In a paper published in the journal Sensors, researchers demonstrated the importance of distributed sensor networks for structural health monitoring (SHM) in seismic contexts by testing self-sensing concrete beams under load, measuring electrical impedance, and assessing cracks with a vision-based system. The study found a high correlation between electrical impedance and applied load, confirming the beams' piezoresistive properties.
Concrete specimen sensorized for electrical impedance measurement and related acquisition board: (a) scheme and pictures of (b) bottom and (c) top views. Image Credit: https://www.mdpi.com/1424-8220/24/2/572
Comparing prediction methods, they found the artificial intelligence (AI) based Prophet model outperformed autoregressive integrated moving average (ARIMA) and seasonal ARIMA with exogenous factors (SARIMAX), with a mean absolute percentage error (MAPE) of less than 1.00%. It highlights the potential of combining electrical impedance sensors, vision-based systems, and AI for enhanced SHM and maintenance predictions.
Related Work
Previous work has emphasized the significance of SHM tools, particularly in seismic contexts, for optimizing structures' lifecycle and performance. Technologies like accelerometers, strain sensors, vision systems, and electrical impedance sensors are key in this field.
Advances include non-contact systems and vibration-based methods, which offer high resolution and cost-effectiveness. Electrical impedance sensors and self-sensing materials enhance conductivity and enable continuous monitoring. Integrating AI and Internet of Things (IoT) technologies allows for real-time data collection and early warnings.
Experimental Setup Overview
The analysts produced six scaled concrete beams using limestone Portland cement (CEM II/A-LL 42.5R) with a water/cement ratio 0.50. Three beams (D, E, F) were used for fracture load measurements, while the remaining three (A, B, C) were used for this study. Aggregates included coarse gravel (10–15 mm), intermediate gravel (5–10 mm), and calcareous sand (0–8 mm).
The analysts added recycled carbon fibers (RCF, 6 mm, 0.05 vol%) and biochar filler (BCH, 0.5 vol%) for self-sensing properties. The mixing involved stirring sand and gravel, then adding cement, BCH, RCF, and finally, water. The mix was poured into 10 cm × 10 cm × 50 cm prismatic molds.
Specimens were cured at 20 ± 1 °C and 50 ± 5% relative humidity. Electrical impedance was measured at intervals over 28 days using sensors embedded during casting. After curing, beams underwent flexural loading tests with a Zwick Roell mechanical press. Tests involved three load levels: 90% of fracture load (t1), fracture load (t2), and load causing ~1 mm crack aperture (t3). Electrical impedance was measured during loading to correlate with applied force, excluding signal peaks during crack formation due to current path interruption.
Cracks were assessed using a vision system with a high-resolution Basler Ace universal serial bus (USB) 3.0 medical camera and Intel RealSense D435i depth sensor. A neural network, UNet, trained on a dataset of real concrete crack images, segmented and measured cracks. The AI model provided sub-pixel resolution for crack aperture width in millimeters. The vision system performed measurements after load tests, noting partial crack closure due to load removal.
The monitoring system used a cloud-based pipeline to collect time series data from concrete specimens. Electrical impedance and force data-informed ARIMA, SARIMAX, and Prophet models. Models trained with 90% of the data and validated with cross-validation predicted future impedance values. Performance metrics included mean absolute error (MAE), MAPE, root mean square error (RMSE), and correlation. The approach enabled the development of a real-time forecasting pipeline and potential early warning system for structural health monitoring.
Electrical Impedance Analysis
The team outlines the findings of electrical impedance measurements conducted during the curing process, emphasizing the increasing trend attributed to material hydration. Filtering techniques are highlighted for enhancing data quality by reducing noise, facilitating correlation evaluation with force signals, and improving applicability for prediction models.
Moreover, the impact of flexural load application on electrical impedance is discussed, with observations indicating a parallel increase due to widening inter-electrode spacing. Despite fluctuations, the correlation between electrical impedance and force signals remains consistently strong, showcasing the material's efficient piezoresistivity property. Additionally, crack formation induces distinct peaks in electrical impedance, enabling the prompt detection of structural changes.
The analysis employs advanced computer vision tools for defect identification and segmentation. The method offers precise crack width measurements across multiple specimen faces. The discussion underscores the alignment of defect assessment methodologies with contemporary standards, highlighting the accuracy and reliability of the employed approach.
Moreover, the study comprehensively evaluates predictive crack detection models, including ARIMA, SARIMAX, and Prophet. Detailed performance metrics analysis shows that Prophet consistently outperforms traditional models, demonstrating superior accuracy and precision across various test scenarios. Notably, Prophet's ability to capture nuanced temporal patterns distinguishes it from linear models like ARIMA and SARIMAX, reinforcing the growing preference for advanced AI approaches in forecasting complex datasets with external variables.
Conclusion
In summary, this study explored the use of self-sensing materials and sensors to measure electrical impedance for potential applications in seismic monitoring. The findings revealed that electrical impedance measurements effectively track external loads and cracking phenomena while integrating self-sensing materials, which enhances a structure's health-monitoring ability.
Vision-based techniques offer precise damage assessment, and AI models demonstrate superior forecasting capabilities, enabling the implementation of early warning systems. This proposed solution could be vital for building resilience in seismic regions and other disaster-prone areas. However, ensuring reliable connectivity and system maintenance remains critical for long-term monitoring solutions.
AI Redefines Open Innovation with Enhanced Practices and New Models