A Study on the Data Compression Technology-Based Intelligent Data Acquisition (IDAQ) System for Structural Health Monitoring of Civil Structures.

sensors Article

A Study on the Data Compression Technology-Based Intelligent Data Acquisition (IDAQ) System for Structural Health Monitoring of Civil Structures Gwanghee Heo

and Joonryong Jeon *

Department of Civil and Environment Engineering, Konyang University, 121 Daehak-ro, Nonsan, Chungnam 320-711, Korea; [email protected] * Correspondence: [email protected]; Tel.: +82-41-730-5312 Received: 1 June 2017; Accepted: 10 July 2017; Published: 12 July 2017

Abstract: In this paper, a data compression technology-based intelligent data acquisition (IDAQ) system was developed for structural health monitoring of civil structures, and its validity was tested using random signals (El-Centro seismic waveform). The IDAQ system was structured to include a high-performance CPU with large dynamic memory for multi-input and output in a radio frequency (RF) manner. In addition, the embedded software technology (EST) has been applied to it to implement diverse logics needed in the process of acquiring, processing and transmitting data. In order to utilize IDAQ system for the structural health monitoring of civil structures, this study developed an artificial filter bank by which structural dynamic responses (acceleration) were efficiently acquired, and also optimized it on the random El-Centro seismic waveform. All techniques developed in this study have been embedded to our system. The data compression technology-based IDAQ system was proven valid in acquiring valid signals in a compressed size. Keywords: data compression technology; embedded software technology; artificial filter bank; band-pass filter optimizing algorithm; peak-picking algorithm; reconstruction error; compressive ratio; spectrum error; structural health monitoring

1. Introduction In general, infrastructure is always exposed to unexpected weather conditions and consequent possibility of damage and strain as well as natural aging, and so its life span decreases [1,2]. As the size of structures is recently becoming larger, casualties and property damage could be accordingly larger in the event of an accident. In order to be well prepared for such threats and secure structural safety, there have been many studies on structural health monitoring (SHM) which makes it possible to inspect structural conditions on a regular basis and detect damage early for an effective preparation against an unexpected disaster or situation [3–8]. On a long-term basis, SHM technology can reduce the time and efforts required for the maintenance of structures [9]. For these reasons, it has recently been introduced to the construction of large bridges, skyscrapers and other major infrastructure. Examples of the application of the SHM technology to bridge construction include the Great Belt East Bridge [10] in Denmark, the Bill Emerson Memorial Bridge [11] in the U.S. and the Hakucho Bridge [12] in Japan. Examples of the application of the technology to skyscrapers are a 17-story building at the UCLA in the U.S. [13] and Republic Plaza [14] in Singapore. In order to inspect structural conditions and early detect any damage using SHM technology, a decent SHM system must be developed. According to previous studies, most of SHM systems use wired sensor and measuring systems to secure a vast number of structural response data from sensors of various types [15] The wired SHM system, using cables for transmitting structural responses from sensors, is advantageous in acquiring data in a stable manner. Since data loggers are expensive, Sensors 2017, 17, 1620; doi:10.3390/s17071620

www.mdpi.com/journal/sensors

Sensors 2017, 17, 1620

2 of 22

however, the system requires a large amount of initial cost for its installation. In addition, since power equipment and sensor location are mostly determined at the stage of design and construction, it is not easy to adjust or change the system later. In the case when new sensors need to be installed or old ones relocated, they all must be re-cabled, making the wiring complicated. If sensors are installed at a distance, then noise should also be considered. To overcome these practical and economic limitations of conventional wired SHM systems, Spencer [16] and Lynch [17] proposed a wireless SHM system based on a wireless sensor network. For example, Nagayama [18], Rice et al. [19] and Park [20] developed the wireless sensor node ‘Imote2’ through ANCRiSST and applied it for SHM and assessment of the structural state of a real bridge. In addition, Kurata et al. [21] developed ‘Narada’ using solar panels and sub-networks and applied it to New Carquinez Bridge in the U.S. for the purpose of SHM. Those examples proved that the wireless SHM system was effective in acquiring structural responses from multiple points on a large structure using wireless sensor nodes and wireless networks. Despite its various obvious advantages, conventional wireless SHM systems are not easy to restructure for either extension or simple change because they are designed and developed with limited resources to carry out only the sensor node-targeted functions. As each sensor node has an independent controller, it also requires a lot of time and efforts if a measuring logic is updated (micro-programmed). Its capacity could be additionally limited in processing consecutive measurement data on a real time basis. To make SHM efficient, therefore, there should be a user-friendly, high-performance SHM system which can: (i) acquire and process a large amount of diverse data on a real-time basis; (ii) easily edit measuring logic quickly and accurately and expand its functions and (iii) provide system safety and data quality up to the level of the commonly used (wired) measuring system even with the proven RF method. For these purposes, a software design technology-adopted, embedded software technology (EST)-based data acquisition (DAQ) system could be an alternative, overcoming the limitations of the conventional hardware-based wired and wireless measuring systems. The EST-based DAQ system can adopt diverse RF systems in a flexible manner and easily expand multi-input/output channel (I/O) for multiple kinds and amounts of measurement. With the memory as large as a high-performance personal computer and with a high-performance CPU included, it is effective in acquiring, computing, processing and storing large data on a real-time basis. If necessary, it can be converted into a standalone system and operated as a sub-SHM system. Thanks to the adoption of the EST, in particular, the system can design and develop diverse logics (e.g., function, algorithm, etc.) which are needed in the acquisition, process and transmission of structural response signals and embed it within the DAQ system. Furthermore, the embedded logic can be operated under the real-time operating systems (RTOS). After all, this kind of EST-based DAQ system can effectively improve the weakness of the conventional hardware-based wired and wireless SHM system and considerably reduce time and efforts needed to develop a user-centered, high-performance SHM system which is directly applicable to actual structure. Meanwhile, even though the EST-based DAQ system is capable of acquiring and storing a large amount of data, bottlenecks, data losses, or data delays can occur because RF protocols (e.g., Bluetooth, Zigbee, Wi-Fi, etc.) are technologically limited in receiving and transmitting data. The RF for a real-time SHM is supposed to work well under long distance, high speed and multi-channel conditions, but it is not so advanced yet. To develop a SHM system by overcoming these RF limitations, Peckens et al. [22] proposed a data compressing technology, a hardware filter bank imitating the human hearing mechanism. However, the hardware filter bank designed and developed for data compression for one particular infrastructure would not be valid anymore when the SHM system is applied to other structures, so a new hardware filter bank would have to be designed and fabricated [23,24]. When the bandwidth of dynamic responses (acceleration) changes in another SHM target structure, the SHM system with this kind of hardware filter bank would become inefficient from a practical standpoint. Such a weakness can be effectively improved through a design and development of a software type of filter bank which can edit/change its logic in a flexible manner. Especially, a digital type of software filter bank can be embedded into the DAQ system by coding the conventional hardware filter bank’s

Sensors 2017, 17, 1620

3 of 22

logic into high-level language, and thus it is basically different from the an analogy type of H/W filter bank in the wireless SHM in terms of filter band design method. In this study, a software type of filter bank named ‘artificial filter bank (AFB)’ was developed for data compression. The AFB can selectively compress only the acceleration signals. Then, the developed AFB was completely embedded into the EST-based DAQ system, and an RF-based SHM system was developed and named an ‘intelligent data acquisition (IDAQ)’ system. To assess its validity, finally tests were conducted based on a random signal (the El-Centro seismic waveform). From the tests, it was found that the IDAQ system having AFB was valid in acquiring dynamic responses (only acceleration) for the SHM in a compressed size, successfully substituting the conventional hardware-based wired and wireless SHM system. In addition, it was confirmed that the system would be available as a user-centered SHM system because the measuring logic requested in a software design system can be easily and quickly developed and embedded. 2. The IDAQ System for Structural Health Monitoring For efficient SHM on structures, this study develops an IDAQ system which is distinguished from a conventional hardware-based wired and wireless SHM system in the following features and advantages: First, this study adopts the leading EST ‘Windows Embedded Standard 7 (WES7)’ and the TROS ‘LabVIEW Real-Time 2012 (LabRT2012)’ as shown in Figure 1a. The adoption of WES7 makes it possible to code the requested logics and functions in high-level language and embed them into the DAQs system. With the adoption of LabRT2012, the DAQ system can likewise be operated in a standalone manner. Then, the targeted structural response can be acquired and processed on a real-time basis, using the diverse logics and functions coded in the LabVIEW. After all, the introduction of EST and RTOS can innovatively reduce time and efforts needed to establish and operate the system and to develop and implement the measuring logics and functions needed for efficient SHM. Second, this study employs a high-performing CPU such as 32 GB memory, 2 GB RAM and Intel core i7 as shown in Figure 1b. Thanks to the introduction of this kind of high-performing PC-level controller, a user is able to efficiently save, compute and process large measurement datasets for SHM. After all, optimum measuring conditions can be provided for efficient SHM. Third, this study adopts a Wi-Fi communication module which enables high-speed mobile communication as shown in Figure 1c. The introduction of Wi-Fi-based access points (APs) makes it possible to transmit large data at high speed and expand to multiple channels. If necessary, a wireless network could be configured for high-speed mobile communication even at a distance by adding low cost APs. In addition, practicality-considered RF communication functions are secured to enable closed-operation (e.g., feedback vibration control, etc.) based on structural response according to users’ purpose and demand through two-way communication. Mobile communication-aimed AP is also configured as an independent hardware from the IDAQ system so that the AP of other wireless communication systems (e.g., Zigbee, Bluetooth, etc.) can be easily changed or adopted depending on user demand. Fourth, this study takes in a Multi-I/O by which it can expand channels according to a type of sensor, as shown in Figure 1d. With the introduction of Multi-I/O, various data needed for SHM can be measured and operated in an integrated manner using the IDAQ system. In particular, it fundamentally solves the problems of conventional SHM—high initial cost for system development and data synchronization. Lastly, this study adopts a data compressing technology of dynamic responses, ‘AFB’, as illustrated in Figure 1e. Here, the AFB is developed to acquire effective dynamic responses in a compressed size and wholly embedded into the IDAQ system. The introduction of this kind of AFB helps the IDAQ system effectively reduce a large volume of dynamic responses needed for SHM. In addition, more efficient data transmission is expected than in the RF system. Therefore, this system can be a great alternative for the RF-based real-time SHM. As stated above, the IDAQ system is compared to

Sensors 2017, 17, 1620

4 of 22

Lastly, this study adopts a data compressing technology of dynamic responses, ‘AFB’, as illustrated in Figure 1e. Here, the AFB is developed to acquire effective dynamic responses in a compressed size and wholly embedded into the IDAQ system. The introduction of this kind of AFB Sensors 2017,IDAQ 17, 1620system effectively reduce a large volume of dynamic responses needed for SHM. 4 of In 22 helps the addition, more efficient data transmission is expected than in the RF system. Therefore, this system can be a great alternative the RF-based real-time As can stated above, the IDAQ system as is the conventional wired andfor wireless SHM system. Its SHM. structure be represented in a diagram compared to the conventional wired and wireless SHM system. Its structure can be represented in a shown below: diagram as shown below:

Figure Figure 1. 1. Design Design of of the the IDAQ IDAQ system. system.

3. Filter Bank Bank (AFB) (AFB) for for the the IDAQ IDAQ System System 3. An An Artificial Artificial Filter In efficiency. AsAs described in Section 2 above, the In this this study, study, the theIDAQ IDAQsystem systemisisdeveloped developedforfor efficiency. described in Section 2 above, IDAQ system features a CPU and memory capacity asashigh the IDAQ system features a CPU and memory capacity highasasthe thelatest latesthigh-performance high-performance personal personal computer allows. It also has the following: WES7 based on EST, LabRT2012 multi computer allows. It also has the following: WES7 based on EST, LabRT2012 based based on on RTOS, RTOS, multi I/O for diverse/multiple measurement in an integrated manner, Wi-Fi AP for high-speed mobile I/O for diverse/multiple measurement in an integrated manner, Wi-Fi AP for high-speed mobile communication, AFB(data (datacompression compressiontechnology technology dynamic responses (only acceleration)). In communication, AFB of of dynamic responses (only acceleration)). In this this section, the mechanism of AFB developed for the IDAQ system is described. The AFB can section, the mechanism of AFB developed for the IDAQ system is described. The AFB can selectively selectively compress only the acceleration signal. AFB’s specific technologies band-pass filter compress only the acceleration signal. AFB’s specific technologies band-pass filter optimizing algorithm optimizing algorithm (BOA) and (PPA) peak-picking algorithm (PPA) The are index also briefly The (BOA) and peak-picking algorithm are also briefly explained. such asexplained. reconstruction index such as reconstruction error (RE), compressive ratio (CR) and spectrum error (SE) by which error (RE), compressive ratio (CR) and spectrum error (SE) by which the AFB’s performance is assessed the AFB’s performance is assessed are also described. are also described. 3.1. 3.1. Artificial Artificial Filter Filter Bank Bank (AFB) (AFB) Mechanism Mechanism In the field field of of signal signal processing, processing, a a filter filter bank bank is is defined defined as as aa special special filtering filtering arrangement arrangement that that In the filters certain frequency elements (information) of interest only based on input signal standards. It filters certain frequency elements (information) of interest only based on input signal standards. handles a series and It handles a seriesofofprocesses processessuch suchasasclassifying classifyinginput inputsignals signals by by frequency, frequency, reconfiguring reconfiguring and printing them. That is, it arranges input signals into each appropriate area of frequency and outputs printing them. That is, it arranges input signals into each appropriate area of frequency and outputs the rearranged ones. ones. The rearrangement means ‘decomposition of signals’ by by decision decision the rearranged The process process of of rearrangement means aa ‘decomposition of signals’ of each filter, filter, and and that that of of output output becomes becomes aa ‘synthesis ‘synthesis of of signals’ signals’ occurring occurring during during filtering. filtering. Such of each Such aa filter bank can be differently designed depending on purposes and frequency levels. As the number filter bank can be differently designed depending on purposes and frequency levels. As the number of band-pass filters filter tank tank increases increases and they become become denser, filtering signals signals get more of band-pass filters in in the the filter and they denser, filtering get more accurate than the theinput inputsignal. signal.However, However,data data operating and processing efficiency declines when accurate than operating and processing efficiency declines when the the number of band-pass filters increases. Therefore, there is a need to optimize the filter bank. To number of band-pass filters increases. Therefore, there is a need to optimize the filter bank. To optimize optimize a filter bank, primary design factors such as number of band-pass filters, bandwidth and a filter bank, primary design factors such as number of band-pass filters, bandwidth and spacing spacing should be considered. Here, the optimization filter is to determine thedesign three factors design should be considered. Here, the optimization of filter of bank is bank to determine the three factors to build the best reproduction capability based on the targeted input signal. For to build the best reproduction capability based on the targeted input signal. For this, numerical this, and numerical and repetitive operations aredifferent requiredcombinations under different combinations the three design repetitive operations are required under of the three design of factors. Meanwhile, factors. Meanwhile, the signals (Figure 2b) decomposed passing through the filter band-pass filter in the signals (Figure 2b) decomposed by passing through thebyband-pass filter in the bank and also the filter bank and also its synthesized and reconstructed ones (Figure 2c) include only certain its synthesized and reconstructed ones (Figure 2c) include only certain frequency information shown as in Figure 2a. Therefore, even though the filter bank can be valid in data acquisition, it will have only the same size of data with the same sampling spacing as the input signals because of the specific features particular to band-pass filters. In other words, although certain reconstructed signals may be selectively better acquired than input signals, the size of the acquired data remains the same. After all,

Sensors 2017, 17, 1620

5 of 22

frequency information shown as in Figure 2a. Therefore, even though the filter bank can be valid in of 22 it will have only the same size of data with the same sampling spacing as the5input signals because of the specific features particular to band-pass filters. In other words, although reconstructed be selectively better acquired than input signals, size of the itcertain turns out to be not so signals efficientmay in terms of RF communication-based transmission andthe management acquired data remains the same. After all, it turns out to be not so efficient in terms of RF of acquired data. communication-based transmission anddynamic management of acquired data.using the limited capacity of Therefore, to efficiently acquire the response of structure Therefore, to efficiently acquire the dynamic response of structure usingconcern the limited capacity of RF communication, data compressing technology is required. What is of most in developing RF communication, data compressing technology is required. What is of most concern a data compression technology for dynamic responses is to reflect frequency information just likein developing data Based compression technology for dynamic responses this is to reflect frequency the filter bankaabove. on several design factors and considerations, study developed an information just like the filter bank above. Based on several design factors and considerations, this AFB by combining the BOA aimed to optimize the filter bank with the PPA designed to compress study developedsignals an AFB by combining BOA aimed optimize of thethe filter with the the reconstructed obtained from thethe filter bank. The to mechanism AFBbank developed for PPA the designed to compress the reconstructed signals obtained from the filter bank. The mechanism of the IDAQ system is shown in Figure 2 below: AFB developed for the IDAQ system is shown in Figure 2 below: Sensors 2017, 17, 1620 data acquisition,

Figure Figure2.2.Mechanism MechanismofofAFB AFBfor forIDAQ IDAQsystem system: :(a) (a)raw rawsignal; signal;(b) (b)filter filterbank bankpath; path;(c) (c)reconstruction reconstruction signal; signal;(d) (d)compression compressiondata data

3.2. 3.2.Band-Pass Band-PassFilter FilterOptimization OptimizationAlgoritnm Algoritnmfor forAFB AFB The signals in Figure 2c can expressed in a form combining individual filtering Thereconstructed reconstructed signals in Figure 2c be can be expressed in aof form of combining individual signals which the which filter bank passes, as stated in as Equation (1)Equation below. Then, the filter bankthe selects filtering signals the filter bank passes, stated in (1) below. Then, filtersome bank data of certain frequency among the original (raw) diverse frequency, and classifies them selects some data of certain frequency among thesignals originalof(raw) signals of diverse frequency, and into several parts for filtering and output. Here, in order that the reconstructed signals sufficiently classifies them into several parts for filtering and output. Here, in order that the reconstructed reflect initial input signals, is required to design therequired filter bank in an optimum With a goal signals sufficiently reflect it initial input signals, it is to design the filtermanner. bank in an optimum ofmanner. optimizing thea filter this studythe develops a BOAthis to determine three design (number With goal bank, of optimizing filter bank, study develops a BOAconditions to determine three ofdesign band-pass filters,(number bandwidth and spacing) in the band-passand filter. conditions of band-pass filters, bandwidth spacing) in the band-pass filter. The calculation process can becan divided the following steps: First, TheBOA’s BOA’s calculation process be into divided into thefour following fourreconstruction steps: First, errors are comparatively numbered with the assumed with number band-pass filters based on the initial reconstruction errors are comparatively numbered theofassumed number of band-pass filters input the number band-pass filters is set as the optimum at the point basedsignal. on theSecond, initial input signal.ofSecond, the number of band-pass filters condition is set as the optimum where the slope of the reconstruction is inflected. Third, a reconstruction is calculated condition at the point where theerror slope of the reconstruction error iserror inflected. Third,bya changing the bandwidth and the central frequency on and the basis of the number of interval, band-pass reconstruction error is calculated by changing the interval, bandwidth the central frequency on filters. Fourth, bandwidth andFourth, mean frequency which have the minimum the basis of theoptimal numberisofthe band-pass filters. optimal is interval the bandwidth and mean frequency reconstruction interval whicherror. have the minimum reconstruction error. InInthis thethe RE,RE, which is required for BOA’s calculation process to optimize the filter thisstudy, study, which is required for BOA’s calculation process to optimize thebank, filter isbank, defined as shownasinshown Equation (2) from the in concept Figure 3 in below. The3 RE defined Equation is defined in Equation (2)concept from the Figure below. TheinRE defined(2) in isEquation represented relative difference of absolute values between original and between reconstructed signals. (2) by is the represented by the relative difference of absolute values original and As the RE is close to ‘0’,As it means original signal is successfully and further, simulated the filter reconstructed signals. the RE the is close to ‘0’, it means the originalsimulated signal is successfully and further, is judged to have excellent bank is judgedthe to filter have bank excellent reconstruction ability: reconstruction ability: N

u(t) ≈ y(t) =

∑ yi ( t )

i =1

(1)

Sensors 2017, 17, 1620

Sensors 2017, 17, 1620

6 of 22

( )≈ ( )=

( )

(1)22 6 of

RT RT ||δ |/| /|u((t)|)| )| ||u((t))−−y((t)|)|/| |/|u((t)| (2) 0 = 0 i RERE== = (2) T T where, u((t))isisanan original signal response while ( ) Tand T represent a reconstructed where, original signal by by response timetime while y(t) and represent a reconstructed signal signal by response time and alength total length by response respectively. though Peckens by response time and a total (s) by (s) response time, time, respectively. Even Even though Peckens and and Lynch [22] lused − for RE, it is a scale for the comparison of the RE’s total energy Lynch [22] used − norm for RE, it is a scale for the comparison of the RE’s total energy against the 2 against the original signal. To clearly express the reconstruction effect of the time of the original signal. To clearly express the reconstruction effect of the time response of theresponse reconstructed reconstructed including study states the reconstruction signal’ssignals error signal includingsignal peak values, this peak study values, states thethis reconstruction signal’s error against original against original on absolute scale.signals on absolute scale.

Figure Figure 3. 3. Concept Concept of of reconstruction reconstruction error error (RE). (RE).

3.3. 3.3. Peak-Picking Algorithm Based Based on on the the Central Central Difference Difference Method Method As As mentioned mentioned above, above, to efficiently acquire the dynamic dynamic responses responses of of structure structure using using RF RF communication limited performance, data data compression technology is required besides communicationtechnology technologywith with limited performance, compression technology is required selective sorting ofsorting data. For this, this PPA to reduce thetosize of thethe acquired besides selective of data. Forstudy this, develops this studya develops a PPA reduce size ofdata the acquired data while including information of certain frequencies of to interest pertaining to the while including information of certain frequencies of interest pertaining the characteristics of the characteristics of the The developed PPA derives peak values from dynamic responses. Thedynamic developedresponses. PPA derives peak values from reconstructed signals, not from reconstructed signals, original signals. the calculation for data compression original signals. Then, not the from calculation process forThen, data compression canprocess be dramatically reduced by can be dramatically reduced peak signal, valuesabased on the reconstructed a set of of deriving peak values based on by the deriving reconstructed set of filtering signals, not thesignal, peak value filtering signals, the filtered signal.not the peak value of the filtered signal. The PPA’s PPA’scalculation calculation process be summarized as follows: first, calculated reconstructed The process cancan be summarized as follows: first, calculated reconstructed signal signal readthe from the optimized filter bank. Second, the reconstructed signal is classified three is readisfrom optimized filter bank. Second, the reconstructed signal is classified intointo three (3) (3) data groups. Third, group’s derived function is calculated on data threeindata each data groups. Third, eacheach group’s derived function is calculated basedbased on three eachingroup. group. Fourth, peak(time values and measuring data) are the of slope of Fourth, peak values and(time measuring data) are derived by derived assessingby theassessing slope (sign) each(sign) group’s each group’s derived function. derived function. In this this study, study, aa central central difference difference method method isis used used as as aa way way to extract peak values while PPA PPA is is In calculating. Here, Here, the thedifference differencemethod methodis is classified backward difference, central difference calculating. classified intointo backward difference, central difference and and forward difference under the assumption thatrelation the relation among neighboring is linear. If forward difference under the assumption that the among neighboring data data is linear. If the the neighboring is spaced suitably enough to the meet the assumption, a derived function with the neighboring data data is spaced suitably enough to meet assumption, a derived function with the lowest lowest can be calculated using difference the central difference method. 4 above the error canerror be calculated using the central method. Figure 4 above Figure reveals the centralreveals difference central difference method’s diagram, and a be derived function can Equation be calculated method’s conceptual diagram,conceptual and a derived function can calculated through (3): through Equation (3): f ( x (i + 1)) − f ( x (i − 1)) f 0 ( x (i )) = (3) x((i + + 1) 1) −−x ((i −( 1− ) 1)) (3) () = ( + 1) − ( − 1) Based on the size of the reconstructed data determined through the Equation (1) above, the effects Based on the size of the reconstructed data determined through the Equation (1) above, the of data compression through PPA can be defined by comparing a relative data size with the extracted effects of data compression through PPA can be defined by comparing a relative data size with the peak value. For this, this study defines data CR as stated in Equation (4) below: extracted peak value. For this, this study defines data CR as stated in Equation (4) below: CR =

NS0 − NSC NS0

(4)

Sensors 2017, 17, 1620

7 of 22

Sensors 2017, 17, 1620

CR =

−

7 of 22

(4)

where, thenumber number of compressive data NS while to the number of where, NSC is the of compressive signalsignal data while the number of reconstructed 0 refers to refers reconstructed the CR gets to ‘0’, the compression effectthe increases. the signal data. Ifsignal the CRdata. gets If closer to ‘0’, thecloser compression effect increases. Here, defined Here, CR is used defined CR is used for the assessment of the compression effects against reconstructed signal. for the assessment of compression effects against reconstructed signal.

Figure ofof central difference Figure4.4.Concept Concept central differencemethod. method.

4. Artificial Filter Bank (AFB) Optimization 4. Artificial Filter Bank (AFB) Optimization The previous chapter defined AFB as a data compressing technology of dynamic responses to The previous chapter defined AFB as a data compressing technology of dynamic responses acquire the efficient dynamic responses (only acceleration) of structures using the IDAQ system by to acquire the efficient dynamic responses (only acceleration) of structures using the IDAQ system combing both BOA and PPA. In this chapter, the AFB is optimized based on a random signal (the by combing both BOA and PPA. In this chapter, the AFB is optimized based on a random signal El-Centro seismic waveform) which is commonly used in the construction industry to assess the (the El-Centro seismic waveform) which is commonly used in the construction industry to assess the IDAQ system’s applicability for SHM on large architectural structures. IDAQ system’s applicability for SHM on large architectural structures. 4.1. 4.1.Reference ReferenceSignal SignalforforAFB AFBOptimization Optimization ItItisis well well known knownthat that El-Centro, Kobe and Northridge the random leading seismic randomwaveforms seismic El-Centro, Kobe and Northridge are theare leading waveforms in the construction industry. Theyanrepresent an circumstance unexpected circumstance which used in theused construction industry. They represent unexpected which may be occur may be occur when designing, constructing and managing structures. To assess the applicability of when designing, constructing and managing structures. To assess the applicability of the IDAQ system the system forstructures, SHM intotherefore, large structures, therefore, the ‘El-Centro’ seismic waveform was forIDAQ SHM into large the ‘El-Centro’ seismic waveform was chosen as an original chosen as an original signal for the optimization of AFB among the random seismic waveforms signal for the optimization of AFB among the random seismic waveforms mentioned above. Figure 5 mentioned above. Figure 5 below the El-Centro seismic waveform LA02 (1940, below shows the El-Centro seismicshows waveform (SAC Name: LA02 (1940, SE))(SAC usedName: as an original signal, SE)) used as an original signal, is expressed inAccording time and frequency responses. to which is expressed in time and which frequency responses. to Figure 5a, a seismicAccording situation lasts Figure 5a, a seismic situation lasts for a short period of time (approximately one minute in the case for a short period of time (approximately one minute in the case of the El-Centro event). In Figure 5b, ofdiverse the El-Centro event). In Figure 5b, diverse frequency factors are distributed acrossthe thefrequency frequencyof frequency factors are distributed across the frequency bands. In particular, bands. In is particular, the frequency of 10 interest is concentrated in less than 10 Here, the seismic interest concentrated in less than Hz. Here, if the seismic situation in Hz. Figure 5 isifassumed, the situation in Figure 5 is assumed, the dynamic responses of the target structure are measured as a dynamic responses of the target structure are measured as a type which includes the target structure’s type includes target structure’sseismic own frequency infrequency the conventional ownwhich frequency factorsthe in the conventional waveform’sfactors random factors. seismic waveform’s random frequency factors. For SHM on structures, after all, the target structure’s original frequency factors in the random For SHM on structures, the target structure’s original frequency in the random frequency factors should beafter fullyall, discovered. Because large structures in thefactors construction industry frequency factors shouldflexible be fully discovered. Because large the construction usually have relatively behavioral characteristics, the structures range of theintarget mode neededindustry for SHM usually have relatively flexible behavioral characteristics, the range of the target mode needed for can be limited to a certain range of frequency (e.g., below 10 Hz). From this perspective, the El-Centro SHM can be limited to a certain range of frequency (e.g., below 10 Hz). From this perspective, the seismic waveform selected as an original signal could be used as a reference response which can El-Centro seismic selected as of anfrequency original signal could be used as aUnder reference response wholly express thewaveform large structure’s rage concerned (below 10 Hz). the assumption which can wholly express the large structure’s rage of frequency concerned (below 10 Hz). that an El-Centro seismic situation has these frequency characteristics, this study designs theUnder AFB in the that an El-Centro situation these frequency thisT:study an assumption optimum manner based on theseismic El-Centro seismichas waveform (total timecharacteristics, length: 50 s, delta 0.02 s, total frame length: 2500).

Sensors 2017, 17, 1620

8 of 22

designs the Sensors 2017, 17, AFB 1620

in an optimum manner based on the El-Centro seismic waveform (total 8time of 22 length: 50 s, delta T: 0.02 s, total frame length: 2500). 4 3

Acceleration(g)

2 1 0

-1 -2 -3 0

5

10

15

20

25 Time(sec)

30

35

40

45

50

(a) 3 2.5

Mag.

2 1.5 1 0.5 0 0

5

10

15

20

25

Freq.(Hz)

(b) Figure 5. El-Centro seismic waveform for reference signal of AFB: (a) Time domain; (b) Frequency Figure 5. El-Centro seismic waveform for reference signal of AFB: (a) Time domain; (b) Frequency domain. domain.

4.2. Optimizing Desing of AFB Using a BOA 4.2. Optimizing Desing of AFB Using a BOA To derive the optimum conditions for a filter bank, this study attempts to determine three design To derive the optimum conditions for a filter bank, this study attempts to determine three factors (number of filters, bandwidth and spacing) for the band-pass filter which would be used in the design factors (number of filters, bandwidth and spacing) for the band-pass filter which would be filter bank. First, to determine the number of band-pass filters, a RE was calculated using Equation (2) used in the filter bank. First, to determine the number of band-pass filters, a RE was calculated by changing the number of filters from 3 to 20 (18 cases in total) through the BOA. Figure 6 below using Equation (2) by changing the number of filters from 3 to 20 (18 cases in total) through the presents the RE results by the number of band-pass filters for the 18 cases. According to the figure, BOA. Figure 6 below presents the RE results by the number of band-pass filters for the 18 cases. when the number of band-pass filters is less than 10, the RE largely depends on the filter conditions. According to the figure, when the number of band-pass filters is less than 10, the RE largely When it is 10 or more, on the contrary, the RE is small depending on the filter conditions. In addition, depends on the filter conditions. When it is 10 or more, on the contrary, the RE is small depending the fluctuation in RE is large when the number of band-pass filters ranges from 3 to 10. When it is 10 on the filter conditions. In addition, the fluctuation in RE is large when the number of band-pass or more, in contrast, fluctuation small. To design efficient filter bank, after all, itTo is filters ranges from 3 the to 10. When it is is relatively 10 or more, in contrast, theanfluctuation is relatively small. advantageous to reduce by decreasing low number of to band-pass Therefore, the the number design an efficient filtera RE bank, after all, itthe is advantageous reduce afilters. RE by decreasing low of band-pass filter is set to ‘10’ in this study. number of band-pass filters. Therefore, the number of band-pass filter is set to ‘10’ in this study. Then, to to determine bandwidth and and spacing, spacing, aa RE RE by Then, determine bandwidth by the the bandwidth bandwidth and and spacing spacing is is calculated calculated using the the 10 The bandwidth bandwidth for for the the band-pass band-pass filter filter is is divided using 10 band-pass band-pass filters. filters. The divided into into 20 20 categories categories (0.1 In addition, spacing for for the the band-pass band-pass filter filter is is classified into 12 (0.1 to to 2.0 2.0 Hz, Hz, increase increase by by 0.1 0.1 Hz). Hz). In addition, spacing classified into 12 conditions (0.2 to 2.4 Hz, an increase by 0.2 Hz). Based on the bandwidth and spacing conditions, conditions (0.2 to 2.4 Hz, an increase by 0.2 Hz). Based on the bandwidth and spacing conditions, REs on on aa total calculated, and and the the results results are are stated stated in in Table Table 11 below. REs total of of 240 240 cases cases are are iteratively iteratively calculated, below. In In addition, they are illustrated in Figure 7. According to Table 1 and Figure 7, when 10 band-pass filters addition, they are illustrated in Figure 7. According to Table 1 and Figure 7, when 10 band-pass are used, bandwidth for the band-pass filter withfilter a minimum RE is 0.7 Hz. for filters arethe used, the bandwidth for the band-pass with a minimum RE Then, is 0.7 the Hz.spacing Then, the the band-pass filter is 1.0 Hz. In conclusion, the conditions of the band-pass filter in the filter bank for spacing for the band-pass filter is 1.0 Hz. In conclusion, the conditions of the band-pass filter in the the optimum acquisition of El-Centro seismic waveforms in Figure 5 are determined as follows: 10 filters, 0.7 Hz in bandwidth and 1.0 Hz of spacing.

Sensors 2017, 1620 Sensors 2017, 17,17, 1620

9 of 9 of 2222

Sensors 17, for 1620the filter2017, bank

9 of 22 optimumacquisition acquisitionofofEl-Centro El-Centroseismic seismicwaveforms waveformsininFigure Figure5 5are aredetermined determined filter bank for the optimum follows:1010filters, filters,0.7 0.7Hz Hzininbandwidth bandwidthand and1.0 1.0Hz Hzofofspacing. spacing. asasfollows:

Reconstruction Error(RE) Reconstruction Error(RE)

0.015 0.015

0.01 0.01

0.005 0.005

0 02 2

44

66

88

1010 1212 1414 Number Filters Number of of Filters

1616

1818

2020

2222

Figure 6. RE results to number of filters. Figure results number filters. Figure 6. 6. RERE results toto number ofof filters.

0.20.2

0.15 0.15

0.10.1 X: 1 X: 1 Y: 0.7 Y: 0.7 Z: 0.01174 Z: 0.01174

0.05 0.05

0 0 0 0

0.50.5

0 0 1 1

1 1 1.51.5

3 3 2 2 Y-Filer Bandwidth(Hz) Y-Filer Bandwidth(Hz)

2 2 X-Filter Spacing(Hz) X-Filter Spacing(Hz)

Figure results filters. Figure 7. 7. RERE results forfor 1010 filters. Figure 7. RE results for 10 filters.

Sensors 2017, 17, 1620

10 of 22

Table 1. RE values for a filter bank with 10 filters.

2.4 2.2 2.0 1.8 1.6 Filter 1.4 Spacing 1.2 (Hz) 1.0 0.8 0.6 0.4 0.2

0.1 0.0351 0.0330 0.0335 0.0326 0.0329 0.0329 0.0316 0.0319 0.0308 0.0289 0.0262 0.022

0.2 0.0328 0.0306 0.0302 0.0295 0.0298 0.0296 0.0279 0.0271 0.0253 0.0221 0.0182 0.0217

0.3 0.0309 0.0287 0.0276 0.0272 0.0272 0.0269 0.0246 0.0226 0.0201 0.0166 0.0150 0.0323

0.4 0.0293 0.0272 0.0254 0.0252 0.0249 0.0244 0.0214 0.0184 0.0158 0.0138 0.0196 0.0444

0.5 0.0280 0.0259 0.0235 0.0233 0.0228 0.0219 0.0184 0.0149 0.0128 0.0145 0.0274 0.0567

0.6 0.0266 0.0247 0.0220 0.0216 0.0207 0.0194 0.0159 0.0127 0.0120 0.0187 0.0360 0.0683

0.7 0.0253 0.0236 0.0206 0.0200 0.0186 0.0172 0.0141 0.0117 0.0139 0.0244 0.0446 0.0791

0.8 0.0243 0.0226 0.0194 0.0186 0.0169 0.0155 0.0130 0.0121 0.0173 0.0306 0.0522 0.0880

Bandwidth of Filters (Hz) 0.9 1.0 1.1 1.2 0.0234 0.0225 0.0217 0.0210 0.0217 0.0207 0.0199 0.0192 0.0183 0.0174 0.0167 0.0161 0.0173 0.0162 0.0153 0.0147 0.0155 0.0145 0.0140 0.0137 0.0142 0.0135 0.0133 0.0136 0.0126 0.0131 0.0141 0.0156 0.0137 0.0160 0.0188 0.0218 0.0212 0.0256 0.0300 0.0344 0.0368 0.0429 0.0487 0.0537 0.0596 0.0666 0.0734 0.0799 0.0960 0.1033 0.1099 0.1154

1.3 0.0203 0.0186 0.0157 0.0143 0.0137 0.0144 0.0174 0.0249 0.0386 0.0586 0.0860 0.1204

1.4 0.0197 0.0180 0.0154 0.0141 0.0141 0.0154 0.0194 0.0281 0.0428 0.0633 0.0919 0.1250

1.5 0.0192 0.0177 0.0153 0.0143 0.0148 0.0167 0.0217 0.0312 0.0469 0.0679 0.0976 0.1292

1.6 0.0188 0.0175 0.0154 0.0146 0.0156 0.0181 0.0240 0.0343 0.0504 0.0724 0.1026 0.1327

1.7 0.0185 0.0174 0.0157 0.0150 0.0165 0.0197 0.0264 0.0373 0.0538 0.0767 0.1075 0.1359

1.8 0.0183 0.0174 0.0160 0.0157 0.0176 0.0214 0.0287 0.0403 0.0572 0.0809 0.1122 0.1389

1.9 0.0181 0.0174 0.0163 0.0164 0.0187 0.0232 0.0311 0.0432 0.0605 0.0850 0.1168 0.1417

2.0 0.0181 0.0075 0.0168 0.0173 0.0199 0.0251 0.0335 0.0459 0.0638 0.0890 0.1212 0.1441

Sensors 2017, 17, 1620 Sensors 2017, 17, 1620

11 of 22 10 of 22

2

2

1

1

0 2

4

6

8

10 12 Number of Filters

14

16

18

Filter Bandwidth (Hz)

Filter Spacing (Hz)

Meanwhile,Figure Figure8 8reveals reveals a change in the bandwidth spacing ofband-pass the band-pass Meanwhile, a change in the bandwidth andand spacing of the filter filter as a as a consequence the change in the number of band-pass filters. According 8, as the consequence of the of change in the number of band-pass filters. According to Figureto8,Figure as the number filters increases, both bandwidth and spacing decrease. other hand, bandwidthand and ofnumber filters of increases, both bandwidth and spacing decrease. On On the the other hand, bandwidth spacingincrease increase when the ToTo getget a maximum amount of information from spacing the number numberofoffilters filtersdecreases. decreases. a maximum amount of information the response using using a limited numbernumber of band-pass filters, then, thethen, filter the bank should beshould optimized. from the response a limited of band-pass filters, filter bank be To determine the numberthe of number band-pass filters, andfilters, also bandwidth and spacing at spacing the same optimized. To determine of band-pass and also bandwidth and attime, the iterative is required.is required. same time,calculation iterative calculation

0 20

Figure8.8.Bandwidth Bandwidthand andfilter filterspacing spacingfor fordifferent differentnumbers numbersofoffilters. filters. Figure

4.3. 4.3.Data DataCompression CompressionUsing UsingaaPPA PPA In InSection Section3.2, 3.2,the theoptimum optimumconditions conditionsofofaafilter filterbank bank(number (numberof ofband-pass band-passfilters, filters,bandwidth bandwidth and are derived derivedbased basedonon BOA. optimized produces reconstructed and spacing) spacing) are thethe BOA. TheThe optimized filterfilter bank bank produces reconstructed signals signals which can simulate original signals most closely. If a filter bank is optimized using a BOA, which can simulate original signals most closely. If a filter bank is optimized using a BOA, after all, after all, the most reconstructed accurate reconstructed signal is even expected, even for random original signals. the most accurate signal is expected, for random original signals. Meanwhile, Meanwhile, though the reconstructed which thefilter optimized filter has bankthe passes has the even thougheven the reconstructed signal whichsignal the optimized bank passes minimum RE minimum RE with original signal, the reconstructed signal’s size is the same as that of the original with original signal, the reconstructed signal’s size is the same as that of the original signal in terms signal in terms of sampling For SHM on large in particular, might be a high of sampling rate. For SHMrate. on large structures, in structures, particular, there might bethere a high demand for a demand for a multi-channel-based precision measurement with a high rate.large Because the multi-channel-based precision measurement with a high sampling rate.sampling Because the dynamic large dynamic responses acquiredbasis on aisreal-time basis is transmitted RF with limited responses acquired on a real-time transmitted using the RF with using limitedthe performances, it can performances, it can cause aofdata loss because of bottleneck. To get large dynamic responses cause a data loss because bottleneck. To get large dynamic responses based on the RF, based a data on the RF, a data compressing technology is essential. compressing technology is essential. For PPA, using usingaacentral centraldifference differencemethod methodmentioned mentionedin Fordata datacompression, compression, this this study study develops a PPA, inFigure Figure 3 and Equation (3).a reference For a reference signal for the of detection the peak value, a 3 and Equation (3). For signal for the detection the peakofvalue, a reconstructed reconstructed signal which the filter bank passes is used. Then, the reconstruction sign’s peak signal which the filter bank passes is used. Then, the reconstruction sign’s peak values are determined values arechanges determined through changes negative and positive signs of onthe thereconstruction derived function of through in negative and positiveinsigns on the derived function signals the reconstruction signals calculated in sequence. Figure 9 bandwidth below reveals CR results by the calculated in sequence. Figure 9 below reveals CR results by the and spacing in which the bandwidth spacinglevel in which the number RE reaches the lowest levelinwhen number of band-pass RE reaches and the lowest when the of band-pass filters Figurethe 6 changes: filtersAccording in Figure 6tochanges: the Figure 9 above, CR is 0.8176 when the number of band-pass filters is 3. The ratio According to thethe Figure 9 above, CR is 0.8176 the to number of band-pass filters 3. The drops to 0.788 when number of band-pass filterswhen increases 20. Here, fluctuations in theisCR over ratio drops when numberfilters of band-pass filters increases to 20. Here, fluctuations in the changes in to the0.788 number ofthe band-pass are minor. In particular, when the number of bass-pass CR over in the number of band-pass In particular, when the number of filters is changes 10 (the optimum condition for the filterfilters bank),are theminor. CR is 79.44, anticipating about 80% of data bass-pass filters is 10 (the optimum condition for the filter bank), the CR is 79.44, anticipating about compressing effects. 80% of data compressing effects.

Sensors 2017, 17, 1620

12 of 22

Sensors 2017, 2017, 17, 17, 1620 1620 Sensors

11 of of 22 22 11 11 0.9 0.9

Compressive Compressive Ratio(CR) Ratio(CR)

0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 00 22

44

66

88

10 12 14 10 12 14 Number of of Filters Filters Number

16 16

18 18

20 20

22 22

Figure Figure9.9. 9.CR CRresults resultsfor fornumber numberof offilters. filters. Figure CR results for number of filters.

4.4.Development Developmentofof ofAFB AFB 4.4. 4.4. Development AFB

55 raw data data raw Filter1 data data Filter1

00 -5 -5 00

Acceleration(g) Acceleration(g)


Inthis thisstudy, study,prior priortoto tothe theIDAQ IDAQsystem’s system’stest, test,an anAFB AFBlogic logicwhich whichwould wouldbe beembedded embeddedinto intothe the In In this study, prior the IDAQ system’s test, an AFB logic which would be embedded into the IDAQ system is developed, and whether or not the AFB could be normally operated is examined. IDAQ IDAQ system system isis developed, developed, and and whether whether or or not not the the AFB AFB could could be be normally normally operated operated is is examined. examined. Then,the theAFB AFB can program band-pass filters using commercial programming language (e.g., C, Then, can program band-pass filters using commercial programming language (e.g., C, C++, Then, the AFB can program band-pass filters using commercial programming language (e.g., C, C++, Matlab, JAVA, etc.). In this study, to develop an AFB, the Matlab M-code is adopted. In Matlab, JAVA, etc.). In this study, to develop an AFB, the Matlab M-code is adopted. In addition, the C++, Matlab, JAVA, etc.). In this study, to develop an AFB, the Matlab M-code is adopted. In addition, seismic the El-Centro El-Centro seismic waveform isAFB. applied into the AFB. AFB. Then, filtering signals signals are El-Centro waveform is applied into the Then, filtering signals are estimated from each addition, the seismic waveform is applied into the Then, filtering are estimated filter from and eachcompared band-passtofilter filter and compared to10each each other. Figures 10 and and 11and below are the the band-pass eachand other. Figuresto andother. 11 below are the time11 frequency estimated from each band-pass compared Figures 10 below are time and and of frequency responses ofdetermined the filter-bank filter-bank of the the conditions determinedwhere optimum conditions where the the responses the filter-bank of the optimum the number of band-pass time frequency responses of the of determined optimum conditions where number of bandwidth band-pass is filters is 10, 10, bandwidth is 0.7 0.7 Hz, andtwo spacing isshow 1.0 Hz. Hz. These twobank figures filters is 10, 0.7 Hz, andbandwidth spacing is 1.0 Hz.Hz, These figures thatThese the filter is number of band-pass filters is is and spacing is 1.0 two figures show that thatdesigned the filter filterusing bank is is optimally designed using the the determined determined conditions. conditions. optimally the determined conditions. show the bank optimally designed using

10 10

20 30 20 30 Time(sec) Time(sec)

40 40

50 50


00 -5 -5 00

10 10

raw data data raw Filter3 data data Filter3

00 -5 -5 00

10 10


40 40

50 50


00 -5 -5 00

10 10


10 10



40 40

50 50

(d) (d) Acceleration(g) Acceleration(g)


-5 -5 00

50 50

55

(c) (c) 55 00

40 40

(b) (b)

55



(a) (a)


40 40

50 50


00 -5 -5 00

(e) (e)

10 10


(f) (f) Figure 10. Cont.

40 40

50 50

Sensors 2017, 17, 1620 Sensors 2017, 17, 1620 Sensors 2017, 17, 1620 raw data raw data Filter7 data Filter7 data

0 0 -5 -5 0 0

10 10

50 50

raw data raw data Filter9 data Filter9 data

0 0 -5 -5 0 0

40 40

(g) (g)

5 5


20 30 20Time(sec)30 Time(sec)

10 10


40 40

50 50



5 5


13 of 22 12 of 22 12 of 22 5 5


0 0 -5 -5 0 0

10 10


50 50

(h) (h)

5 5

raw data raw data data Filter10 Filter10 data

0 0 -5 -5 0 0

40 40

10 10

(i) (i)


40 40

50 50

(j) (j)

3 3 2 2 1 1 0 00 0

3 3 2 2 1 1 0 00 0

3 3 2 2 1 1 0 00 0

5 5

10 15 10 15 Frequency(Hz) Frequency(Hz)

20 20

25 25

3 3 2 2 1 1 0 00 0

25 25

3 3 2 2 1 1 0 00 0

25 25

3 3 2 2 1 1 0 00 0

25 25

3 3 2 2 1 1 0 00 0

Magnitude Magnitude

(a) (a)

5 5


20 20

Magnitude Magnitude


(c) (c)


5 5


20 20

Magnitude Magnitude

3 3 2 2 1 1 0 00 0

25 25

3 3 2 2 1 1 0 00 0


(e) (e)


5 5


20 20

Magnitude Magnitude

3 3 2 2 1 1 0 00 0

(g) (g)


5 5


(i) (i)

20 20

Magnitude Magnitude

Magnitude Magnitude

Magnitude Magnitude

Magnitude Magnitude

Magnitude Magnitude

Magnitude Magnitude

Figure 10. Time domain response of optimal condition AFB: (a) filter 1 data; (b) filter 2 data; (c) filter Figure Timedomain domain response of optimal condition (a) filter data; (b)2 data; filter (c) 2 data; Figure 10. 10. Time response of optimal condition AFB:AFB: (a) filter 1 data;1 (b) filter filter 3 data; (d) filter 4 data; (e) filter 5 data; (f) filter 6 data; (g) filter 7 data; (h) filter 8 data; (i) filter 9 (c) filter (d) 3 data; filter (e) 4 data; filter(f) 5 data; filter(g) 6 data; filter data; 8(h) filter data;9 3 data; filter(d) 4 data; filter (e) 5 data; filter 6(f)data; filter 7(g) data; (h)7 filter data; (i) 8filter data; (j) filter 10 data. (i) filter data;10 (j)data. filter 10 data. data; (j)9filter raw data raw data Filter2 data Filter2 data

5 5


20 20

25 25

(b) (b)


5 5


20 20

25 25

(d) (d)


5 5


20 20

25 25

(f) (f)


5 5


20 20

25 25

(h) (h)

raw data raw data data Filter10 Filter10 data

5 5


20 20

25 25

(j) (j)

Figure 11. Frequency domain responses of optimal condition AFB: (a) filter 1 data; (b) filter 2 data; (c) Figure 11. Frequency domain responses of optimal condition AFB: (a) filter 1 data; (b) filter 2 data; (c) Figure Frequency responses of optimal filter (h) 1 data; filter (i) 2 data; filter 311. data; (d) filter 4domain data; (e) filter 5 data; (f) filtercondition 6 data; (g)AFB: filter(a) 7 data; filter(b) 8 data; filter filter 3 data; (d) filter 4 data; (e) filter 5 data; (f) filter 6 data; (g) filter 7 data; (h) filter 8 data; (i) filter (c) filter (j) 3 data; 9 data; filter(d) 10 filter data. 4 data; (e) filter 5 data; (f) filter 6 data; (g) filter 7 data; (h) filter 8 data; (i) filter data;(j) (j)filter filter10 10data. data. 99 data;

differ. Then, a user can configure the AFB optimally by determining the reference signal for the purpose. 5. Estimation of the IDAQ System In this chapter, Sensors 2017, 17, 1620

the IDAQ system is completed by embedding the AFB developed under the 14 of 22 optimum conditions into the EST-based DAQ system. Based on the El-Centro seismic waveform (SAC Name: LA02 (1940, SE)), then, the IDAQ system’s test is performed. As for the test, it is done Here, the it can be confirmed that the is successfully designed the determined optimum through IDAQ system’s input byAFB forming the El-Centro seismicunder waveform using the LabVIEW’s conditions (six band-pass filters, 0.6 Hz in bandwidth and 1.0 Hz of spacing). In addition, time and function generator. Then, the AFB’s output signals (reconstructed signal, compressive signal, etc.) frequency information is more implicated in the filters 1 through 6. As a result, it is confirmed that the embedded in the IDAQ system was acquired wirelessly from the host PC. Finally, whether or not AFB-added system optimized in the waveform is effective for the acquisition the IDAQIDAQ system properly carries outEl-Centro a seriesseismic of targeted operations (real-time acquisition, ofdecomposition, dynamic responses and SHM on large and relatively flexible structures which have(acceleration)) a range of reconstruction, compression and transmission of dynamic responses frequency 6 Hz. in If the AFB is designed with Figure a random theof El-Centro for SHMless is than assessed a standalone manner. 12areference below signal, revealsnot thewith logic the AFB seismic waveform, however, optimum conditions can differ. Then, a user can configure programmed based on the optimal conditions mentioned earlier, while Figure 12b showsthe the AFB IDAQ optimally determining signal for the purpose. system’sby controller usedthe for reference embedding the programmed AFB's. In this study, NI’s cDAQ-9139 controller is used to completely embed (systematize) the 5. Estimation of the IDAQ System developed AFB into the Matlab M-code. The cDAQ-9139 controller has a 1.33 (max: 2.4 GHz) dual InIntel this chapter, the IDAQ system completed embedding the AFB under theto core i7 processor, 2 GB of RAM is and 32 GB of by data storage space in thedeveloped stand-alone chassis optimum conditions intosystem the EST-based DAQ system.2.Based on the El-Centro seismic waveform accomodate the IDAQ mentioned in Section Therefore, it offers a dramatically improved (SAC Name: LA02 (1940, then, the processing IDAQ system’s is performed. As forand the large test, itdata is done system environment forSE)), acquisition, and test storage of high-speed on a through the IDAQ system’s input by forming the El-Centro seismic waveform using the LabVIEW’s real-time basis. In addition, it features WES7 and Linux-based RTOS which guarantee the function generator. Then, the AFB’s signals (reconstructed compressive signal,can etc.)be embedded system’s stability andoutput real-timeness. Therefore, thesignal, developed AFB logic embedded in the IDAQ system was acquired theInhost PC. Finally, whether or not the completely embedded and operated on awirelessly real-time from basis. addition, a MOXA SWK-3121 AP IDAQ system properly series of targeted decomposition, module designed forcarries multiout I/Oa is applied for an operations integrated(real-time multiple acquisition, (various) measurement and reconstruction, compression of dynamic responses (acceleration)) for SHM is assessed high-speed and two-wayand RFtransmission (Wi-Fi) communication. Thus, the IDAQ system in this paper is indesigned a standalone manner. 12a below reveals the logic of the AFB programmed based on the to carry out a Figure series of standalone operations for the real-time acquisition, decomposition, optimal conditionscompression mentioned earlier, while Figureof12b shows the IDAQ system’s controller used foron reconstruction, and transmission large dynamic responses (acceleration) based embedding the programmed AFB's. the built-in AFB.

(a)

(b)

Figure H/Wand andS/W S/Wfor forthe theIDAQ IDAQsystem: system:(a) (a)EST-based EST-basedDAQ DAQ (Controller); Embedded AFB Figure 12.12. H/W (Controller); (b)(b) Embedded AFB (optimal conditions). (optimal conditions).

In this study, NI’s cDAQ-9139 controller is used to completely embed (systematize) the developed AFB into the Matlab M-code. The cDAQ-9139 controller has a 1.33 (max: 2.4 GHz) dual core Intel i7 processor, 2 GB of RAM and 32 GB of data storage space in the stand-alone chassis to accomodate the IDAQ system mentioned in Section 2. Therefore, it offers a dramatically improved system environment for acquisition, processing and storage of high-speed and large data on a real-time basis. In addition, it features WES7 and Linux-based RTOS which guarantee the embedded system’s stability and real-timeness. Therefore, the developed AFB logic can be completely embedded and operated on a real-time basis. In addition, a MOXA SWK-3121 AP module designed for multi I/O is applied for an integrated multiple (various) measurement and high-speed and two-way RF (Wi-Fi) communication. Thus, the IDAQ system in this paper is designed to carry out a series of standalone operations for the real-time acquisition, decomposition, reconstruction, compression and transmission of large dynamic responses (acceleration) based on the built-in AFB.

Sensors 2017, 17, 1620 Sensors 2017, 17, 1620

15 of 22 14 of 22

5.1. 5.1.Reconstructed ReconstructedSignal Signalofofthe theIDAQ IDAQSystem System To Totest testthe thevalidity validityofofthe theIDAQ IDAQsystem systemdeveloped developedfor forthe theefficient efficientSHM, SHM,a atest testisisconducted. conducted. Then, outputsignals, signals,coming coming from original signals after inputting the El-Centro seismic Then, the the output from original signals after inputting the El-Centro seismic waveform, waveform, are classified into reconstructed signals (a set of signals which have passed through the are classified into reconstructed signals (a set of signals which have passed through the band-pass filter), band-pass filter), and compressive signals are made after peak based and compressive signals which are made which after abstracting only abstracting peak valuesonly based on values reconstructed on reconstructed signals. two signals arein assessed RECR in inEquation CR in signals. Those two signalsThose are assessed with RE Equationwith (2) and Equation(2) (4) and respectively. Equation First, the time and frequency response the reconstructed signals are First, the (4) timerespectively. and frequency response of the reconstructed signals areofestimated from the IDAQ system estimated the IDAQ system as shown in Figures 13 and 14, respectively. as shownfrom in Figures 13 and 14, respectively.

Acceleration(g)

4 raw data Recon. data

2 0 -2 -4

0

5

10

15

20

25 Time(sec)

30

35

40

45

50

(a) 4 raw data Recon. data

2

Acceleration(g)

Acceleration(g)

4

0 -2 -4

0

2

4

6 Time(sec)

8

raw data Recon. data

2 0 -2 -4

10

2

(b)

2.2 2.4 Time(sec)

2.6

(c)

Figure Time Figure13. 13.Time Timedomain domainofofthe thereconstructed reconstructedsignal signalfrom fromthe theIDAQ IDAQsystem: system:(a)(a) Timefrom from0 0toto5050s; s; (b) Detail 1; (c) Detail 2. (b) Detail 1; (c) Detail 2.

Sensors 2017, 17, 1620

15 of 22

Magnitude

Magnitude

Magnitude

According to3 Figure 13, it is found that the time response of the reconstructed signals obtained raw data under the optimum conditions sufficiently simulates that of the original signal, by means of only 10 Recon. data band-pass filters.2 The estimated RE shows 0.01174, and approximately 70% of reconstruction capability is found against total original signals. In particular, the maximum acceleration is 3.417 g 1 at 2.14 s, when the seismic acceleration of original signal is the highest. Then, the maximum acceleration of reconstructed signals is 2.878 g. Therefore, the reproduction capacity turns out to be 0 0 10 15 20 25 85% at the highest acceleration 5of the reconstructed signals, which is excellent. Frequency(Hz) Meanwhile, the frequency response of the reconstructed signal which passed through the AFB (a) embedded in3 the IDAQ system should reflect all range of optimized frequency information besides 3 data raw data the reproduction capability on time response as shown in Figure 13. To assessrawthe frequency Recon. data Recon. data 2 2 response’s reproduction performances in the reconstructed signal, this study defines the reconstructed signals’ spectrum error (SE) as shown in Equation (5) from the concept in Figure 15. 1

0

0

2

SE =4

| ( ) − ( )|/| ( )| 6 Frequency(Hz)

8

10

=

|

1

|/| ( )| 0

(5) 1

1.5 2 2.5 Frequency(Hz)

3

Here, ( ) is the original signal’s frequency response while ( ) refers to (c) the reconstructed (b) signal’s frequency response. In addition, ‘F’ represents the total length (Hz) of the spectrum Figure 14.14.Frequency domain of the reconstructed signal fromthe the IDAQ system: (a) frequency Figure Frequency domain of the reconstructed signal from IDAQ from estimated through the fast fourier transform (FFT) analysis. Then, ifsystem: the SE(a) infrequency Equation (5) 0gets from 0 Hz; tothe 25 ; (b) 1; Detail 1; (c) Detail to (b)Hz Detail (c) Detail 2. closer to25‘0’, reconstruction effect on2.the reconstructed signal’s frequency response is excellent, just like the RE defined in Equation (2). The reconstructed signal’s SE from the frequency response of the According reconstructed signal 13, in Figure 14 isthat 0.01725. Compared to of thethe total original signal’s frequency to Figure it is found the time response reconstructed signals obtained response, 83% of spectrumsufficiently reconstruction abilitythat is found. under theabout optimum conditions simulates of the original signal, by means of only 10

Sensors 2017, 17, 1620

15 of 22 3

Magnitude

raw data Recon. data 2

Sensors 2017, 17, 1620

16 of 22 1

Magnitude

Magnitude

band-pass filters. The estimated RE shows 0.01174, and approximately 70% of reconstruction capability 0 0 5 10 15 20 25 is found against total original signals. In particular, the maximum acceleration is 3.417 g at 2.14 s, Frequency(Hz) when the seismic acceleration of original signal is the (a) highest. Then, the maximum acceleration of reconstructed3 signals is 2.878 g. Therefore, the reproduction capacity turns 3 out to be 85% at the highest raw data data acceleration of the reconstructed signals, which raw is excellent. Recon. data Recon. data 2 2 Meanwhile, the frequency response of the reconstructed signal which passed through the AFB embedded in the IDAQ system should reflect all range of optimized frequency information besides the 1 1 reproduction capability on time response as shown in Figure 13. To assess the frequency response’s reproduction0 performances in the reconstructed signal, this study defines the reconstructed signals’ 0 0 2 4 6 8 10 1 1.5 2 2.5 3 Frequency(Hz) spectrum error (SE) as shown in Equation (5) from the concept in Figure 15. Frequency(Hz) (b) (c) RF RF 0 | δMi | / | u ( f )| 0 | u ( f ) − y ( f )| / | u ( f )| Figure 14. Frequency domain the reconstructed signal the IDAQ system: (a) frequency (5) = from SE = of F F from 0 to 25 Hz; (b) Detail 1; (c) Detail 2.

Figure 15. Concept of spectrum error (SE). Figure 15. Concept of spectrum error (SE).

Also, the SE on the originally targeted frequency range (less than 10 Hz) is found to be 0.00928, Here,about u( f ) is91% the of original signal’s frequency response while y( f ) refers to the reconstructed showing spectrum reconstruction capability, therefore the reconstructed signal signal’s frequency response. In addition, ‘F’ represents the total length (Hz) of the spectrum estimated estimated under the optimum conditions through the AFB can sufficiently simulate the original through the fast fourier transform analysis. Then, if the SE in Equation (5) gets to ‘0’, signal’s frequency response with 10(FFT) band-pass filters only. Specifically, it simulates all closer information the reconstruction effect on the reconstructed signal’s frequency response is results excellent, just like RE on the original signal’s frequency range (less than 10 Hz). Based on the above, the the IDAQ defined in Equation (2). The reconstructed signal’s SE from the frequency response of the reconstructed system this study works successfully with the AFB’s logic optimized in the El-Centro seismic signal in Figure is 0.01725. to the in total original signal’s frequency response, about 83% of waveform. It is14also found Compared to be effective estimating reconstructed signals including mode spectrum reconstruction abilitymeets is found. information concerned which the AFB’s optimum conditions. Also, the SE on the originally targeted frequency range (less than 10 Hz) is found to be 0.00928, showing 91% ofofspectrum capability, therefore the reconstructed signal estimated 5.2. Data about Compression the IDAQreconstruction System under the optimum conditions through the AFB can sufficiently simulate the original signal’s frequency To test the validity of the IDAQ system developed for the efficient SHM, then, the data response with 10 band-pass filters only. Specifically, it simulates all information on the original signal’s compression ability was assessed. The reconstructed signal’s time response derived in Figure 13 is frequency range (less than 10 Hz). Based on the results above, the IDAQ system in this study works about 70% against the original signal while the reconstructed signal’s frequency response from successfully with the AFB’s logic optimized in the El-Centro seismic waveform. It is also found to be Figure 14 is about 83% against the original signal in terms of reconstruction ability. Even so, the size effective in estimating reconstructed signals including mode information concerned which meets the of the estimated reconstructed signal’s time response is the same as the data size of original signal. AFB’s optimum conditions. To carry out SHM by acquiring a great number of dynamic response data (acceleration) using the limited data compression technology 5.2. DataRF, Compression of the IDAQ System is needed. For this, this study develops a PPA using the central difference method. Here, the PPA is designed to extract peak values only based on To test the validity of the IDAQ reveals system the developed for the efficient then, thesignal data reconstructed signals. Figure 16 below peak-picking results of theSHM, reconstructed compression ability was assessed. The reconstructed signal’s time response derived in Figure 13 from the IDAQ system. is about 70% against the original signal while the reconstructed signal’s frequency response from Figure 14 is about 83% against the original signal in terms of reconstruction ability. Even so, the size of the estimated reconstructed signal’s time response is the same as the data size of original signal. To carry out SHM by acquiring a great number of dynamic response data (acceleration) using the limited RF, data compression technology is needed. For this, this study develops a PPA using the central difference method. Here, the PPA is designed to extract peak values only based on

Sensors 2017, 17, 1620

17 of 22

reconstructed signals. Figure 16 below reveals the peak-picking results of the reconstructed signal 16 of 22 from the IDAQ system.

Sensors 2017, 17, 1620

Acceleration(g)

4 Recon. data Picking data

2 0 -2 -4

0

5

10

Sensors 2017, 17, 1620

15

20

25 Time(sec)

30

35

40

4 2 0 -2 -4

-2

4

Recon. data Picking data

Acceleration(g)

Acceleration(g)

0

5

10

15

20

-4

25 Time(sec)

30

35

40

(a) 2 Acceleration(g)

0

16 of 22

4

4

6 Time(sec)

8


2 0 -2

10

45

50

2

2.2 2.4 Time(sec)

4


(b)

2


-4

Acceleration(g)

Acceleration(g)

0

50

(a)

4 2

45

0

2.6


(c)

2 0

Figure 16. Peak-picking results of the reconstructed signal from the IDAQ system: (a) Time from 0 to -2 -2 Figure 16. Peak-picking results of the reconstructed signal from the IDAQ system: (a) Time from 0 to 5050s; s;(b) Detail 1; (c) Detail 2. -4 (b) Detail 1; -4(c) Detail 2. 0

2

4

6 Time(sec)

8

10

2

2.2 2.4 Time(sec)

2.6

(b) method-based PPA is able to pick up InInFigure Figure16, 16,the thecentral centraldifference difference method-based PPA is able to pick (c) uponly onlythe thereconstructed reconstructed signal’s peak values. A total of 520 peak values are derived. The CR against 2500, Figure 16. Peak-picking the reconstructed signal from the IDAQ (a) Time 2500, from 0the tothenumber signal’s peak values. A total ofresults 520 ofpeak values are derived. The system: CR against numberofof 50 s; (b) Detail 1; (c) Detail 2. reconstructed for fifty seconds, 0.7920. When PPA is combined the AFB, reconstructedsignals signals for fifty seconds, isis0.7920. When PPA is combined withwith the AFB, aboutabout 80% of80% data ofare data are expected to be compressed under the conditions optimized with the El-Centro seismic expected to be compressed under the conditions optimized with the El-Centro seismic waveform. In Figure 16, the central difference method-based PPA is able to pick up only the reconstructed waveform. Based on the results above, itpeak is the operates IDAQ system operates well based onwhich the signal’s peak values. of 520 valuesthat are derived. The CR against 2500, number of Based on the results above, it Ais total found that thefound IDAQ system well based onthe the AFB’s logic AFB’s logic which is added to the PPA developed using the central difference method. Among the reconstructed signals for fifty seconds, is 0.7920. When PPA is combined with the AFB, about 80% is added to the PPA developed using the central difference method. Among the reconstructed signal’s of data are expected toresponses, be compressed underonly the conditions optimized with the El-Centro seismic reconstructed signal’s time then, the peak values which represent frequency time responses, then, only the peak values which represent frequency information (periodicity) are picked waveform. Based on the results above, it is found that the IDAQ system operates well based on the information arein picked up socompressed that it is proven valid in estimating compressed dynamic up so that AFB’s it(periodicity) is proven valid estimating dynamic responses. logic which is added to the PPA developed using the central difference method. Among the responses. reconstructed signal’s time responses, then, only the peak values which represent frequency 5.3. Compressive Signal of the IDAQ System information (periodicity) are picked up so that it is proven valid in estimating compressed dynamic responses. 5.3. Compressive Signal of the IDAQ System To test the validity of the IDAQ system developed for the efficient, only the peak values (Figure 16) Compressive Signal of theIDAQ IDAQ To the validity of system the system developed for RF, theand efficient, only the peak values picked test up5.3. from the IDAQ areSystem transmitted using the the test is performed to have (Figure 16) picked up host from the Then, IDAQ system arevalues transmitted the RF,only andis the test is performed them stored in PC. peak stored in host PC topeak be values restored into a To the test the validity of the the IDAQ system developed forusing thethe efficient, the (Figure 16)type picked up from theSHM. IDAQ In system are transmitted usingin the RF, host and the istoperformed tosignal-analyzable have them stored in the host PC. Then, thethis peak values the PCtest ispeak bevalues restored into for future study, thestored signals restored with only are to have them stored infuture the hostSHM. PC. Then, the peak valuesthe stored in the restored host PC is to be restored into a defined signal-analyzable type for In this study, signals with peak values only as ‘compressive signals’. For SHM, compressive signals of the same size (time length) as the a signal-analyzable type for future SHM. In this study, the signals restored with peak values only are defined as ‘compressive signals’. For SHM, compressive signals of theTosame size (time length) as spacing (delta t) equivalent to the original sampling rate are required. crate compressive signals are defined as ‘compressive signals’. For SHM, compressive signals of the same size (time length) as the spacing (delta t) equivalent to the original sampling rate are required. To crate compressive using the peaks only, which are acquired and stored based on irregular spacing, linear interpolation is the spacing (delta t) equivalent to the original sampling rate are required. To crate compressive signals using the peaks only, which are acquired and stored based on irregular spacing, linear signals using the peaks only, which are acquired and from stored onPPA irregular spacing, linear used. This method can preserve peak values determined thebased original and create compressive interpolation is used. This method can preserve peakvalues values determined fromfrom the original PPA andPPA and interpolation is used. This method can preserve peak determined the original signals which are the same in terms of reconstructed signal and spacing within the peak range. create compressive signals which are the same in terms of reconstructed signal and spacing within create compressive signals which the sameshow in terms of and reconstructed signal and spacing within Figures 17the and 18 range. below show theare time domains of compressive restored using peak Figures 17 and 18 and belowfrequency the time frequency domains ofsignals compressive the 17 and 18 below showrespectively. the time and frequency domains of compressive thepeak linearrange. interpolation, respectively. signalsFigures restored using the linear interpolation, signals restored using the linear interpolation, respectively. Acceleration(g)

4

Acceleration(g)

4 2 0

Recon. data Comp. of recon. data

2

Recon. data Comp. of recon. data

0 -2 -4

0

5

10

15

20

-2 -4

25 Time(sec)

30

35

40

45

50

(a) 0

5

10

15

20

25

30

Time(sec) Figure 17. Cont.

(a)

35

40

45

50

Sensors 2017, 17, 1620 Sensors2017, 2017,17, 17,1620 1620 Sensors

18 of 22 17of of22 22 17

Recon.data data Recon. Comp.ofofrecon. recon.data data Comp.

22



44

00

-2-2 -4-40 0

22

44

66 Time(sec) Time(sec)

88

10 10

88 66

Recon.data data Recon. Comp.ofofrecon. recon.data data Comp.

44 22 00 -2-2 -4-4

(b) (b)

22

2.2 2.4 2.2 2.4 Time(sec) Time(sec)

2.6 2.6

(c) (c)

Figure 17.17. Time domain ofcompressive compressive signal using peak-picking values: (a)Time Time fromfrom to50 50 (b)s; Figure Time domain of compressive signal using peak-picking values: (a) Time 0 to 50 Figure 17. Time domain of signal using peak-picking values: (a) from 00to s;s;(b) Detail 1; (c) Detail 2. (b) Detail 1; (c) Detail 2. Detail 1; (c) Detail 2.

Magnitude Magnitude

33

Recon.data data Recon. Comp.ofofrecon recondata data Comp.

22 11 00 0 0

55

10 10

Frequency(Hz) Frequency(Hz)

15 15

20 20

25 25

(a) (a) Recon.data data Recon. Comp.ofofrecon recondata data Comp.

22 11 000 0

11

22

33

44 55 66 Frequency(Hz) Frequency(Hz)

(b) (b)

77

88

99

10 10

44 Magnitude Magnitude

Magnitude Magnitude

33

33

Recon.data data Recon. Comp.ofofrecon recondata data Comp.

22 11 00 1 1

1.5 2.5 1.5 22 2.5 Frequency(Hz) Frequency(Hz)

33

(c) (c)

Figure Frequency domainofofcompressive compressivesignal signalusing usingpeak-picking peak-picking values: values: (a) (a) frequency frequency from Figure 18.18. Frequency domain from000 to Figure 18. Frequency domain of compressive signal using peak-picking values: (a) frequency from 25 Hz; (b) Detail 1; (c) Detail 2. to25 25Hz; Hz;(b) (b)Detail Detail1;1;(c) (c)Detail Detail2.2. to

InIn the compressive signal’s time response derived from Figure 17, the the RE is 0.00756, 0.00756, compared comparedto the compressive signal’stime timeresponse responsederived derivedfrom fromFigure Figure17, RE is is 0.00756, In the compressive signal’s to the reconstructed signal. As a result, the compressive signal is well able to restore data compared well able toto restore data compared with to the the reconstructed reconstructedsignal. signal.As Asaaresult, result,the thecompressive compressivesignal signalis is well able restore data compared with the reconstructed reconstructed signal (the restoration restoration ability approximately 81%).the Specially, the the reconstructed signalsignal (the restoration ability isability approximately 81%). Specially, reconstructed with the (the isis approximately 81%). Specially, the reconstructed signal’s peak values values picked up from from the PPA PPAreflected are properly properly reflected and and simulated.an signal’s peaksignal’s values peak picked up from the PPA are properly and simulated. Nevertheless reconstructed picked up the are reflected simulated. Nevertheless an RE REaoccurs occurs because section of the the empty time dataiswithout without peak values isis RE occurs because sectionbecause of the empty time data without peak values linearized in restoring the Nevertheless an aa section of empty time data peak values linearized in restoring the compressive signal’s time response using the linear interpolation. compressive signal’s time response using the linear interpolation. Because the peak values of time linearized in restoring the compressive signal’s time response using the linear interpolation. Because theare peak values ofof time response are usedas asone one of major majorinterest interest information in executing executing response used as one major interestare information in executing SHM, the linear interpolation-based Because the peak values of time response used of information in SHM, the linear linear interpolation-based compressive signal restoration restoration technology accomplishes accomplishes the the compressive signal restoration technology accomplishes the design goal. SHM, the interpolation-based compressive signal technology design goal. While the reconstructed signal remarks especially mode information about frequency range of design goal. While the reconstructed signal remarks especially mode information about frequency range of original (1 to 10 Hz) insignal Figureremarks 14, the compressive signalinformation expresses it about all across the totalrange frequency Whilesignal the reconstructed especially mode frequency of original signal (1 to to 10 10 signals Hz) in inas Figure 14, 18 the compressive signal expresses alldata across the total total range of reconstructed in Figure because the peak values, the reference of compressive original signal (1 Hz) Figure 14, the compressive signal expresses itit all across the frequency range of reconstructed reconstructed signals as in in Figure Figure 18 because because the peak peak values, the reference reference data signals, are derived throughout signals the reconstructed signal’s response time.values, The estimated compressive frequency range of as 18 the the data ofsignal’s compressive signals, are shows derived throughout the reconstructed reconstructed signal’s response response time. The frequency response 0.01445 of SE, compared to the reconstructed signal. Intime. comparison of compressive signals, are derived throughout the signal’s The estimated compressive signal’sfrequency frequencyresponse, response shows 0.01445 of SE, SE, compared to to the the is with the compressive total original signal’s about 85%0.01445 of spectrum reconstruction ability estimated signal’s frequency response shows of compared reconstructed signal. In comparison with the total original signal’s frequency response, about 85% observed. Based on In thecomparison results above, therefore, compressive signal’s frequency response properly reconstructed signal. with the totalthe original signal’s frequency response, about 85% ofsimulates spectrumallreconstruction reconstruction ability isisthe observed. Basedsignal’s on the thefrequency results above, above, Lastly, therefore, the information throughout reconstructed range. to test the of spectrum ability observed. Based on results therefore, the compressive signal’s frequency response properly simulates all information throughout the validity of the IDAQ system developed for efficiency, the compressive signal’s time and frequency compressive signal’s frequency response properly simulates all information throughout the reconstructed signal’s frequency range. Lastly, to test test the the as validity of the the IDAQ IDAQ system developed responses are shownfrequency in Figuresrange. 19 and 20, respectively the final results of the test based on the reconstructed signal’s Lastly, to validity of system developed fororiginally efficiency, the compressive compressive signal’s time time and and frequency frequency responses responses are are shown shown in in Figures Figures 19 19 and and targeted original signals. for efficiency, the signal’s 20, respectively as the final results of the test based on the originally targeted original signals. 20, respectively as the final results of the test based on the originally targeted original signals.

Sensors 2017, 17, 1620 Sensors 2017, 17, 1620 Sensors 2017, 17, 1620

19 of 22 18 of 22 18 of 22


4 4

raw data raw data Comp. of recon. data Comp. of recon. data

2 2 0 0 -2 -2 -4 -40 0

5 5

10 10

15 15

20 20

25 25 Time(sec) Time(sec)

30 30

35 35

40 40

45 45

50 50

(a) (a)


2 2



4 4

0 0 -2 -2 -4 -40 0

2 2

4 4

6 6 Time(sec) Time(sec)

8 8

10 10

8 8 6 6 4 4 2 2 0 0 -2 -2 -4 -4

(b) (b)


2 2

2.2 2.4 2.2 2.4 Time(sec) Time(sec)

2.6 2.6

(c) (c)

Figure raw signal compressive signal :: (a) from to 50 s;s; (b) Figure19. 19.Time Time domain domain signal vs.vs. compressive signal: (a) Time from to 5000s; Figure 19. Time domainofof ofraw raw signal vs. compressive signal (a) Time Time 0 from to(b) 50Detail (b) 1;

Detail 1; (c) Detail 2. Detail 1; (c) (c)Detail Detail2. 2.

Magnitude Magnitude

3 3


2 2 1 1 0 00 0

5 5

10 10

Frequency(Hz) Frequency(Hz)

15 15

20 20

25 25

(a) (a)

4 4


2 2

Magnitude Magnitude

Magnitude Magnitude

3 3

1 1 0 00 0

2 2


(b) (b)

8 8

10 10

3 3


2 2 1 1 0 01 1

1.5 2 2.5 1.5 2 2.5 Frequency(Hz) Frequency(Hz)

3 3

(c) (c)

Figure Figure20. 20.Frequency Frequencydomain domainof ofraw rawsignal signal vs. vs.compressive compressivesignal: signal: (a) (a)frequency frequencyfrom from00to to25 25Hz; Hz;(b) (b) Figure 20.Detail Frequency domain of raw signal vs. compressive signal: (a) frequency from 0 to 25 Hz; Detail 1; (c) 2. Detail 1; (c) Detail 2. (b) Detail 1; (c) Detail 2.

In In Figure Figure 19, 19, the the compressive compressive signal’s signal’s time time response response generated generated based based on on the the reconstructed reconstructed Inpeak Figure 19, the compressive signal’sthe time response generated based on theRE reconstructed signal’s signal’s values sufficiently simulates original signal’s periodicity. The signal’s peak values sufficiently simulates the original signal’s periodicity. The RE with with the the original original peakisvalues sufficiently simulates the original signal’s periodicity. The RE with the originalsignal. signal is signal 0.01702, showing about 60% of reconstruction ability against the reconstructed signal is 0.01702, showing about 60% of reconstruction ability against the reconstructed signal. 0.01702, showing about 60% of reconstruction ability is against the reconstructed signal. Unlike the case Unlike Unlike the the case case in in Figure Figure 17, 17, aa relatively relatively large large error error is observed observed in in the the compressive compressive signal’s signal’s time time in Figure 17, a relatively large errorItisappears observed in the compressive signal’s time response against the response response against against the the original original signal. signal. It appears that that itit originates originates from from the the reconstructed reconstructed signal’s signal’s RE RE original signal. It appears that it originates from the reconstructed signal’s RE against the original against the original signal. As a way to overcome the compressive signal’s RE, a user can increase against the original signal. As a way to overcome the compressive signal’s RE, a user can increase signal. As aof way to overcome the RE, compressive signal’s RE, a6, canmeasuring increase the number of filters the number filters against Figure after the efficiency in the number of filters against the the RE, as as described described Figure 6,user after measuring the efficiency in against the RE, as described Figure 6, after measuring the efficiency in configuring the filter bank. configuring the filter bank. In Figure 20, just like the characteristics of the reproduction of the configuring the filter bank. In Figure 20, just like the characteristics of the reproduction of the In Figure 20,signal’s just like frequency the characteristics of against the reproduction of the reconstructed reconstructed response the signal shown Figure 14, reconstructed signal’s frequency response against the original original signal shown in insignal’s Figure frequency 14, the the response against the original signal shown Figure 14, the compressive signalssignal’s intensively simulate compressive signals intensively simulate the information on frequency compressive signals intensively simulate theinmode mode information on the the original original signal’s frequency the mode information range (less 10 range (less than than 10 Hz). Hz). on the original signal’s frequency range (less than 10 Hz). Compared to the original original compressive signal’s frequency response is 0.03598 Compared signal, the estimated compressive signal’s frequency response is Compared to to the the original signal, signal,the theestimated estimated compressive signal’s frequency response is in terms of the SE, showing about 63% of spectrum reconstruction ability against the total original 0.03598 in terms of the SE, showing about 63% of spectrum reconstruction ability against the total 0.03598 in terms of the SE, showing about 63% of spectrum reconstruction ability against the total original original signal’s signal’s frequency frequency response. response. In In particular, particular, the the SE SE on on the the originally originally targeted targeted frequency frequency range range

Sensors 2017, 17, 1620

20 of 22

signal’s frequency response. In particular, the SE on the originally targeted frequency range (less than 10 Hz) is 0.02831, showing about 71% of spectrum reconstruction ability. Based on the results, even though only several band-pass filters are used through the BOA, the AFB-added, the data compression technology-based IDAQ system developed in this study turns out to be effective in acquiring effective dynamic responses which can sufficiently express the targeted original signal’s time and frequency information by acquiring compressed peak values only through the PPA. In the restoration of compressive signals using peak values only, in addition, linear interpolation is effective in creating compressive signals within the complete peak value section while preserving conventional peak value information. 6. Conclusions This study develops a data compression technology-based IDAQ system for the efficient by overcoming the limitations of the conventional hardware-based wired and wireless SHM system and adopting the latest measuring system and technology. To test the validity of the developed IDAQ system, the test is carried out. Then, the study results are as follows: 1.

2.

3.

4.

5.

6.

The BOA developed for the AFB highlights only the signals of the frequency range including the mode of interest among random signal’s broad frequency factors. The reconstructed signal reveals 70% and 83% of reconstruction effects in time and frequency responses against the original signal respectively. In particular, when the AFB is optimized using the El-Centro seismic waveform, it could be available as a filter technology needed to acquire dynamic responses in large flexible architectural structure (less than 10 Hz). The central difference method-based PPA developed for the AFB selectively resamples only peak values including effective modal information among the reconstructed signals. Then, about 80% of data compression effects are expected. This kind of data compressing technology can overcome the limitations of RF communication and be available as a technology for efficient operation and management (big-data problem) of database at a long-term monitoring. The AFB developed by using high-level language is embedded into the IDAQ system in a fast and accurate manner. As a result, it is able to acquire compressed effective dynamic responses on a real-time basis, specially, because the AFB can add or deduct diverse logics and functions depending on a user’s needs and easily edit and add the filter bank’s logic when response changes. Therefore, it can innovatively improve the conventional hardware-based filter bank’s design and implementation method in terms of efficiency and economy. The linear interpolation applied for the restoration of the compressive signal is found to be effective in creating compressive signals for SHM by keeping the conventional reconstructed signal’s peak values intact. Then, in time and frequency responses against reconstructed signals, the compressive signal reveals 81% and 85% of expected reconstruction effects, respectively. This kind of linear interpolation can enhance the IDAQ system’s practical availability. The IDAQ system overcomes the conventional hardware-based wired and wireless SHM system’s limitations by adopting high-performance hardware, multi-input/output channel (I/O), high-speed RF module, EST and RTOS based on the latest measuring system and technology. With the addition of AFB, in particular, it can substitute the conventional hardware-based filter bank, showing a possibility as a brand-new SHM system which enables efficient dynamic responses acquisition from structures. In further studies, the IDAQ system would be continuously upgraded and applied to actual structures with a goal of evolving into a bio-inspired structural system equipped with intelligent sensing, judgment and response capabilities.

Acknowledgments: This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (Grant Number: NRF-2016R1A2A1A05005499, Grant Number: NRF-2017R1A2B4001836).

Sensors 2017, 17, 1620

21 of 22

Author Contributions: Gwanghee Heo participated in data analysis and revised the manuscript. Joonryong Jeon performed all stages of the study including data collection, analysis, interpretation, and substantial revision of the manuscript. All authors collaboratively wrote the paper, and design the data compression technology-based intelligent data acquisition (IDAQ) system. Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations The following abbreviations are used in this manuscript: SHM IDAQ EST RTOS DAQ RF AP AFB BOA PPA RE CR SE FFT

Structural Health Monitoring Intelligent Data Acquisition Embedded Software Technology Real-Time Operating System Data Acquisition Radio Frequency Access Point Artificial Filter Bank Band-pass filter Optimizing Algorithm Peak-Picking Algorithm Reconstruction Error Compressive Ratio Spectrum Error Fast Fourier Transform

References 1. 2. 3. 4. 5.

6. 7.

8. 9. 10. 11.

Peeters, B.; De Roeck, G. One-year monitoring of the Z24-Bridge: Environmental effects versus damage events. Earthq. Eng. Struct. Dyn. 2001, 30, 149–171. [CrossRef] Sohn, H. Effect of environmental and operational variability on structural health monitoring. Philos. Trans. A 2007, 365, 539–560. [CrossRef] [PubMed] Giraldo, D.F.; Dyke, S.J.; Caicedo, J.M. Damage detection accommodating varying environmental conditions. Struct. Health Monit. 2006, 5, 155–172. [CrossRef] Tomaszewska, A. Influence of statistical errors on damage detection based on structural flexibility and mode shape curvature. Comput. Struct. 2010, 88, 154–164. [CrossRef] Baptista, F.G.; Budoya, D.E.; Almeida, V.A.D.; Ulson, J.A.C. An Experimental Study on the Effect of Temperature on Piezoelectric Sensors for Impedance-Based Structural Health Monitoring. Sensors 2014, 14, 1208–1227. [CrossRef] [PubMed] Na, W.S.; Baek, J. Impedance-Based Non-Destructive Testing Method Combined with Unmanned Aerial Vehicle for Structural Health Monitoring of Civil Infrastructures. Appl. Sci. 2017, 17, 15. [CrossRef] Zhang, J.; Tian, G.Y.; Marindra, A.M.J.; Sunny, A.I.; Zhao, A.B. A Review of Passive FRID Tag Antenna-Based Sensors and Systems for Structural Health Monitoring Applications. Sensors 2017, 17, 265. [CrossRef] [PubMed] Salmanpour, M.S.; Khodaei, Z.S.; Aliabadi, M.H. Impact Damage Localisation with Piezoelectric Sensors under Operational and Environmental Conditions. Sensors 2017, 17, 1178. [CrossRef] [PubMed] Lee, Y.-J.; Cho, S. SHM-Based Probabilistic Fatigue Life Prediction for Bridges Based on FE Model Updating. Sensors 2016, 16, 317. [CrossRef] [PubMed] Andersen, E.Y.; Pedersen, L. Structural monitoring of the Great Belt East Bridge. In Proceedings of the Third Symposium on Strait Crossing, Ålesund, Norway, 12–15 June 1994; pp. 189–195. Celebi, M.; Purvis, R.; Hartnagel, B.; Gupta, S.; Clogston, P.; Yen, P.; O’Connor, J.; Franke, M. Seismic instrumentation of the Bill Emerson Memorial Mississippi River Bridge at Cape Girardeau (MO): A cooperative effort. In Proceedings of the 4th International Seismic Highway Conference, Memphis, TN, USA, 9–11 February 2004.

Sensors 2017, 17, 1620

12.

13.

14.

15. 16. 17. 18. 19.

20. 21.

22. 23. 24.

22 of 22

Abe, M.; Fujino, Y.; Yanagihara, M.; Sato, M. Monitoring of hakucho suspension bridge by ambient vibration measurement. In Proceedings of the Nondestructive Evaluation of Highways, Utilities, and Pipelines IV, Newport Beach, CA, USA, 7–9 March 2000. Baek, J.; Hansen, M.; Nigbor, R.; Tileylioglu, S. Elevators as an Excitation Source for Structural Health Monitoring in Buildings. In Proceedings of the 4th World Conference on Structural Control and Monitoring, Los Angeles, CA, USA, 11–12 July 2006. Khemlani, L. CORENET e-PlanCheck: Singapore’s Automated Code Checking System. AECbypte. Available online: http://www.aecbytes.com/buildingthefuture/2005/CORENETePlanCheck_pr.html (accessed on 27 December 2011). Ko, J.M.; Ni, Y.Q. Technology developments in structural health monitoring of large-scale bridges. Eng. Struct. 2005, 27, 1715–1725. [CrossRef] Spencer, B.F.; Ruiz-Sandoval, M.E.; Kurata, N. Smart sensing technology: Opportunities and challenges. J. Struct. Control Health Monit. 2004, 11, 349–368. [CrossRef] Lynch, J.P.; Loh, K.J. A summary review of wireless sensors and sensor networks for structural health monitoring. Shock Vib. Dig. 2006, 28, 91–126. [CrossRef] Nagayama, T.; Spencer, B.F.; Mechitov, K.A.; Agha, G.A. Middleware services for structural health monitoring using smart sensors. Smart Struct. Syst. 2009, 5, 119–137. [CrossRef] Rice, J.A.; Spencer, B.F. Structural health monitoring sensor development for the Imote2 platform. In Proceedings of the Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems, San Diego, CA, USA, 4–8 April 2008. Park, J.-W.; Sim, S.-H.; Jung, H.-J.; Spencer, B.F. Development of a Wireless Displacement Measurement System Using Acceleration Responses. Sensors 2013, 13, 8377–8392. [CrossRef] [PubMed] Kurata, M.; Kim, J.; Lynch, J.P.; van der Linden, G.; Sedarat, H.; Thometz, E.; Hipley, P.; Sheng, L. Internet-Enabled Wireless Structural Monitoring Systems: Development and Permanent Deployment at the New Carquinez Suspension Bridge. J. Struct. Eng. 2013, 139, 1688–1702. [CrossRef] Peckens, C.A.; Lynch, J.P. Utilizing the cochlea as a bio-inspired compressive sensing technique. Smart Mater. Struct. 2013, 22, 1–16. [CrossRef] Galbraith, C.J. Cochlea-Based RF Channelizing Filters. IEEE Trans. Circuits Syst. 2008, 55, 969–979. [CrossRef] Albuquerque, D.F. Analog Filter Bank for Cochlear Radio. In Proceedings of the 2010 IEEE International Microwave Workshop Series on RF Front-ends for Software Defined and Cognitive Radio Solutions (IMWS), Aveiro, Portugal, 22–23 February 2010; pp. 1–4. © 2017 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/).

Integrated system of structural health monitoring and intelligent management for a cable-stayed bridge.

An on-line data acquisition microcomputer: system for monitoring physiological variables during rest and exercise.

Case study on the maintenance of a construction monitoring using USN-based data acquisition.

Intelligent data acquisition blends targeted and discovery methods.

A Sensor Data Fusion System Based on k-Nearest Neighbor Pattern Classification for Structural Health Monitoring Applications.

Damage detection with streamlined structural health monitoring data.

New Quality Control Algorithm Based on GNSS Sensing Data for a Bridge Health Monitoring System.

A sample computer system for physiological data acquisition and analysis.

A data acquisition system for three-dimensional treatment planning.

Data Acquisition System for In Situ Monitoring of Chemoelectrical Potential in Living Plant Fuel Cells.

Analysis of ECG data, for data compression.

TCGA Expedition: A Data Acquisition and Management System for TCGA Data.

Intelligent management of epidemiologic data.

Multivariate data analysis based on a computerized patient monitoring system.

Data acquisition system for harmonic motion microwave Doppler imaging.

wireless) acquisition data system for monitoring of cultural heritage physical parameters in Smart Cities.

Compressive sensing of full wave field data for structural health monitoring applications.

A direct-to-drive neural data acquisition system.

Data Collection for Monitoring Hepatitis Testing Programs: The HepTLC Data Management System.

Big Data: Implications for Health System Pharmacy.

Data-driven matched field processing for Lamb wave structural health monitoring.

Data compression of large document data bases.

A global assessment of civil registration and vital statistics systems: monitoring data quality and progress.

A data acquisition protocol for a reactive wireless sensor network monitoring application.