Structural Health Monitoring

A structured learning path from SHM fundamentals through advanced data-driven damage detection. Master monitoring systems for bridges, buildings, and critical infrastructure.

Start Learning Full Handbook

Level 1

Beginner — Introduction to SHM

Start here if you are new to structural health monitoring.

Introduction to SHM

Structural Health Monitoring (SHM) is the continuous or periodic measurement of structural response to assess condition and detect damage at an early stage. SHM complements traditional visual inspection by providing quantitative data on structural behavior under operational and environmental loads. Benefits include early damage detection (reducing repair costs by 20-40%), extended service life through condition-based maintenance, reduced lifecycle costs, and improved safety for critical infrastructure such as long-span bridges, high-rise buildings, and dams.

The SHM hierarchy has four levels defined by Rytter: Level 1 confirms that damage exists (anomaly detection), Level 2 identifies the damage location, Level 3 quantifies damage severity, and Level 4 predicts remaining useful life (prognosis). Most operational SHM systems currently operate at Level 1-2, with Level 4 being an active research area. Effective SHM requires understanding the structure's baseline behavior, distinguishing damage-induced changes from environmental effects (temperature, humidity, traffic), and selecting appropriate sensor types and data acquisition strategies.

Sensor Technologies for SHM

SHM relies on a wide range of sensors. Strain gauges: foil resistance gauges (120-350 ohm, gauge factor 2.0), vibrating wire strain gauges (long-term stability, resolution 1 microstrain), fiber optic FBG sensors (multiplexing capability, corrosion resistant, up to 80 sensors per fiber). Accelerometers: piezoelectric (ICP type, frequency range 0.5Hz-10kHz), MEMS capacitive (low cost, DC response, range +/-2g to +/-50g), servo force-balance (high sensitivity, 0-100Hz, used for seismic monitoring). Displacement transducers: LVDT (linear range up to +/-500mm), laser Doppler vibrometers (non-contact, remote measurement), GPS (cm-level static, mm-level with differential correction).

Selection criteria for SHM sensors include: sensitivity (minimum detectable signal), measurement range, frequency bandwidth, environmental robustness (temperature range, humidity, corrosion resistance), power consumption (especially for remote wireless monitoring), long-term stability (drift over years), and cost per channel. Emerging technologies include distributed acoustic sensing (DAS) using fiber optic cables for pipeline and railway monitoring, smart aggregates with embedded piezoelectric sensors, and wireless smart sensor networks with onboard processing for real-time damage detection.

Data Acquisition Systems

Data acquisition (DAQ) systems convert analog sensor signals to digital data for analysis. Key components include: signal conditioning (amplification to match ADC range, low-pass anti-aliasing filtering at Nyquist frequency, excitation voltage for strain gauges and RTDs), analog-to-digital conversion (resolution 16-24 bits, sampling rate determined by highest frequency of interest), multiplexing (scanning multiple channels sequentially), and data transmission (wired Ethernet or fiber optic, wireless Wi-Fi/LoRa/cellular). Synchronization across multiple DAQ units is critical for modal analysis — achieved through GPS time stamping or dedicated timing cables.

Sampling rate selection follows the Nyquist criterion: sample at least twice the highest frequency of interest. For modal analysis of buildings (first mode typically 0.5-5Hz), a sampling rate of 50-200Hz is adequate. For bridge cable vibration (frequencies up to 20Hz), 100-500Hz is used. For acoustic emission (high-frequency elastic waves up to 500kHz), rates of 1-10MHz are required. Data storage requirements: a 100-channel system sampling at 200Hz, 24-bit resolution generates approximately 500 MB/day. Continuous monitoring over years requires data compression strategies: event-triggered recording, feature extraction on-site, and hierarchical data storage (raw data for short term, features for long term).

Level 2

Intermediate — Monitoring Methods and Analysis

Build on fundamentals with practical monitoring and analysis techniques.

Vibration-Based Monitoring

Vibration-based SHM uses changes in modal parameters (natural frequencies, mode shapes, damping ratios) to detect and locate damage. Natural frequencies are the most commonly used damage indicator because they can be measured accurately from a single sensor location. However, frequency shifts due to damage are typically small (1-5% for significant damage), and environmental effects (temperature 0.1-0.3 Hz/10°C for concrete bridges) can mask damage-induced changes. Operational Modal Analysis (OMA) extracts modal parameters from ambient vibration (wind, traffic, microtremors) without artificial excitation. Methods include peak picking on averaged FFT spectra, Frequency Domain Decomposition (FDD), and Stochastic Subspace Identification (SSI).

Mode shape-based methods offer spatial damage localization. The mode shape curvature method computes the second derivative of mode shapes, which is highly sensitive to local stiffness reductions. The Modal Assurance Criterion (MAC) quantifies the correlation between two mode shape vectors — values close to 1 indicate good agreement, while deviations suggest damage. Coordinate Modal Assurance Criterion (COMAC) localizes the damage to specific DOF. Strain mode shapes are more sensitive to local damage than displacement mode shapes. Compensation for temperature effects is essential: regression models, PCA, or cointegration remove environmental variability before damage detection.

Static and Quasi-Static Monitoring

Static monitoring measures long-term deformations, strains, and displacements under quasi-static loads (traffic, temperature, creep). Total stations and GNSS (Global Navigation Satellite Systems) measure 3D displacements with mm-level accuracy for long-span bridges and dams. Tiltmeters (electrolytic, MEMS) monitor rotation changes with resolution 0.001 degree. Hydrostatic leveling systems measure differential settlement across large structures using water-filled tubes and pressure transducers. Fiber optic sensors (FBG, Brillouin scattering) provide distributed strain and temperature measurements along the entire sensor length (up to 50km for Brillouin).

Controlled load testing applies known loads (test trucks for bridges, water filling for tanks) and measures structural response to calibrate analytical models and verify design assumptions. Temperature-induced movements dominate long-period measurements — expansion joints in bridges move 20-50mm daily due to temperature cycles. Creep and shrinkage in prestressed concrete bridges cause gradual deflection changes over years. Separating these time-dependent effects from damage-induced changes requires long-term baseline data (typically 2-3 years) and physics-based models. Statistical process control (SPC) with control charts (Shewhart, CUSUM, EWMA) detects significant deviations from baseline behavior.

Damage Detection Algorithms

Damage detection algorithms extract features from sensor data that are sensitive to damage. Time-domain methods: Auto-Regressive (AR) and Auto-Regressive Moving Average (ARMA) models fit a time-series model to the structural response under undamaged conditions; damage is indicated by significant changes in model coefficients or residual errors. Novelty (outlier) detection identifies data points that deviate significantly from the baseline distribution using Mahalanobis distance or one-class SVM. Frequency-domain methods: damage indices from Frequency Response Function (FRF) changes, damage detection using transmissibility functions (ratio of output spectra between sensor pairs).

Wavelet transform analysis provides time-frequency localization, detecting transient events (crack propagation, impact) that are invisible in frequency spectra. Machine learning approaches: autoencoders learn a compressed representation of undamaged structural response — reconstruction error increases when unseen damage patterns appear. Convolutional Neural Networks (CNNs) detect cracks and corrosion from images with accuracy exceeding 95% in controlled conditions. Support Vector Machines (SVMs) classify damage states from extracted features. The key challenge is obtaining labeled training data for damaged conditions — typically addressed through numerical simulation, laboratory experiments, or transfer learning from similar structures.

Level 3

Advanced — Application Systems and Data Management

For senior students and practicing engineers.

Bridge Monitoring Systems

Bridges are the most common SHM application. Suspension bridges: monitoring cable tension (vibration method, elastomagnetic sensors), wind-induced vibration (accelerometers on deck and towers), deck deflection (GPS, total stations), fatigue-critical details (strain gauges at welded connections). Cable-stayed bridges: stay cable force monitoring (vibration frequency method: T=4mf²L²), cable damping (rain-wind induced vibration control), deck profile (longitudinally distributed tiltmeters). Girder bridges: strain at critical sections (midspan, supports), bearing displacement monitoring (LVDT, potentiometer), corrosion monitoring in prestressing tendons (acoustic emission for wire break detection).

Long-span bridges typically have 100-500+ sensor channels. The Tsing Ma Bridge in Hong Kong has over 1000 sensors including accelerometers, strain gauges, GPS, anemometers, and temperature sensors. Data from these systems guides maintenance decisions: threshold alarms trigger inspections when strain, displacement, or acceleration exceed predetermined limits. Load rating verification uses measured strains under known traffic loads to update analytical load ratings. Fatigue life assessment uses rainflow cycle counting on strain histories to compute cumulative damage per Palmgren-Miner rule. SHM-informed bridge management reduces inspection frequency and targets maintenance resources to actual needs.

Building and Infrastructure Monitoring

Tall buildings: wind-induced vibration monitoring to verify occupant comfort criteria (acceleration limits per ISO 10137: 10-15 milli-g for residential, 20-25 milli-g for office). Seismic monitoring uses strong-motion accelerometer arrays at basement, mid-height, and roof levels to record earthquake response. Foundation settlement monitoring involves hydrostatic leveling arrays and tiltmeters. Dams: seepage monitoring (weirs, flow meters), pore pressure (piezometers), crack opening (crack meters, joint meters), tilt (pendulum, invert level). Tunnels: convergence monitoring (laser scanning, total station arrays), lining stress (pressure cells), groundwater pressure (piezometers), and fire detection (linear heat sensors).

Pipelines: leak detection (acoustic emission, negative pressure wave, fiber optic temperature sensing for gas leaks), pressure monitoring for surge detection, corrosion monitoring (electrochemical noise, linear polarization resistance, coupon weight loss). Offshore platforms: structural integrity monitoring (accelerometers for global vibration, strain gauges for wave-induced stress, tilt for platform inclination), cathodic protection monitoring (reference electrodes for potential measurement), riser and mooring line monitoring (tension, angle, fatigue). Wind turbines: tower acceleration and tilt, blade strain (FBG sensors, accelerometers), gearbox vibration monitoring, foundation settlement and scour monitoring for offshore turbines.

Data Management and Decision Support

SHM generates large volumes of data that require systematic management. Data compression strategies: downsampling (store only features rather than raw time series), event-triggered recording (store only when thresholds exceeded), and in-network processing (compute features at the sensor node). Data quality assessment: detect sensor faults (drift, bias, noise, saturation, disconnection), missing data imputation (interpolation, PCA, matrix completion), and signal processing (detrending, filtering, normalization). Threshold setting for alarms: probabilistic methods (assume Gaussian distribution, threshold at mean + 3 sigma for 99.7% confidence), extreme value statistics for rare events, and time-dependent thresholds accounting for environmental cycles.

SHM-informed maintenance transitions from time-based to condition-based strategies. Maintenance decision criteria: safety (risk of failure), economic (lifecycle cost optimization), and operational (minimize disruption). Digital twin integration creates a real-time virtual replica of the structure that receives monitoring data for model updating and predictive simulations. Finite element model updating uses measured modal properties (frequencies, mode shapes) or static strains to calibrate uncertain parameters (stiffness, boundary conditions, mass). The updated model predicts response under extreme events and estimates remaining fatigue life. Integration with Building Information Modeling (BIM) provides a common data environment for visualization and facility management.

Practice Exercises

Exercise 1: Frequency-Based Damage Detection

A simply supported beam of span 10m with EI=200 MNm2 and mass per unit length 5000 kg/m. Damage is simulated by a 20% reduction in EI over the middle 1m section. Calculate the first 3 natural frequency shifts due to this damage. Are these shifts detectable with typical accelerometer accuracy of 0.1% frequency resolution? Discuss implications for practical SHM.

Exercise 2: Operational Modal Analysis

A 5-story shear building with floor masses of 50t and inter-story stiffness of 200 MN/m per floor. Simulate acceleration records under ambient excitation (white noise base acceleration). Apply OMA using peak picking on averaged FFT spectra (Welch method, 50% overlap, Hanning window). Identify the first 3 natural frequencies and compare with the theoretical values. What minimum sampling rate and record length are needed?

Exercise 3: Strain-Based Bridge Load Rating

A steel girder bridge is instrumented with strain gauges at midspan. Under a test truck weighing 320kN, the measured strain is 85 microstrain. The steel section modulus S=0.012 m3 and E=200 GPa. Calculate the applied moment, load distribution factor to the instrumented girder, and compare with AASHTO LRFD distribution factor equations. Determine the bridge load rating (RF) for an HL-93 design truck.

Exercise 4: Anomaly Detection in Monitoring Data

A 30-day record of hourly temperature and strain measurements from a bridge girder. Apply a 24-hour moving average filter to remove diurnal cycles. Remove the temperature effect using linear regression of strain vs. temperature. Compute the residual strain and identify outlier events exceeding plus or minus 3 standard deviations. What physical events (temperature, traffic, damage) might cause these anomalies?

References

Balageas, D., Fritzen, C.P., and Guemes, A. Structural Health Monitoring. ISTE, 2006.
Farrar, C.R. and Worden, K. Structural Health Monitoring: A Machine Learning Perspective. Wiley, 2013.
Adams, D. Health Monitoring of Structural Materials and Components. Wiley, 2007.
Chang, F.K. Structural Health Monitoring: Current Status and Perspectives. DEStech, 2005.
Doebling, S.W. et al. Damage Identification and Health Monitoring of Structural and Mechanical Systems. Los Alamos National Laboratory Report, 1996.
Civil Engineering Handbook — SHM sensors and monitoring methods chapter.
Engineering Formula Library — Modal analysis and signal processing formulas.
Engineering Standards Reference — Bridge inspection, ISO 10137 vibration criteria.
Engineering Glossary — Definitions of SHM and NDT terms.