image missing

International Journal of Innovation and Entrepreneurship

(ISSN: 2753-6297) Open Access Journal
Rss Feed:

Int. J. Innov. Entrep. , 3(1), 4; doi:10.56502/IJIE3010004

Article
AI-Enhanced Structural Health Monitoring: Advancing Accuracy and Predictive Capabilities in Infrastructure
Luca Wang 1
Crestwood Preparatory College, North York, Toronto, ON M3A 2T7, Canada; luca.wang@crestwood.on.ca
How to cite: Wang, L. AI-Enhanced Structural Health Monitoring: Advancing Accuracy and Predictive Capabilities in Infrastructure. Int. J. Innov. Entrep., 2024, 3(1): 4; doi:10.56502/IJIE3010004.
Received: 13 September 2024 / Accepted: 18 September 2024 / Published: 27 September 2024

Abstract

:
Structural health monitoring (SHM) is crucial for ensuring the safety and longevity of infrastructure by detecting and assessing structural damage through advanced sensing and analytical techniques. The integration of artificial intelligence (AI) into SHM offers significant advantages over traditional methods, enabling real-time, continuous monitoring and predictive maintenance. This approach enhances the accuracy of structural issue detection, allowing for timely intervention before issues become critical, thereby reducing maintenance costs and improving safety. This article examines the application of various AI methods, including supervised, unsupervised, and deep learning, to optimise SHM under different conditions. By exploring hybrid approaches that combine these AI techniques, this study demonstrates how such integrations can improve the accuracy, adaptability, and reliability of SHM systems, ultimately contributing to safer and more cost-effective infrastructure management.
Keywords:
structural health monitoring; supervised learning; unsupervised learning; deep learning

Introduction

Structural health monitoring (SHM) is a rapidly evolving field dedicated to ensuring the safety and functionality of infrastructure by detecting damage, assessing structural integrity, and predicting failures. SHM systems utilise a combination of sensors, data acquisition technologies, and analytical techniques to monitor structures such as bridges, buildings, aircraft, and pipelines [1]. Conventional SHM methods include visual inspection [2], fibre optic sensing [3], and vibration analysis [4]. Although these methods could provide immediate results, long diagnoses, and trend analysis, they have limitations, as shown in Figure 1. SHM has been improved through artificial intelligence (AI) techniques, such as supervised learning, unsupervised learning, and deep learning, which analyse vast amounts of data to detect anomalies and structural issues with increased speed and accuracy [5]. Supervised learning, unsupervised learning, and deep learning are foundational concepts in machine learning, each with distinct approaches and applications. Supervised learning involves training a model on labelled data, where the input–output pairs are known. This model learns to predict the output from the input by minimising the error between its predictions and the actual labels. This method is widely used in applications such as image classification, speech recognition, and medical diagnosis [6]. Unsupervised learning, on the other hand, deals with unlabelled data. This model tries to uncover hidden patterns or intrinsic structures from the input data and is commonly used in clustering and anomaly detection tasks, such as identifying abnormal events in video surveillance [7]. Deep learning is a subset of machine learning that utilises neural networks with multiple layers to model complex patterns in data. Deep learning models can be applied using both supervised and unsupervised learning approaches, excelling in areas such as computer vision, natural language processing, and autonomous systems [8]. Convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are two types of neural networks that are widely used in artificial intelligence. CNNs are primarily used for image and video recognition due to their ability to capture spatial features through convolutional layers. This technique has advantages in dealing with tasks such as object detection, image classification, and real-time environmental monitoring because it can efficiently extract and process visual information [9]. RNNs, on the other hand, are designed for sequence data, making them well-suited for tasks like speech recognition, language translation, and time series prediction. They have a feedback loop that allows them to remember previous inputs, enabling them to make predictions based on past data, which is crucial in applications such as natural language processing and autonomous systems [10].
Image001
Figure 1. Comparison of the pros and cons between different SHM techniques.
Figure 1. Comparison of the pros and cons between different SHM techniques.
Image001

Main

Supervised Learning in SHM

Supervised learning significantly enhances structural health monitoring by improving the detection, localisation, and assessment of structural damage through pattern recognition and predictive modelling. Supervised learning algorithms, such as support vector machines, convolutional neural networks, and other neural networks, are trained on labelled datasets, allowing them to identify damages like cracks, stiffness loss, and corrosion with high accuracy. For instance, supervised learning models have been successfully applied in monitoring systems for bridges, pipelines, aircraft structures, and railways to accurately diagnose and predict structural failures by analysing vibrations, signals, and images. These models are particularly useful in automating damage detection processes, thereby reducing the time and human effort that are needed for regular inspections, enhancing the safety and maintenance of critical structures [11,12]. However, relying solely on supervised learning has drawbacks. One of the main limitations is the need for extensive labelled datasets, which can be costly and time-consuming to collect, especially when dealing with diverse structural conditions and types of damage. The performance of these models heavily depends on the quality and quantity of labelled data, making them less effective when facing new, unseen damage types. Additionally, supervised models are sensitive to data noise and may struggle with real-world conditions where data variability is high. Moreover, they typically require retraining when significant changes in structural conditions occur, which can lead to increased computational costs and maintenance requirements [11]. Supervised learning significantly improves structural health monitoring by automating damage detection through pattern recognition and predictive modelling, but it requires extensive labelled datasets and struggles with new or noisy data. While effective in diagnosing structural issues, the models can be costly to maintain and may need tuning when structural conditions change.

Unsupervised Learning in SHM

Unsupervised learning methods have significantly advanced structural health monitoring by detecting anomalies and damage without labelled training data, which is a major advantage over supervised learning. These methods are particularly effective in SHM because they can learn from normal operational data of structures, identifying deviations that may indicate damage and structural issues. Techniques such as autoencoders are commonly used for anomaly detection, allowing for the continuous assessment of structures like bridges and buildings based on vibration data or sensor inputs. Unsupervised learning can be particularly useful in environments where obtaining labelled data is costly or impractical, providing an efficient way to monitor structural health in near-real time and automatically flag potential issues before they escalate into critical failures [13,14]. However, unsupervised learning has its limitations. One of the main drawbacks is the challenge of distinguishing between harmless anomalies and actual damage, which can lead to false alarms and missed detections. This is because these models are not trained on specific damage cases, making them sensitive to events that are not necessarily harmful. Additionally, these methods usually struggle with adapting to changing environmental conditions, which can introduce noise into the data and affect the model’s performance. Another challenge is the difficulty in interpreting the results, as unsupervised models can identify an anomaly without clearly explaining its cause, requiring further analysis by human experts to confirm and act on the findings [14,15]. Unsupervised learning methods enhance structural health monitoring by detecting anomalies without the need for labelled data, making them ideal for environments where labelled data are costly or impractical. However, these methods may struggle with harmless anomalies from actual damage, adapting to changing conditions, and providing clear explanations for the detected anomalies.

Deep Learning in SHM

Deep learning significantly improves structural health monitoring by automating damage detection, localisation, and severity assessment through complex feature extraction from complex datasets like vibration signals, images, and acoustic data. Techniques such as convolutional neural networks excel at processing large-scale data, enabling the early identification of structural damage that might go unnoticed with traditional methods. These models offer a higher accuracy rate in detecting anomalies, reducing the reliance on manual inspections, and enhancing predictive maintenance by analysing long-term structural behaviours [16]. However, there are drawbacks of deep learning in SHM applications. One significant challenge is the need for large amounts of high-quality data to train the models effectively, which can be difficult to obtain for unique or complex structures. Additionally, deep learning models are usually sensitive to changes in environmental conditions, which lead to inaccurate data if the model is not properly tuned or if it encounters data outside its training set. High computational requirements for processing data can also limit the practical application of these models, particularly in resource-constrained settings [17]. Deep learning enhances structural health monitoring by automating damage detection and assessment through advanced feature extraction from complex datasets, leading to higher accuracy and reduced reliance on manual inspections. However, the effectiveness of these models depends on large and high-quality datasets, they are sensitive to environmental changes, and they require significant computational resources, which greatly limit their application in some settings of structural health monitoring.

Conclusion

In conclusion, supervised, unsupervised, and deep learning methods have significantly advanced structural health monitoring by improving damage detection, localisation, and assessment in infrastructures. Supervised learning, using algorithms like recurrent neural networks and convolutional neural networks, requires extensive labelled datasets and can struggle with new or noisy data, but it excels at accurately diagnosing specific types of structural damage. Unsupervised learning, which does not require labelled data, is effective for anomaly detection but may produce false alarms and requires professional analysis to understand the anomalies it detects. Deep learning methods, including convolutional neural networks and recurrent neural networks, enhance structural health monitoring by automating the analysis of complex datasets, though they require large amounts of high-quality data and are sensitive to environmental changes. Each learning method offers unique benefits and challenges, highlighting the importance of choosing the appropriate approach based on the specific needs of the structural health monitoring application. Overall, structural health monitoring should prioritise hybrid approaches that integrate supervised, unsupervised, and deep learning methods to enhance accuracy and adaptability, particularly under varying environmental conditions. Focusing on transfer learning and real-time adaptation could further reduce data dependency and improve system reliability.

Funding

This research received no external funding.

Conflicts of Interest

The author declares no competing financial interests.

References

  1. Singh, T.; Sehgal, S.; Prakash, C.; Dixit, S. Real-Time Structural Health Monitoring and Damage Identification Using Frequency Response Functions along with Finite Element Model Updating Technique. Sensors 2022, 22, 4546. [Google Scholar] [CrossRef]
  2. Zakaria, M.; Karaaslan, E.; Catbas, F.N. Advanced bridge visual inspection using real-time machine learning in edge devices. Adv. Bridge Eng. 2022, 3, 27. [Google Scholar] [CrossRef]
  3. Sakiyama, F.I.H.; Lehmann, F.; Garrecht, H. A Novel Runtime Algorithm for the Real-Time Analysis and Detection of Unexpected Changes in a Real-Size SHM Network with Quasi-Distributed FBG Sensors. Sensors 2021, 21, 2871. [Google Scholar] [CrossRef] [PubMed]
  4. Zhang, G.Q.; Wang, B.; Li, J.; Xu, Y.L. The application of deep learning in bridge health monitoring: A literature review. Adv. Bridge Eng. 2022, 3, 22. [Google Scholar] [CrossRef]
  5. Luckey, D.; Fritz, H.; Legatiuk, D.; Peralta Abadía, J.J.; Walther, C.; Smarsly, K. Explainable Artificial Intelligence to Advance Structural Health Monitoring. In Structural Health Monitoring Based on Data Science Techniques. Structural Integrity; Cury, A., Ribeiro, D., Ubertini, F., Todd, M.D., Eds.; Springer: Cham, Switzerland, 2022; Volume 21. [Google Scholar] [CrossRef]
  6. Abdulrazzaq, M.M.; Ramaha, N.T.A.; Hameed, A.A.; Salman, M.; Yon, D.K.; Fitriyani, N.L.; Syafrudin, M.; Lee, S.W. Consequential Advancements of Self-Supervised Learning (SSL) in Deep Learning Contexts. Mathematics 2024, 12, 758. [Google Scholar] [CrossRef]
  7. Kiran, B.R.; Thomas, D.M.; Parakkal, R. An Overview of Deep Learning Based Methods for Unsupervised and Semi-Supervised Anomaly Detection in Videos. J. Imaging 2018, 4, 36. [Google Scholar] [CrossRef]
  8. Taye, M.M. Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Future Directions. Computers 2023, 12, 91. [Google Scholar] [CrossRef]
  9. Martin, S. What’s the Difference between a CNN and an RNN? NVIDIA Blog. 2018. Available online: https://blogs.nvidia.com/blog/whats-the-difference-between-a-cnn-and-an-rnn/ (accessed on 3 September 2024).
  10. Klingler, N. Convolutional Neural Networks (CNNs): A 2024 Deep Dive. Viso, 2024. Available online: https://viso.ai/deep-learning/convolutional-neural-networks/ (accessed on 3 September 2024).
  11. Jia, J.; Li, Y. Deep Learning for Structural Health Monitoring: Data, Algorithms, Applications, Challenges, and Trends. Sensors 2023, 23, 8824. [Google Scholar] [CrossRef]
  12. Altabey, W.A.; Noori, M. Artificial-Intelligence-Based Methods for Structural Health Monitoring. Appl. Sci. 2022, 12, 12726. [Google Scholar] [CrossRef]
  13. Boccagna, R.; Bottini, M.; Petracca, M.; Amelio, A.; Camata, G. Unsupervised Deep Learning for Structural Health Monitoring. Big Data Cogn. Comput. 2023, 7, 99. [Google Scholar] [CrossRef]
  14. Eltouny, K.; Gomaa, M.; Liang, X. Unsupervised Learning Methods for Data-Driven Vibration-Based Structural Health Monitoring: A Review. Sensors 2023, 23, 3290. [Google Scholar] [CrossRef]
  15. Souza, L.; Yano, M.O.; da Silva, S.; Figueiredo, E. A Comprehensive Study on Unsupervised Transfer Learning for Structural Health Monitoring of Bridges Using Joint Distribution Adaptation. Infrastructures 2024, 9, 131. [Google Scholar] [CrossRef]
  16. Zhang, Z.; Pan, H.; Wang, X.; Lin, Z. Deep Learning Empowered Structural Health Monitoring and Damage Diagnostics for Structures with Weldment via Decoding Ultrasonic Guided Wave. Sensors 2022, 22, 5390. [Google Scholar] [CrossRef] [PubMed]
  17. Toh, G.; Park, J. Review of Vibration-Based Structural Health Monitoring Using Deep Learning. Appl. Sci. 2020, 10, 1680. [Google Scholar] [CrossRef]