| Author: | Wong, Tin Wai Phoebe |
| Title: | Imaging and diagnosis of corrosion in reinforced concrete structures by using ground penetrating radar |
| Advisors: | Lai, Wai-Lok Wallace (LSGI) Poon, Chi-sun (CEE) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Reinforced concrete -- Corrosion Reinforced concrete construction Ground penetrating radar Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Land Surveying and Geo-Informatics |
| Pages: | xxviii, 283 pages : color illustrations |
| Language: | English |
| Abstract: | Steel-reinforced concrete has been widely applied for constructing various infrastructures because concrete is highly durable and economical. One of the major concerns with steel-reinforced concrete structures is ageing and accompanied deterioration due to the ingress of aggressive agents from the atmosphere, ground and ocean. Chloride intrusion in marine and offshore structures is a common and major problem when concrete structures are located close to the sea. This is because the penetration of free chloride ions corrodes the steel reinforcement and could lead to structural failure like cracks of various kinds in a long run. Engineers have been searching for effective non-destructive diagnosis methods for decision-making of concrete maintenance and repair without compromising the structural integrity and serviceability. The most common existing inspection methods like the measurement of corrosion potential (HCP) is an electrochemical semi-destructive and small-scale methods based on setting thresholds mapping the probability of corrosion. However, HCP suffices from its over-simplified one-dimensional contour map for diagnosis, and the simple thresholds (ASTM C876) is not ideal for correct diagnosis due to the variety of concrete properties like mixes, exposure conditions, surface-wetting conditions, etc. Another NDT method, ground penetrating radar (GPR) is a suitable technology because concrete properties are much less susceptible to GPR than HCP, in addition to its superior robustness and efficiency of data collection and imaging compared with HCP. However, GPR signals and images do not convey semantic or physical meanings of material properties for classification and then diagnostic purposes in the context of this thesis, i.e., corrosion. This thesis contributes to the understanding and definition of concrete corrosion problems on the basis of GPR-based classification and machine-learning approaches from three perspectives. The first is the understanding and definition of concrete corrosion problems. The specific factors for an accurate diagnosis of chloride-induced corrosion by GPR imaging are largely unknown. For instance, how the GPR data in 2D and 3D domains can be used to represent and classify corrosion indicators and phases for the sake of diagnosis. In fact, over the last decade, there has been a debate in the international research community on the effect of corrosion on the GPR reflection amplitude. Some researchers state that the GPR wave signal strength measured as reflected amplitude decreases as corrosion progresses, while some observed an opposite change because corrosion is not a well-defined terminology in these research. On this note, this thesis first attempts to define the underlying corrosion mechanism by classifying it into chloride intrusion and active corrosion in severely deteriorated concrete at different times of service life. The second contribution is the use of GPR for concrete assessment. In this thesis, a procedure for numerically simulating GPR signals as a basis for characterising the complex dielectric permittivity of concrete was developed and validated on real concrete specimens. An empirical method for harmonising the amplitudes acquired from different GPR systems was proposed and validated. This new and original data pre-processing step facilitates the application of machine learning on GPR data. Additionally, a more comprehensive ray-based approach was introduced to investigate the effects of chloride-induced corrosion on GPR signals. This approach makes use of the amplitude and frequency GPR attributes extracted from the direct wave and reflected wave, which is more extensive than existing methods. Third, a supervised classifier trained by both ray-based and imaged-based GPR attributes were developed. Image-based learning with only hyperbolas as training materials is a common approach. The hybrid approach present in this thesis is novel as it integrates the advantages of both sides. For example, a deformed hyperbolic reflection from the rebar is likely associated with severe deterioration, and the ray-based GPR attributes provide more detailed information that is influenced by the effect of chloride-induced corrosion on reflected GPR signals. Together with the well-defined corrosion phases in the laboratory and field experiments, the processed GPR data were labelled and utilised for training a classification model. Ultimately, GPR can become a stand-alone evaluation tool for corrosion assessment without other electrochemical methods. The implementation of the above three contributions is described as follows. An amplitude conversion method was proposed in this thesis to harmonise the data acquired by different GPR instrumentations into a single fingerprinting database. Then, the fingerprinting database was produced by using ground-truthed or laboratory-referenced data (5 field concrete structures and 8 laboratory concrete specimens, compiling more than 15,000 GPR scans for training). The contributing attributes were extracted from the rebar reflected waves, direct waves and whether a flat layer of reflection is observed at the associated traces of the rebar's hyperbolic reflection. This comprehensive study shows that the amplitude attributes change more prominently in different corrosion phases than the frequency attributes. The database was used for training a supervised classification model (logistic regression, LR) which served as a training set and was applied to classify the corrosion phases of five test sets with different conditions as blind tests. The LR model and the amplitude thresholding method were compared. The limitations of the latter method were identified, and the supervised LR model was found to be a better solution. The prediction results of the LR model have a high cross-validation score (>99%) and could reflect the actual corrosion phase of the unseen test concrete specimens in general with a low false alarm rate (<4%), and high sensitivity and specificity (>96%). The LR classifier performs similarly to the baseline performance based on human-level performance (a level comparable to or indistinguishable from a human expert). Continuous expansion of the database and refinement of the decision boundaries derived by supervised learning are suggested in the future. In summary, this work integrates knowledge from three domains: infrastructure in civil engineering, ground penetrating radar and machine learning, to establish a robust and reliable workflow for diagnosing chloride-induced corrosion in concrete with a fully non-destructive approach. This turns GPR into an evaluation tool and paves the path for transferring the same techniques to cope with other concrete-related problems with GPR. It also opens a gateway which demonstrates how a concrete structure can be diagnosed with a completely non-destructive method. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12516