Structural Damage Diagnosis
Integrates modal-based and machine learning (ML) based methods to develop a more damage-sensitive, noise robust, and generalised damage diagnosis method
Research Scopes
Integration of modal-based and machine learning (ML) methods for structural damage diagnosis
Unsupervised ML method
Unsupervised methods are data-driven techniques that aim to identify and characterize structural damage without relying on labeled or pre-classified data. These methods extract patterns or anomalies from the structural response data and use them to detect and assess the presence of damage. Our team has implemented k-means and hierarchical clustering for damage diagnosis and the developed approach can go up to damage identification.
Hybrid Modal-ML Method
As structural damages are rare, a conventional supervised-based method may face the cold-start issue, i.e., having limited training examples at earlier stages. Our team has developed an integration strategy to combine both modal-based and machine learning methods to overcome this issue, and the proposed approach is able to locate damages on both plate and beam-like structures.
Hybrid Digital Twin
Our team aims to integrate a physics-based model with ML methods to improve the extrapolation capacity of the current ML method. It is an ongoing research and this research aims to enhance the interpretability of the data-driven or ML model.
Unsupervised ML Method
The hierarchical clustering method allows automatic grouping of the upstream data based on their similarities or dissimilarities. A real-time-based program was developed using LabVIEW software. By using the upstream data (Frequency Response Functions or FRFs) as the input features and ISMA as the modal analysis method, the developed program is damage-sensitive, noise robust, and can achieve up to locating and quantifying the damage.
Hybrid Modal-ML Method
This approach integrates modal-based and machine learning (ML) based methods to develop a more damage-sensitive, noise robust, and generalised damage diagnosis method. Both unsupervised and supervised ML methods were used, and a mode shape assessment method was strategically integrated as a physics-based method to allow damage identification even when there is no trained supervised model yet at earlier stages. The real-time-based program was developed using LabVIEW software, and ISMA was implemented as the modal analysis method.
Motivation of This Research
Cold-Start Problem faced by ML methods
Damage events are rare, which causes the cold-start problem for machine learning methods that usually requires several labelled examples of healthy and damaged conditions for training.
The need for a noise robust modal analysis method for damage diagnosis
Conventional modal analysis requires shutdown conditions, which is not practical for modal-based damage diagnosis methods.
Damage sensitive features required for unsupervised methods
Modal parameters, i.e., natural frequencies are commonly used as the damage-sensitive feature for unsupervised methods. However, as modal parameters are not the upstream data (FRFs), they could incur extraction-errors that affect their sensitivity towards damage.
Main Features of the Proposed Approach
Damage-sensitive, noise robust, and solves the cold-start problem
ISMA
Makes the program to be noise robust
PCA-FRF
Used as a damage-sensitive feature for 1st level unsupervised damage presence detection
MSA
Used as a single damage classifier to solve the cold-start issue and auto labels the damage samples
Hybrid ML
Combines unsupervised and supervised methods to diagnose unseen single and multiple damages
Tested on Plate-like and Beam-like Structures
Generalised enough to locate unseen single and multiple damages for both plate and beam-like structures
Research Outputs
- Siow, P. Y., Ong, Z. C., Khoo, S. Y., & Lim, K. S. (2023). Hybrid modal-machine learning damage identification approach for beam-like structures. Journal of Vibration and Control. https://doi.org/10.1177/10775463231209008
- Siow, P. Y., Ong, Z. C., Khoo, S. Y., Lim, K. S., & Chew, B. T. (2023). Hybrid machine learning with mode shape assessment for damage identification of plates. Smart Structures and Systems, 31(5), 485-500. https://doi.org/10.12989/sss.2023.31.5.485
- Siow, P. Y., Ong, Z. C., Khoo, S. Y., & Lim, K. S. (2023). Noise robustness of an operational modal-based structural damage-detection scheme using Impact-Synchronous Modal Analysis. Journal of Zhejiang University-Science A, 24(9), 782–800. https://doi.org/10.1631/jzus.A2200620
- Gordan, M., Siow, P. Y., Deifalla, A. F., Chao, O. Z., Ismail, Z., & Yee, K. S. (2022). Implementation of a Secure Storage Using Blockchain for PCA-FRF Sensor Data of Plate-Like Structures. Ieee Access, 10, 84837-84852. https://doi.org/10.1109/access.2022.3197776
- Siow, P. Y., Ong, Z. C., Khoo, S. Y., & Lim, K. S. (2021). Damage Sensitive PCA-FRF Feature in Unsupervised Machine Learning for Damage Detection of Plate-Like Structures. International Journal of Structural Stability and Dynamics, 21(2), 29. https://doi.org/10.1142/s0219455421500280
- Chen, S., Ong, Z. C., Lam, W. H., Lim, K.-S., & Lai, K. W. (2020). Operational Damage Identification Scheme Utilizing De-Noised Frequency Response Functions and Artificial Neural Network. Journal of Nondestructive Evaluation, 39(3), 66. https://doi.org/10.1007/s10921-020-00709-x
- Chen, S. L., Ong, Z. C., Lam, W. H., Lim, K. S., & Lai, K. W. (2020). Unsupervised Damage Identification Scheme Using PCA-Reduced Frequency Response Function and Waveform Chain Code Analysis. International Journal of Structural Stability and Dynamics, 20(8), 26. https://doi.org/10.1142/s0219455420500911
