Dense Sensor Arrays Research Articles

Autonomous decentralized structural health monitoring using smart sensors

Though smart sensor technology has seen substantial advances during recent years, implementation of smart sensors on full-scale structures has been limited. Direct replacement of wired sensing systems with wireless sensor networks is not straightforward as off-the-shelf wireless systems are unlikely to provide the data users expect. The difficulty arises, in part, because centralized systems common for wired measurements are not scalable to large numbers of smart sensors. Decentralized computing is required to harvest the rich information that a dense array of smart sensors can make available for structural health monitoring (SHM). Decentralized system can be effectively managed by each sensor node and cluster autonomously performing tasks such as sensing and damage detection. Taking advantage of recent developments in middleware services which provide basic functionality for structural vibration measurements (e.g. synchronized sensing and reliable communication) this paper proposes an autonomous decentralized SHM approach using smart sensors. A set of algorithms that mesh well with the decentralized approach are first proposed and implemented on smart sensors. Finally, experimental validation of the implemented system is given and its damage detection capability is discussed. Copyright © 2009 John Wiley & Sons, Ltd.

A decentralized procedure for structural health monitoring of uncertain nonlinear systemsprovided with dense active sensor arrays

A numerical simulation study is conducted to evaluate the feasibility of a proposedalgorithm for testing and analyzing the vibration signature of complex, nonlinear anduncertain structural systems provided with dense, active sensor arrays that have thecapability to generate local probing signals. The proposed algorithm for ‘decentralized’identification of the parameters of a highly reduced-order model can be easily embedded insimple processors that are incorporated in modern wireless sensors, without the need fordemanding computational resources or inter-sensor communications. It is shown that theproposed estimation approach, in which a reduced-order equivalent single-degree-of-freedomlinear system is identified to match the observed response of a specific sensor/actuatorpair, is capable of detecting relatively small changes in the estimated systemparameters. These parameters can be directly correlated with physically meaningfulmeasures of the structural dynamic properties of the system being monitored.A parametric study is conducted to determine the robustness of the proposedapproach with regard to variation in ‘damage’ location relative to the sensorlocation, ‘damage’ magnitude, scatter in the estimated system parameters due to thestochastic nature of the nonlinear system and variability (uncertainty) in thereference system parameters. It is shown that the proposed approach is viable forimplementations involving sensor ‘motes’ with limited storage and computationalresources.

Environmental controls and the influence of vegetation type, fine roots and rhizomorphs on diel and seasonal variation in soil respiration.

Characterization of spatial and temporal variation of soil respiration coupled with fine root and rhizomorph dynamics is necessary to understand the mechanisms that regulate soil respiration. A dense wireless network array of soil CO2 sensors in combination with minirhizotron tubes was used to continuously measure soil respiration over 1 yr in a mixed conifer forest in California, USA, in two adjacent areas with different vegetation types: an area with woody vegetation (Wv) and an area with scattered herbaceous vegetation (Hv). Annual soil respiration rates and the lengths of fine roots and rhizomorphs were greater at Wv than at Hv. Soil respiration was positively correlated with fine roots and rhizomorphs at Wv but only with fine roots at Hv. Diel and seasonal soil respiration patterns were decoupled with soil temperature at Wv but not at Hv. When decoupled, higher soil respiration rates were observed at increasing temperatures, demonstrating a hysteresis effect. The diel hysteresis at Wv was explained by including the temperature-dependent component of soil respiration and the variation dependent on photosynthetically active radiation. The results show that vegetation type and fine root and rhizomorph dynamics influence soil respiration in addition to changes in light, temperature and moisture.

Automated Modal Parameter Estimation by Parallel Processing within Wireless Monitoring Systems

With recent advances in low-cost wireless sensing and data acquisition technology, it has become feasible to instrument a large structure with a dense array of wireless sensors. Furthermore, analog-to-digital conversion and data processing capabilities of current wireless sensor prototypes offer the ability to efficiently distribute data processing tasks across a large network of wireless sensing nodes. For decades, the structural engineering community has been adapting input-output modal identification techniques for use in large-scale civil structures. However, unlike in mechanical or aerospace engineering, it is often difficult to excite a large civil structure in a controlled manner. Thus, additional emphasis has been placed on developing a number of output-only modal identification methods for use in structural engineering applications. In this paper, three of these output-only methods peak picking, random decrement, and frequency domain decomposition are modified for implementation in a distributed array of processors embedded within a network of wireless sensor prototypes. The software architecture proposed emphasizes parallel data processing and minimal communication so as to ensure scalability and power efficiency. Using the balcony of a historic theater in metropolitan Detroit as a testbed, this network of wireless sensors is allowed to collect and process acceleration response data during a set of vibration tests. The embedded algorithms proposed in this study are used to autonomously determine the balcony's modal properties with network-derived results found to be comparable to those derived from traditional offline techniques.

A novel approach for the structural identification and monitoring of a full-scale 17-storybuilding based on ambient vibration measurements

For reliable and practical application of structural health monitoring approaches inconjunction with dense sensor arrays deployed on ‘smart’ systems, there is a need todevelop and evaluate alternate strategies for efficient problem decomposition to rapidlyand accurately determine the occurrence, location and level of small changesin the underlying structural characteristics of a monitored system based on itsvibrational signature. Furthermore, there is also a need to quantify the level ofuncertainties in the identified system characteristics so as to have a measurablelevel of confidence in the parameters to be relied on for the detection of genuinechanges (damage) in the monitored system. This study presents the results of twotime-domain identification techniques applied to a full-scale 17-story building, based onambient vibration measurements. The Factor building is a steel frame structurelocated on the UCLA campus. This building was instrumented permanently with adense array of 72-channel accelerometers, and the acceleration data are beingcontinuously recorded. The first identification method used in this study is theNExT/ERA, which is regarded as a global (or centralized) approach, since it dealswith the global dynamic properties of the structure. The second method is atime-domain identification technique for chain-like MDOF systems. Since in thismethod the identification of each link of the chain is performed independently,it is regarded as a local (or decentralized) identification methodology. For thesame reason, this method can be easily adopted for large-scale sensor networkarchitectures in which the centralized approaches are not feasible due to massivestorage, power, bandwidth and computational requirements. To have a statisticallymeaningful results, 50 days of recorded data are considered in this study. Themodal parameter and chain identification procedures are performed over timewindows of 2 h each and with 50% overlap. Using the NExT/ERA method, 12dominant modes of the building were identified. It was observed that variations in thefrequency estimation are relatively small; the coefficient of variation is about 1–2%for most of the estimated modal frequencies. Chain system identification wassuccessfully implemented using the output-only data acquired from the Factorbuilding. Probability distributions of the estimated coefficients of displacement andvelocity terms in the interstory restoring functions (which are the mass-normalizedlocal stiffness and damping values) that were found based on the chain systemidentification are presented. The variability of the estimated parameters due totemperature fluctuations is investigated. It is shown that there is a strong correlationbetween the modal frequency variations and the temperature variations in a 24 hperiod.

A Sinkhole Field Experiment in the Eastern Alps

Because sinkholes are an excellent natural laboratory for studying processes leading to the formation, maintenance, and dissipation of temperature inversions, an extended set of meteorological field experiments was conducted in limestone sinkholes of various sizes and shapes in the eastern Alps during the period from 17 October 2001 through 4 June 2002. The experiments were conducted in an area surrounding the Gruenloch Sinkhole, which in earlier years had recorded the lowest surface minimum temperature in Central Europe, −52.6°C. A dense array of surface temperature sensors and three automatic weather stations were operated continuously during the experimental period, and special experiments enhanced with tethersondes and other equipment were conducted from 2 to 4 June 2002. An overview of the experiments is presented and first results are given.

Classifying functional time series

We consider the problem of classifying a high-dimensional time series into a number of disjoint classes defined by training data. Techniques of this type are an important component of a number of emerging technologies. These include the use of dense sensor arrays for condition monitoring, brain–computer interfaces for communications and control, the detection of moving pedestrians from sequences of images and the study of cognitive function using high-resolution electroencephalography (EEG). We propose a novel approach to problems of this type using the parameters of an underlying functional auto-regression model. We compare the performance of this approach using two contrasting data sets. The first is based on simulated series with different characteristics and sampling schemes and a second based on high-dimensional times series generated by multi-channel EEG. Both experiments show that our approach outperforms conventional time series methods by exploiting low-intrinsic dimensionality (smoothness). In addition, our simulation experiments show that good performance can be maintained for data generated by non-stationary sampling schemes, the latter causing large reductions in the performance of conventional procedures. These experiments suggest that meaningful information can be extracted from high-resolution EEG.

Stimulus novelty and emotion perception: the near absence of habituation in the visual cortex

In rapid serial visual presentation of pictures, an early event-related brain potential component shows enlarged negativity over occipital regions for emotional pictures compared with neutral pictures. The present study examined whether the processing of emotional target pictures varies as a function of stimulus repetition. Accordingly, pictures of erotica, neutral contents, and mutilations were repeatedly presented (90 times) while the electroencephalogram was recorded with a 129 dense sensor array. As in previous studies, emotional pictures were associated with a larger posterior negativity than neutral pictures. Furthermore, differential emotion processing did not vary as a function of stimulus repetition and was similarly expressed across blocks of picture presentation. These findings suggest the near absence of habituation in differential emotion processing during perceptual processing.

Spatial and Temporal Variability of Internal Wave Forcing on a Coral Reef

Abstract The deployment of a dense spatial array of temperature sensors on a coral reef in the Florida Keys provided a unique view of the interaction of cool water incursions generated by internal waves with the three-dimensional reef bathymetry. Water temperature on the reef surface was sampled every 5 s at 100 points on a 100 m by 150 m grid with concomitant measurements of water column velocity and temperature from mid-May through mid-August 2003. Episodic incursions of cool, subsurface water were driven by periods of strong semidiurnal internal tide and higher-frequency internal wave activity. For every time step in the data record the mean profile of temperature as a function of depth is calculated with a 3-m vertical averaging length scale. Subtracting this mean profile from the raw record yields a within depth, horizontal temperature anomaly. Visualization through time of the anomaly mapped onto the measured reef bathymetry reveals episodic variability of thermal patchiness across the reef as well as persistent features associated with reef bathymetry. Variation in the nature of the cooling and resulting thermal heterogeneity among events and seasons suggests multiple modes of cool water incursion ranging from unbroken, tidal period internal waves to packets of higher-frequency, energetic, broken internal bores.

Response Execution in Lexical Decision Tasks Obscures Sex-specific Lateralization Effects in Language Processing: Evidence from Event-related Potential Measures during Word Reading

A common hypothesis about sex differences in language processing attributes these differences to a bilateral contribution of language-related brain areas in females and a left-hemispheric dominated activation in males. However, most imaging studies failed to find such a generalized lateralization effect and reported a left-lateralized activation in both sexes instead. In a previous semantic priming study, we found a sustained (approximately 190-640 ms) bilateral positivity in the ERP waveforms, which was larger for the female group. Word reading and lexical decision were confounded in that study. In the present study we used a delayed response to separate semantic processing from response selection and execution. The modification of the task design, together with a dense sensor array, showed that females developed a bilateral sustaining posterior positivity/frontal negativity during reading/semantic processing. In contrast, males showed an attenuated positivity at left posterior sites and an attenuated negativity at right frontal sites. This sex-specific lateralization effect disappeared during response processing, evoking a bilaterally distributed activation for both sexes (frontal negative and posterior positive), which was larger for the female subjects. We conclude that, at least under specific conditions, language processing evokes a bilateral activation in females and a lateralization effect in males. However, the processing of the response, which is dominated by a 'P300-like' component evoked by this process, evokes a larger activation in both sexes which obscures the sex-specific lateralization effect when semantic processing and response processing are not separated.

Parallel high-throughput microanalysis of materials using microfabricated full bridge device arrays

An array of microfabricated full bridge devices has been implemented for the rapid thermal microanalysis of polymers. In each microelectromechanical system device, four strain gauges were formed in silicon cantilevered microbeams and were configured as a Wheatstone bridge circuit. Glass transition temperatures Tg were measured by the quantitation of the strain produced in the sensor by the stress applied by a polymer layer to the cantilevered microbeams. The measured strain was analyzed as a function of chip temperature for the change in the slope, which was indicative to Tg. Resolution of Tg determinations of amorphous and crystalline polymers was <0.25 °C and <2.0 °C, respectively, being attractive for combinatorial screening of polymers. Our approach is a practical alternative to known methods for Tg determinations because of the immunity to the variations in the amount of deposited material and its viscosity, vapor pressure of employed solvent, and ease of multiplexing into dense sensor arrays.

Sex-related ERP differences in deviance detection

The effect of sex on neural mechanisms of auditory mismatch detection was examined using dense sensor array (128 channel) event-related potential recordings (ERPs). ERPs of 32 right-handed subjects (16 males) were recorded to frequent (85%, 880 Hz) and infrequent (15%, 1480 Hz) tones. There were no sex differences in mismatch negativity (80–180 ms), however, the fronto-central P2 (180–260 ms) was less positive in males ( F=12.56, P<0.005) and the N2 (260–340 ms) was more negative in males ( F=6.28, P<0.05). The increased negativity in males spanning the P2 and N2 may index a top–down process of attention bias towards novelty. This result supports the hypothesis of an adaptive, sexually dimorphic processing of novel events in humans.

Low power serial-parallel dynamic shift register

Serial-to-parallel shift registers have a wide range of applications. These registers are commonly found in communication systems and interfaces between electronic peripherals. Presented is a unique low power area efficient 128 bit serial-to-parallel shift register design that contains only four transistors per stage. The new register uses the capacitive bootstrapping technique to overcome the threshold voltage drop of MOSFETs. This logic family is named non-ratioed bootstrap logic (NRBL). Target applications are dense smart sensor arrays and image sensors.

Detection of a single magnetic microbead using a miniaturized silicon Hall sensor

Using a highly sensitive silicon Hall sensor fabricated in a standard complementary metal–oxide–semiconductor (CMOS) technology, we detect a single magnetic microbead of 2.8 μm in diameter. The miniaturized sensor has an active area of 2.4×2.4 μm2, a sensitivity of 175 V/AT and a resistance of 8.5 kΩ. Two detection methods, both exploiting the superparamagnetic behavior of the bead, are experimentally tested and their performances are compared. This work opens the way to the fabrication of low cost microsystems for biochemical applications based on the use of dense arrays of silicon Hall sensors and CMOS electronics.

Statistical control of artifacts in dense array EEG/MEG studies

With the advent of dense sensor arrays (64-256 channels) in electroencephalography and magnetoencephalography studies, the probability increases that some recording channels are contaminated by artifact. If all channels are required to be artifact free, the number of acceptable trials may be unacceptably low. Precise artifact screening is necessary for accurate spatial mapping, for current density measures, for source analysis, and for accurate temporal analysis based on single-trial methods. Precise screening presents a number of problems given the large datasets. We propose a procedure for statistical correction of artifacts in dense array studies (SCADS), which (1) detects individual channel artifacts using the recording reference, (2) detects global artifacts using the average reference, (3) replaces artifact-contaminated sensors with spherical interpolation statistically weighted on the basis of all sensors, and (4) computes the variance of the signal across trials to document the stability of the averaged waveform. Examples from 128-channel recordings and from numerical simulations illustrate the importance of careful artifact review in the avoidance of analysis errors.

Analog Metal-Oxide-Silicon IC Implementation of Motion-Detection Network Based on a Biological Correlation Model

In this paper, fabrication of a simple analog network for one-dimensional motion detection using the complementary metal-oxide-silicon (CMOS) process is described. The network is constructed with a number of local motion sensors based on biological elementary motion detectors. Experimental and simulation results indicate that the fabricated network computes the local velocity and direction of the moving object in the velocity range of 10-4 m/s to 3.2 ×102 m/s. The results also show that the velocity sensitivity of the network can be changed flexibly. The circuit structure of the network is compact and suitable for highly dense motion sensor arrays.

Dense sensor array topography of the event-related potential to task-relevant auditory stimuli

High spatial density recording and better topographic mapping algorithms have improved the spatial resolving power of the event-related potential (ERP), adding to its already excellent temporal resolution. This study used a 64 channel recording array and spherical spline interpolation to create topographic descriptions of the voltage and current density scalp distributions of the ERP in an auditory oddball paradigm. Frequent (standard) and infrequent (target) tones were presented at a rate of one every approximately 2500 ms to a group of 20 college undergraduates in passive listening and active (count the infrequent tones) task blocks. ANOVAs and topographic analyses were performed on the primary deflections in the `late' portion of the ERP: the P1, N1, P2, N2 and P3. A target minus standard difference wave was also created for each task. The difference wave contained a mismatch negativity (MMN), an N2b and a P3d. The MMN did not differ between the passive and active tasks and had a topography similar to the N1; also the difference wave P3d was topographically similar to the target P3. The N2b, which occurred only to targets in the active condition, and was the first index of target detection, had a scalp distribution consistent with generation in frontal and superior temporal cortex, suggesting activity in cortical areas of selective attention and auditory stimulus representation.

Spatial sampling and filtering of EEG with spline laplacians to estimate cortical potentials.

The electroencephalogram (EEG) is recorded by sensors physically separated from the cortex by resistive skull tissue that smooths the potential field recorded at the scalp. This smoothing acts as a low-pass spatial filter that determines the spatial bandwidth, and thus the required spatial sampling density, of the scalp EEG. Although it is better appreciated in the time domain, the Nyquist frequency for adequate discrete sampling is evident in the spatial domain as well. A mathematical model of the low-pass spatial filtering of scalp potentials is developed, using a four concentric spheres (brain, CSF, skull, and scalp) model of the head and plausible estimates of the conductivity of each tissue layer. The surface Laplacian estimate of radial skull current density or cortical surface potential counteracts the low-pass filtering of scalp potentials by shifting the spatial spectrum of the EEG, producing a band-passed spatial signal that emphasizes local current sources. Simulations with the four spheres model and dense sensor arrays demonstrate that progressively more detail about cortical potential distribution is obtained as sampling is increased beyond 128 channels.

Spatial sampling of head electrical fields: the geodesic sensor net

In studying brain electrical activity from scalp sensors (electrodes), the optimal measurement would sample the potential field over the entire surface of the braincase, with a sufficient density to avoid spatial aliasing of the surface electrical fields. The geodesic sensor organizes an array of sensors, each enclosed in a saline sponge, in a geodesic tension structure comprised of elastic threads. By fixing a sensor pedestal at each geodesic vertex, the geometry of the tension structure insures that the sensory array is distributed evenly across the the accessible head surface. Furthermore, the tension of the network is translated into compression that is divided equally among the sensor pedestals and directed along head-radial vectors. Various geodesic partitioning frequencies may be selected to provide an even surface distribution of the dense sensor arrays (e.g., 64, 128, or 256) that appear to be necessary to provide adequate spatial sampling of brain electrical events.