PaperlayerView exact revision on arXiv

Source-linked AI summary

Dependable Structural Helath Monitoring Using Wireless Sensor Networks

Md Zakirul Alam Bhuiyan, G. Wang, J. Wu, J. Cao

arXiv:1509.06065v1cs.DCeess.SY

TL;DR

WSN-based SHM needs application-specific dependability because sensor faults and resource constraints can undermine monitoring results. The paper presents DependSHM, a distributed framework combining online fault detection with signal reconstruction, and evaluates it through simulations and a prototype. The reported results show that recovery from faulty signals can support dependable WSN-based SHM, subject to stated detection limitations.

  • Problem

    WSN-based SHM lacks application-specific dependability mechanisms that detect sensor faults online and preserve meaningful monitoring under resource constraints.

  • Method

    DependSHM combines distributed automated sensor-fault detection with online Kalman-filter reconstruction of faulty sensor signals.

  • Results

    Simulations and a WSN prototype implementation show that careful recovery from faulty signals can lead to a dependable WSN-based SHM system.

  • Takeaways & Limitations

    DependSHM can maintain SHM quality in the presence of sensor faults without requiring sensor grouping, avoidance, masking, isolation, or replacement.

Abstract

from arXiv · show

As an alternative to current wired-based networks, wireless sensor networks (WSNs) are becoming an increasingly compelling platform for engineering structural health monitoring (SHM) due to relatively low-cost, easy installation, and so forth. However, there is still an unaddressed challenge: the application-specific dependability in terms of sensor fault detection and tolerance. The dependability is also affected by a reduction on the quality of monitoring when mitigating WSN constrains (e.g., limited energy, narrow bandwidth). We address these by designing a dependable distributed WSN framework for SHM (called DependSHM) and then examining its ability to cope with sensor faults and constraints. We find evidence that faulty sensors can corrupt results of a health event (e.g., damage) in a structural system without being detected. More specifically, we bring attention to an undiscovered yet interesting fact, i.e., the real measured signals introduced by one or more faulty sensors may cause an undamaged location to be identified as damaged (false positive) or a damaged location as undamaged (false negative) diagnosis. This can be caused by faults in sensor bonding, precision degradation, amplification gain, bias, drift, noise, and so forth. In DependSHM, we present a distributed automated algorithm to detect such types of faults, and we offer an online signal reconstruction algorithm to recover from the wrong diagnosis. Through comprehensive simulations and a WSN prototype system implementation, we evaluate the effectiveness of DependSHM.

I. INTRODUCTION

WSN-based SHM offers practical deployment benefits but requires application-specific dependability under sensor faults and resource constraints. DependSHM addresses this challenge by detecting faulty sensors online, reconstructing corrupted signals, and evaluating dependable monitoring through simulations and a prototype.

  • Motivation: WSNs offer a low-cost, easily installed platform for monitoring structural changes and issuing early safety alerts.Applications include monitoring damage, cracks, and corrosion in structures such as buildings, bridges, and aerospace vehicles.
  • Problem: Application-specific dependability requires meaningful SHM results despite sensor faults, with online fault detection and immediate recovery actions.WSN resource constraints can reduce data quality, while centralized collection complicates scrutiny of faulty signals.
  • Problem: Existing decision-fusion and threshold-based fault-tolerance schemes use simplified data and do not function properly with SHM methods based on raw measured signals.SHM signal analysis also requires domain knowledge involving vibration, strain, damping, and structural models.
  • DependSHM: DependSHM combines distributed online fault detection using mutual information independence with Kalman-filter reconstruction of faulty sensor signals.The approach is intended for spatially and temporally correlated signals and avoids sensor grouping, masking, isolation, or replacement.
  • Problem: Faulty sensor signals can make undamaged locations appear damaged or damaged locations appear undamaged, producing false positive or false negative diagnoses.Both structural damage and sensor faults can affect the only available sensor-data information.
  • Evaluation: Simulations with real SHM datasets and a TinyOS/Imote2 prototype indicate that careful recovery from faulty signals can produce dependable WSN-based SHM.The prototype was verified on a test structure.

II. RELATED WORK

Prior SHM dependability schemes often do not reliably distinguish faulty sensor readings from structural damage, especially under WSN constraints. The paper therefore develops DependSHM as a distributed framework for detecting faults, reconstructing signals, and improving dependable monitoring.

  • Work from Generic WSN Applications Related to SHM Dependability: Existing WSN fault-tolerance research largely targets event or target detection, with many schemes operating offline and centrally.Common approaches include correlation analysis, 0/1 decisions, value fusion, decision rules, and thresholds.
  • Work from Generic WSN Applications Related to SHM Dependability: SHM dependability differs fundamentally from binary detection or value fusion because the system must identify what occurs in the structure.A 0/1 scheme may flag sensor 5 as faulty but cannot determine whether changes around sensors 4 and 6 indicate faulty signals or damage.
  • Work from Generic WSN Applications Related to SHM Dependability: Existing schemes show high false-positive and false-negative rates, resulting in low system dependability for structural-health detection.The reported weakness remains evident when dependability is treated as the ability to detect structural health conditions.
  • Work from SHM Applications Related to SHM Dependability: Centralized, computationally intensive FDI techniques were generally developed for wired networks, while upper-stream filtering can struggle with high-resolution raw acceleration data and WSN constraints.The constraints include processing large data volumes and satisfying application requirements such as limited energy and bandwidth.
  • Work from SHM Applications Related to SHM Dependability: SPEM’s dependability performance decreases from 87% to 28% as the number of sensors increases, reflecting insufficient dependability support.By contrast, NFMC ranges from 96% to 76% and performs better than the other schemes, but the paper reports that this remains insufficient for confidence in WSN-based SHM.
  • Work from SHM Applications Related to SHM Dependability: DependSHM addresses the gap with distributed fault detection, faulty-data removal, MII-based fault indication, and tolerance through signal reconstruction.The paper also targets simultaneous structural damage and sensor faults and recovery from short-duration faults without requiring sensor masking or isolation.

III. MODELS AND PROBLEM FORMULATION

The paper formulates DependSHM as a distributed WSN-based SHM framework with local sensing, analysis, fault detection, and communication choices. It models sensor faults, neighborhood-based detection, signal prediction, and energy costs under structural-monitoring requirements.

  • A. Network Model: DependSHM distributes acquisition, local analysis, faulty-reading identification, mode-shape computation, and transmission of non-faulty data across sensors and the base station.The proposed framework is contrasted with a traditional WSN-based SHM framework.
  • A. Network Model: Sensors are deployed at candidate structural locations, with communication ranges supporting local neighbor exchange and direct base-station transmission.Rmin maintains local topology for damage and fault detection, while Rmax supports direct communication to the base station.
  • A. Network Model: Adjustable communication range is intended to reduce transmission energy cost while preserving neighborhood-based sensing and fault-detection interactions.Sensors form a communication graph whose edges represent direct communication.
  • 1) Fault Model:: The fault model includes debonding, faulty signals, offset or bias errors, amplification-gain faults, noise faults, and missing or failed nodes.Several faulty sensors may continue communicating and appear operational while returning incorrect values or decisions.
  • 1) Fault Model:: Faulty sensors can produce incorrect readings despite normal communication, so DependSHM exchanges signals among neighboring sensors for fault detection.The model assumes sensors exchange signals with their neighbors.
  • 2) Fault Detection Model:: When sensors near a potential damage area provide mixed-quality signals, the model separates them into disjoint subsets for fault detection.The framework emphasizes continuous monitoring and sensors whose signals change significantly during discrete sampling periods.
  • 2) Fault Detection Model:: MII measures how measurement correlations between presumed non-faulty and faulty sensor groups deviate from a reference correlation model.It uses consecutive signal sets and can support local decisions or recovery, with neighborhood selection controlled by communication range, size, or density.
  • C. Energy Cost Model (cost(ei)): The energy-cost model includes transmission, local computation, sampling, and overhead for fault detection, reconstruction, buffering, and network latency.The total cost is evaluated under shortest-path routing and repeated monitoring rounds.

D. Problem Statement

The paper formulates dependable WSN-based SHM as jointly identifying structural damage and faulty sensors under delivery and connectivity constraints. DependSHM targets lower detection cost and dependable mode-shape information while addressing the effects of faulty measurements on damage localization.

  • Problem formulation: The problem seeks sensors participating in damage detection while separating non-faulty sensors N from faulty sensors F.The sensor subset D must satisfy data-delivery and connectivity constraints.
  • Objectives: The stated objectives are to minimize sensor-set disagreement and communication cost while maximizing dependable mode shape.The framework is divided into distributed event detection, faulty-sensor detection, and faulty-reading reconstruction.
  • SHM basis: Structural damage identification uses changes in local or global responses and relies on mode shape Φ and natural frequency f.Sensors measure responses from ambient or forced excitation using devices such as accelerometers, strain gauges, or displacement sensors.
  • WSN constraints: Traditional centralized state-space computation can be costly for resource-limited WSNs, motivating local processing of structural responses.DependSHM modifies the model for each sensor location and reduces system order through mode coordinates.
  • Problem formulation: Faulty measurements can affect computed mode shapes, particularly when sensors occupy optimal structural locations.The paper therefore considers faulty-sensor detection alongside damage detection and signal reconstruction.

C. General Overview of DependSHM

DependSHM distributes monitoring across neighboring sensors: nodes collect local signals, detect faults with mutual information, reconstruct faulty readings, and send compact mode-shape results to the base station. This local workflow supports online detection and preserves signal reconstruction when a sensor fails or is missing.

  • Distributed workflow: Each monitoring round collects multiple signals at every sensor, which uses MII to identify faulty sensors and triggers neighboring-sensor reconstruction.A faulty sensor may also reconstruct its own signal if it remains operational.
  • Distributed workflow: A missing or failed sensor still receives signal reconstruction, without assuming sensor isolation or replacement.Replacement is treated as costly.
  • Data reduction: Sensors locally compute final Φ and transmit the relatively small result to the BS, which assembles the received mode shapes for damage identification.Each final Φ is described as a number of bytes, whereas raw Φ data can require kilobytes.
  • Fault detection: MII changes when a sensor fault occurs because the faulty signal is absent from reference signals.The algorithm computes MII across sensor-output combinations and uses reductions involving one channel to localize faults.
  • Fault detection: The MII method detects faults across different combinations of sensor faults.The supplied overview reports this capability without specifying a numerical evaluation result.

B. Algorithm 2: Faulty Sensor Detection

Algorithm 2 performs faulty-sensor detection locally using neighboring signals and MII, rather than centralizing all raw measurements at the BS. Its distributed design reduces communication demands, but node disappearance can remain difficult to detect.

  • Scope and limitation: Centralized collection of all sensor signals is unsuitable for resource-constrained WSNs, especially in large-scale deployments.High-frequency SHM data should be reduced before transmission.
  • Distributed detection: Under centralized detection, the BS handles damage and faulty-sensor detection, whereas Algorithm 2 lets each sensor decide locally.The distributed algorithm is explicitly contrasted with centralized detection.
  • Decision rule: A sensor is marked faulty when its local relative-change decision λω exceeds 0.5.The algorithm includes sensor self-reporting and neighbor reporting when decisions are not transmitted.
  • Resource handling: The distributed detector is nearly immediate and online because it requires synchronization only among neighbors within one hop.Detected faulty-signal sets are not forwarded to the BS, lowering communication and energy costs.
  • Scope and limitation: MII does not depend on a particular fault type and can detect the modeled fault kinds.The method may fail to detect a node that is missing or failing.

VI. FAULTY SIGNAL RECONSTRUCTION

DependSHM reconstructs faulty wireless-sensor signals with a Kalman filter based on a structural state-space model. The recursive estimator combines measurements with state predictions using uncertainty-dependent Kalman gain.

  • Kalman-filter reconstruction: DependSHM applies a Kalman Filter technique to reconstruct signals from faulty wireless sensors.The paper uses the KF as a recursive estimator for linear systems and for faulty-signal processing.
  • Model assumptions: The structural model accounts for excitation, transition matrices, and measurement noise in time-discrete, time-invariant cases.Sensor measurements are assumed to contain noise, with zero mean and independent normally distributed noise under the stated model.
  • Kalman-filter reconstruction: At each time instant, the KF estimates the system state by weighting the actual measurement and predicted state.The weights depend on estimated uncertainties, with lower uncertainty receiving higher weight through Kalman gain K_t.
  • Recursive update: The reconstruction process first computes a priori state estimates and then updates posterior estimates using the measured value and Kalman gain.The measurement-estimate difference is weighted during the correction step.
  • Recursive update: The KF updates its gain using priori error covariance and updates that covariance with posterior error covariance.Process and measurement noise covariance values are required for state estimation.

B. Sensor Signal Reconstruction Algorithm

DependSHM reconstructs faulty sensor signals using a state-space model and Kalman filtering, allowing multiple incorrect or missing signals to be recovered within WSN constraints.

  • Reconstruction procedure: The reconstruction algorithm detects faulty sensors from changes and rebuilds the structure’s state-space model for the affected sensor locations.The model uses the structure’s M and K matrices and accommodates faulty or nonexistent sensors.
  • Reconstruction procedure: The algorithm sets the assumed faulty sensors’ measurement-noise covariance values high before solving the state-space equation with a Kalman filter.Increasing the faulty-sensor covariance reduces the filter’s reliance on incorrect measurements.
  • Kalman-filter reconstruction: The Kalman filter reconstructs all signals, including incorrect ones, using the remaining signals and the structural model.The approach can reconstruct more than one signal simultaneously.
  • Scalability and constraints: The number of simultaneously reconstructed signals depends on neighboring nodes in distributed WSNs, all network nodes in centralized systems, and model quality.For a single malfunctioning sensor, the procedure can identify and reconstruct the sensor using only the Kalman filter.
  • Evaluation: The evaluation compares DependSHM with SPEM and NFMC using simulated and real SHM data under injected faults and measures detection, dependability, mode-shape recovery, and energy cost.The simulations include a 100-sensor case and evaluate distributed and centralized processing schemes.

2) Results:

DependSHM achieved stronger fault-detection and dependability results than comparison schemes while recovering distorted mode shapes and reducing communication energy in simulations and a prototype.

  • Simulation results: DependSHM achieved the smallest MII among the evaluated schemes, whereas SPEM performed poorly because centralized processing caused substantial packet loss.Heavy data losses reduced SPEM’s MII performance.
  • Simulation results: 98% detection accuracy was achieved by DependSHM, compared with less than 80% for SPEM and 75%–85% for NFMC.Accuracy is computed as (true positive + true negative)/all.
  • Dependability verification: Faulty sensor signals distorted the actual mode-shape curvature, but signal reconstruction successfully recovered it.The result indicates that monitoring without appropriate fault detection and tolerance is not dependable.
  • Prototype results: The prototype detected injected faults at the fifth and tenth sensors and distinguished persistent mode-shape changes associated with damage from faulty-sensor effects.Neighboring sensors also detected partially affected signals, while recovered neighbor information clarified the damage case.
  • Prototype results: 92% reconstruction quality was obtained relative to fault-free baseline results, while prototype communication energy savings were at least threefold over counterparts.Computation and overhead represented 5%–8% of total energy cost per monitoring round.

VIII. CONCLUSION

The paper concludes that DependSHM combines distributed sensor-fault detection with signal reconstruction to maintain SHM quality on resource-constrained WSNs.

  • Conclusion: DependSHM integrates complementary algorithms for automatic sensor-fault detection and faulty-sensor signal reconstruction.The framework targets both engineering and computer-science requirements for WSN-based SHM.
  • Conclusion: The framework maintains SHM quality in the presence of sensor faults without network maintenance for fault detection and recovery.The paper reports that recovery does not consume significant WSN resources.
  • Future work: Future work will study decentralized computing architectures intended to reduce data traffic while integrating computing-system and structural-engineering techniques.This identifies a future direction rather than a demonstrated result of the current framework.
  • Conclusion: The paper addresses data-intensive SHM and WSN energy cost as central resource constraints.The conclusion frames resource use as an ongoing concern for WSN-based SHM.

APPENDIX A

The appendix defines the energy-cost model and motivates local mode-shape computation over natural-frequency features for fault-tolerant WSN-based SHM.

  • Energy-cost model: The total sensor energy cost includes measurement, computation, transmission, and overhead components.Overhead covers activities such as fault detection, signal reconstruction, local buffering, and network latency.
  • Energy-cost model: Transmission cost is modeled over shortest paths, with sensor power dynamically adjusted between Rmin and Rmax.The routing model tracks paths from sensors toward neighboring nodes or the base station.
  • Energy-cost model: Computational energy depends on processing cycles, processor frequency, hardware constants, and task complexity O(w).The model estimates basic operations such as averages, additions, and multiplications.
  • Structural features: Natural frequency f is a structure-specific vibration characteristic, whereas mode shape Φ represents a spatial deformation pattern.Mode shapes correspond to particular structural responses under excitation.
  • Structural features: Natural frequency is unsuitable for some SHM settings because it is weakly sensitive to minor damage, lacks spatial information, and requires transmitting large data sets.Lost frequency data can substantially affect damage detection, particularly under WSN resource limitations.
  • Structural features: Mode shape Φ contains sensor-level spatial information and its derivatives are sensitive features for damage detection.DependSHM therefore computes mode shapes locally from measurements around each sensor’s vicinity.

APPENDIX C

Appendix C describes the state-space and Kalman-filter basis for estimating structural responses, detecting missing or faulty sensors, and reconstructing faulty signals.

  • State-space model: The structural response model includes mass and stiffness matrices but neglects damping when estimating individual sensor measurements.Undamped mode shapes can nevertheless be accurately estimated when low-damping modes are strongly excited and faults produce sufficiently large response changes.
  • Kalman-filter estimation: The Kalman filter recursively estimates sensor signals from a state-space representation of the structural system.The model uses structural excitation, transition matrices, and measurement-noise terms at each time instant.
  • Limitations: The detection algorithm cannot guarantee detection when a sensor fails, becomes unreachable because of communication constraints, or has an unknown failure reason.The method requires sufficient neighboring information and an available initial frequency for signal identification.
  • Fault detection: KL-KF compares measured and estimated signal distributions, using their divergence as an indicator of sensor faults.If a faulty sensor is excluded from estimation, the divergence is minimal; otherwise, it becomes higher.
  • Fault detection: The KL-KF method detects missing or failed sensors, including the experimentally removed fifth sensor, and enhances detection of pure bias faults.The fifth sensor was identified as missing in the real experiment.
  • Signal recovery: DependSHM uses a recovery algorithm to reconstruct faulty sensor signals after fault detection.The framework targets fault types that produce faulty readings while operating under WSN constraints.

APPENDIX F

Appendix F evaluates WSN-based SHM dependability through sensor-fault detection and structural-event detection. DependSHM outperforms the comparison schemes, while removing recovery sharply reduces event-detection ability.

  • Dependability verification: DependSHM has substantially better sensor-fault detection ability than cSHM, NFMC, and SPEM, with NFMC showing higher detection errors than DependSHM.The comparison is based on fault-detection results under random fault injection.
  • Comparison schemes: NFMC is limited to natural-frequency matching and fails to detect other fault types.Its frequency-matching process can also isolate one or more clusters from the network.
  • Comparison schemes: SPEM has low fault-detection ability because losses of non-faulty readings increase the amount of faulty readings.Both SPEM and NFMC also require substantial resources when recovering from faults.
  • Structural-event detection: 93% to 97.2%: DependSHM’s structural health event detection ability, exceeding SPEM’s 75% to 92%, NFMC’s 74% to 95%, and cSHM’s 87% to 95%.The rate is calculated as 1 − (false positive rates + false negative rates) across 50 simulation runs.
  • Structural-event detection: As faulty sensor nodes increase, structural-event detection becomes lower in NFMC and SPEM than in DependSHM and cSHM.The analysis uses false-positive and false-negative cases to assess health-event detection.
  • Structural-event detection: 65%: structural event detection ability in a system with no recovery from sensor faults.Faulty sensors can produce undetected false-positive and false-negative damage diagnoses, reducing detection ability.
  • Prototype implementation: The proof-of-concept system uses TinyOS and Imote2 motes on a ten-floor test structure, with one mote monitoring horizontal acceleration on each floor.A base station relays wireless sensor data to a PC for structural-health calculation and visualization.

B. Sensor Identified Natural Frequencies

This section reports experimental natural-frequency measurements, fault detection, signal reconstruction, and energy-cost comparisons for DependSHM and related schemes.

  • Sensor-identified natural frequencies: Natural frequencies from the first five sensors are used to create baseline mode shapes for an initially undamaged, fault-free structure.The baseline mode shape is intended to adapt to changing environments and environmental noise.
  • Sensor-identified natural frequencies: The experimental MII values are low across different frequencies, supporting consistency among sensor measurements in the baseline system.The result is reported in Fig. 11.
  • Fault detection and reconstruction: The fifth sensor was detected as faulty in the experimental fault-detection results, and its measured drift was corrected by the estimated signal.The reconstruction is shown for the fifth sensor in Fig. G2.
  • Fault detection and reconstruction: 2% to 4%: MII values for locally processed sensor nodes under the fifth-sensor fault injection, which were not considered faulty.Remarkable signal changes at the fifth and tenth sensors produced the best MII values and indicated improved practical detection accuracy.
  • Energy cost: 0.072 mAh versus 0.197 mAh: DependSHM’s energy cost compared with SPEM for one monitoring round with Td=1.Raw signal transmission to the base station is identified as the major energy consumer, while DependSHM computes locally.
  • Energy cost: 0.165 mAh: average transmission-energy saving attributed to DependSHM’s local computation, which reduces CC2420 radio activity.A sensor without a fault can also save an average of 0.0027 mAh by avoiding reconstruction computation.
  • Energy cost: DependSHM significantly outperforms NFMC over five monitoring rounds, including costs associated with cluster and network maintenance.Fault isolation and transmitting cluster-head mode shapes contribute to the comparison.
  • Energy cost: 5% to 8%: computation and slight-overhead cost for fault detection and signal reconstruction.DependSHM’s distributed processing avoids frequent retransmission and reduces transmitted data relative to centralized approaches.
Loading 1509.06065v1…