01 · Low-cost sensing · Monitoring
Automated bridge monitoring with low-cost accelerometers
The problem
Industrial-grade sensors are expensive. That cost creates significant barriers to the widespread use of structural health monitoring, especially in projects with tight budgets.
What was done
The low-cost LARA accelerometer was compared with a traditional commercial accelerometer (PCB 607A61) during a field campaign on the Andoain Road Bridge. The study examined noise density, internet-based synchronisation of the data and the bridge's dynamic characteristics, and developed an automated operational modal analysis procedure that integrates data acquisition and the modal calculation. The measured data were then used to calibrate a finite element model of the bridge.
What it means in practice
According to the authors, LARA closely matched the commercial accelerometer, with minimal differences in natural frequencies and high Modal Assurance Criterion (MAC) values, and the calibrated model aligned closely with the measured modal properties. For an owner, this means vibration-based monitoring and model calibration with low-cost sensors, and with processing that runs automatically.
Delgado Zhagui, E.; Komarizadehasl, S.; Torralba, V.; Diaz-Rozo, J.; Turmo, J. (2025). Cost-effective bridge health monitoring via automated operational modal analysis using low-cost adaptable and reliable accelerometers. Structure and Infrastructure Engineering, 1–16. https://doi.org/10.1080/15732479.2025.2521479
03 · Low-cost sensing · Monitoring
Commercial sensors for training, low-cost sensors for permanent monitoring
The problem
Structural health monitoring is crucial for the safety of essential infrastructure, but the high cost of traditional sensors often limits its use.
What was done
Commercial sensors were installed temporarily next to low-cost accelerometers on a bridge to support the training of a deep learning model based on long short-term memory (LSTM) neural networks. Once training is complete, the commercial sensors are removed and the calibrated low-cost accelerometers stay in place for continuous monitoring. In the case study, the bridge carried six low-cost and six commercial sensors, and the mode shapes and natural frequencies from the commercial, low-cost and calibrated low-cost sensors were compared.
What it means in practice
The authors report that this comparison corroborated the approach. For an owner, the commercial sensors are needed only during training, while permanent monitoring runs on low-cost hardware.
Emadi, S.; Komarizadehasl, S.; Xia, Y. (2025). Application of Intelligent Low-Cost Accelerometers for Bridge Monitoring With a Deep Learning Approach. Structural Control and Health Monitoring, 2025, 9835353. https://doi.org/10.1155/stc/9835353
04 · Low-cost sensing · Damage detection
An affordable inclinometer for damage detection on bridges
The problem
Structural damage detection with inclinometers is getting wide attention from researchers, but their high price limits them to unique structures with a relatively high monitoring budget.
What was done
LARA, a low-cost inclinometer that combines five gyroscopes and five accelerometers, was developed with an Internet of Things microcontroller for wireless data streaming and free commercial software for data acquisition. Its accuracy, resolution, Allan variance and standard deviation were studied with different numbers of combined circuits, and its readings were compared in the laboratory with numerical slope calculations and with a commercial inclinometer (HI-INC).
What it means in practice
In a load test on a simple beam model, LARA reached an accuracy of 0.003 degrees. The authors conclude that its affordability and high accuracy make LARA applicable for structural damage detection on bridges using inclinometers. These results come from laboratory tests.
Komarizadehasl, S.; Komary, M.; Alahmad, A.; Lozano-Galant, J.A.; Ramos, G.; Turmo, J. (2022). A Novel Wireless Low-Cost Inclinometer Made from Combining the Measurements of Multiple MEMS Gyroscopes and Accelerometers. Sensors, 22(15), 5605. https://doi.org/10.3390/s22155605
05 · Long-term monitoring
Flagging when a long-term tilt sensor may need recalibrating
The problem
Low-cost sensors make monitoring of buildings and infrastructure affordable, but their long-term durability and reliability is a critical concern, and no published method existed to check that they keep performing over time.
What was done
An artificial-intelligence method was developed for evaluating the long-term performance of low-cost inclinometers, using a low-cost adaptable reliable anglemeter. It automatically compares the inclinations measured on site with the values predicted for the real environmental conditions, and flags when the discrepancy exceeds a predefined statistical threshold. It was demonstrated in a long-term study on a real steel frame in Spain.
What it means in practice
A flag may mean that the sensor has lost accuracy and needs recalibrating, or that the structure itself shows an anomaly. It addresses a gap the authors identify: the literature presented no solution for ensuring the long-term performance of low-cost sensors.
Lozano, F.; Emadi, S.; Komarizadehasl, S.; González-Arteaga, J.; Xia, Y. (2024). Enhancing performance evaluation of low-cost inclinometers for the long-term monitoring of buildings. Journal of Building Engineering, 87, 109148. https://doi.org/10.1016/j.jobe.2024.109148
06 · Damage detection
Low-cost accelerometers and fibre optics working together to detect damage
The problem
Distributed fibre optic systems such as Distributed Acoustic Sensing (DAS) are accurate, but they are costly, computationally demanding and energy-hungry. Low-cost Arduino-based sensors are gaining popularity, and this work compared the accuracy of one of them with that of a DAS system in detecting structural damage.
What was done
The design of the LARA accelerometer, which uses Arduino and Raspberry Pi technology, was improved. LARA and a DAS system were then used in the laboratory to track changes in the natural frequencies of a beam model with induced localised damage, and LARA was assessed as a trigger for the DAS system.
What it means in practice
Both LARA and DAS identified changes in the natural frequencies of the damaged structure, with deviations as small as 3.68 %. LARA therefore demonstrated its potential as a trigger for DAS, significantly reducing the computational demands while enriching the analysis. The tests were carried out in a laboratory.
Komarizadehasl, S.; González Jiménez, M.A.; Pérez Casas, J.M.; Lozano-Galant, J.A.; Turmo, J. (2024). Eigenfrequency analysis using fiber optic sensors and low-cost accelerometers for structural damage detection. Engineering Structures, 318, 118684. https://doi.org/10.1016/j.engstruct.2024.118684
07 · Drive-by monitoring
Bridge frequencies from one crossing of an instrumented vehicle
The problem
Conventional monitoring of large or long-span bridges is held back by installation complexity, maintenance demands and high costs.
What was done
A drive-through method was developed in which several MEMS sensors mounted on a vehicle turn it into a mobile sensing platform. The method was validated on the Yangpu and Nanpu cable-stayed bridges in Shanghai under normal traffic, and combining the data of several sensors improved the signal-to-noise ratio.
What it means in practice
Bridge natural frequencies were extracted from a single crossing under normal traffic, with sensors mounted on a passing vehicle, significantly reducing the need for multiple passes. The authors present it as a practical, cost-effective and non-intrusive option for monitoring bridges at large scale.
Komarizadehasl, S.; Shen, Z.; Xia, Y.; Song, M.; Turmo, J. (2025). An innovative drive-through approach for structural testing and experimental insights from two cable stayed bridges. Developments in the Built Environment, 22, 100653. https://doi.org/10.1016/j.dibe.2025.100653
08 · Drive-by monitoring
Getting more from a single drive-by pass by combining sensors
The problem
Drive-by monitoring could make bridge monitoring scalable, but a single pass of a vehicle typically gives a short, noisy record that is strongly affected by the vehicle's own dynamics.
What was done
Three accelerometer nodes were mounted on the roof of a passenger car, which crossed the three-span Viaduct EN203 in Viana do Castelo, Portugal, three times. The results were compared with a stationary reference test using PCB 356A16 accelerometers mounted on the bridge. Two ways of combining the sensors were tested: adding their signals together (spatial summation) and joining them end to end (temporal concatenation).
What it means in practice
Joining the records end to end recovered the bridge's fundamental frequency from a single pass in each of the three crossings, with errors of 2.54–5.72 % (mean 3.60 %), even though the car sensors had a noise floor 6.6 times higher than the reference. Adding the signals came close to the reference under stationary excitation (0.61 % error for the first mode, 1.43 % for the second) but failed completely in drive-by conditions. The practical lesson: how well the sensors agree with one another (their coherence) should decide how their data are combined, and the paper gives a quantitative basis for designing sensor arrays for single-pass monitoring.
Komarizadehasl, S.; Shen, Z.; Xia, Y.; Matos, J. (2026). Single drive-by bridge modal frequency identification using multi-sensor data fusion. Structures, 91, 112685. https://doi.org/10.1016/j.istruc.2026.112685
09 · Drive-by monitoring
Drive-by measurement against fixed sensors on two large truss bridges
The problem
Drive-by monitoring, from a vehicle crossing the bridge, is gaining attention as a cost-effective alternative to conventional direct measurement of bridge vibration, and this study compares the two in the field.
What was done
Direct and drive-by measurements were compared on two large steel truss bridges in Vietnam, Thang Long Bridge and Chuong Duong Bridge. Fixed sensors on the deck recorded vertical acceleration, while accelerometers on a test vehicle recorded crossings at controlled speeds from 20 to 50 km/h.
What it means in practice
On Thang Long Bridge, the fixed sensors identified nine vertical bending modes below 5 Hz; the vehicle consistently captured the first two, with errors as low as 0.2 % at 30–35 km/h and up to about 15 % at 50 km/h. On Chuong Duong Bridge, two dominant modes were identified with errors of 1.5–9.9 %, best at speeds near 30 km/h. Under suitable operating conditions, drive-by monitoring can provide reliable estimates of low-order vertical bending frequencies; higher-order and torsional modes remain hard to identify, and vehicle speed matters.
Tran, M.; Cam Nhung, Nguyen Thi; Van, Thuc Ngo; Komarizadehasl, S. (2026). Innovative experimental assessment of direct and drive-by monitoring on two truss bridges. Measurement, 278, 121693. https://doi.org/10.1016/j.measurement.2026.121693
10 · From data to reports
From raw monitoring data to an automated report
The problem
Monitoring reports are central to bridge maintenance and management, but producing them still depends on manual analysis.
What was done
An automated framework was built in which a Python-based system processes heterogeneous monitoring data and a domain-specific multimodal large language model, fine-tuned for the task, acts as the reasoning component of an AI agent. Once the data are standardised and the system configured, it carries out the analysis, interprets the graphs and writes summaries at several levels of detail. It was verified on the Mingzhou Bridge and applied to 10 in-service bridges with 5466 GB of data in total.
What it means in practice
Across these applications, the system processed about 2500 figures in a total of 308 minutes and generated about 837,000 words on typical personal computer resources. The authors present this as demonstrating the framework's potential for scalable, automated analysis, interpretation and report generation, once the data have been standardised.
Xia, Y.; Pan, C.; Gong, F.; Lozano Galant, F.; Zhang, L.; Komarizadehasl, S.; Sun, L. (2026). Bridge monitoring data analysis and automated report generation using multimodal large language model. Structural Health Monitoring. https://doi.org/10.1177/14759217261453811