEE Internet Things J. 2018. [CrossRef] Popovic, N.; Feltrin, G.; Jalsan, K.E.; Wojtera, M. Event-driven strain cycle monitoring of railway bridges using a wireless sensor network with sentinel nodes. Struct. Control Health Monit. 2017, 24, e1934. [CrossRef] Hung, S.L.; Ding, J.T.; Lu, Y.C. Developing an energy-efficient and low-delay wake-up wireless sensor network-based structural health monitoring system using on-site earthquake early warning system and wake-on radio. J. Civ. Struct. Health Monit. 2018, 1–13. [CrossRef] Dorvasha, S.; Lib, X.; Pakzada, S.; Chengb, L. Network architecture design of an agile sensing system with sandwich wireless sensor nodes. Proc. SPIE 2012, 8345, 83450H. Analog Devices. ADXL362 Datasheet. 2016. Available online: http://www.analog.com/media/en/ technical-documentation/data-sheets/ADXL362.pdf (accessed on 18 November 2018). Analog Devices. ADXL372 Datasheet. 2017. Available online: http://www.analog.com/media/en/ technical-documentation/data-sheets/adxl372.pdf (accessed on 18 November 2018). STMicroelectronics. LIS3DSH Datasheet. 2017. Available online: https://www.st.com/resource/en/ datasheet/lis3dsh.pdf (accessed on 18 November 2018).

Sensors 2018, 18, 4480

29. 30.

31. 32. 33. 34.

35.

36. 37.

17 of 17

InvenSense. MPU-6050 Datasheet. 2013. Available online: https://www.invensense.com/products/motiontracking/6-axis/mpu-6050/ (accessed on 18 November 2018). Kane, M.; Zhu, D.; Hirose, M.; Dong, X.; Winter, B.; Häckell, M.; Lynch, J.P.; Wang, Y.; Swartz, A. Development of an extensible dual-core wireless sensing node for cyber-physical systems. In Proceedings of the Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2014, San Diego, CA, USA, 9–13 March 2014; Volume 9061, p. 90611U. BeanAir. AX-3D Datasheet. 2016. Available online: http://www.beanair.com/wa_files/Datasheet-WirelessAccelerometer-BeanDevice-AX-3D.pdf (accessed on 18 November 2018). Libelium. Waspmote v15 Datasheet. 2016. Available online: http://www.libelium.com/downloads/ documentation/waspmote_datasheet.pdf (accessed on 18 November 2018). LORD Sensing. G-Link-200 Datasheet. 2017. Available online: http://www.microstrain.com/sites/default/ files/g-link-200_datasheet_8400-0102_2_0.pdf (accessed on 18 November 2018). Spencer, B.F., Jr.; Park, J.W.; Mechitov, K.A.; Jo, H.; Agha, G. Next Generation Wireless Smart Sensors toward Sustainable Civil Infrastructure. In Proceedings of the Sustainable Civil Engineering Structures and Construction Materials (SCESCM 2016), Bali, Indonesia, 5–7 September 2016. Fu, Y.G.; Mechitov, K.A.; Hoskere, V.; Spencer, B.F., Jr. Development of RTOS-based wireless SHM system: Benefits in applications. In Proceedings of the International Conference on Smart Infrastructure and Construction, Cambridge, UK, 27–29 June 2016. Nagayama, T.; Spencer, B.F., Jr. Structural Health Monitoring Using Smart Sensors; Newmark Structural Engineering Laboratory, University of Illinois at Urbana-Champaign: Urbana, IL, USA, 2007. Moreu, F.; Kim, R.E.; Spencer, B.F., Jr. Railroad bridge monitoring using wireless smart sensors. Struct. Control Health Monit. 2017, 24, e1863. [CrossRef] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).