- Title
- Repeatability of self-healing in ECC with various mineral admixtures
- Creator
- Chen, Guangwei
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2021
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Concrete structures are susceptible to cracking that is primarily responsible for the reduction of the strength and stiffness of the concrete structure. This has substantial negative influences on the durability and sustainability of concrete, as well as significantly reducing the service life of the whole concrete infrastructure. For prolonging the service life of concrete infrastructure affected by cracks, self-healing of cementitious materials has attracted more and more attention. Engineered Cementitious Composite (ECC) has an intrinsic ability to control the width of cracks and to promote self-healing, which makes ECC the most promising self-healing material for improving the durability and serviceability of concrete structures. The aim of this research is to evaluate the self-healing behaviour of ECC when incorporated with different minerals, focusing on self-healing capability, repeatability and modelling prediction. In addition, it is expensive and time-consuming to quantify self-healing capability by conducting experiments, and difficult to mathematically predict self-healing based on available data. Therefore, an accurate and reliable self-healing prediction model will be designed to reduce time and costs for enhancing the durability design of ECC. For this purpose, ECC mixtures were created in which Fly Ash (FA) was partially replaced by 5%, 10% and 15% of Hydrated Lime Powder (LP) or Silica Fume (SF), and an ECC mixture without LP or SF was used as a control. The samples were precracked using a newly developed splitting tensile test apparatus at the age of 28 days. After that, the specimens were exposed to curing conditions with 10 Wet-Dry (W/D) cycles for self- healing of cracks. To study the self-healing capability and repeatability, the load was re-applied to the specimens after each 10 W/D cycles. A Rapid Chloride Permeability Test (RCPT) and splitting tensile test were utilized to assess the self-healing repeatability of ECC in terms of chloride ion permeability and mechanical properties, respectively. The recovery rate of crack width associated with different mixtures was supported and analysed by digital microscope observations. The precipitations formed at the surface of cracks were detected by Scanning Electron Microscope (SEM) equipped with an Energy Dispersive Spectroscopy (EDS), and then further examined by X-ray Diffraction Analysis (XRD). Experimental results show that all mixtures exhibited self-healing with slight differences. The microstructure was also assessed using SEM–EDS and XRD analysis. The microstructural analysis of healed cracks in LP incorporated ECC mixtures showed the presence of calcite, portlandite and Calcium Silicate Hydrates (C-S-H) gels as well as monocarboaluminate, which confirmed a possible reaction between FA and LP. Digital microscope analysis of crack width recovery showed that the addition of 5% SF is only beneficial for improving the recovery of crack width that is around 20 µm. The addition of 15% LP significantly improves the crack width recovery ability and repeatability compared to the reference sample. When SF and LP are added at the same time, the synergistic effect of 5% SF and 10% LP shows the best crack width recovery results. RCPT results show that all mixtures exhibited repeated self-healing with slight differences. The addition of 10% SF shows two times significant self-healing ability under three times repetitive loading; however, 15% SF is not conducive to repeated self-healing under multiple loading. The addition of 10% LP shows the best recovery rate. When SF and LP were simultaneously added to ECC mixtures, the RCPT results of FA55-SF10-LP5 and FA55-SF5-LP10 are all below the low level before and after three times of preloading and self-healing, in accordance with ASTM C1202. For the case where the load is applied only once, the exposure time of the W/D environment affects the self-healing capability of splitting tensile strength. The addition of 5% or 10% of SF to FA70 can increase the splitting tensile strength recovery rate of ECC samples within 30 W/D cycles. The addition of 15% SF to FA70 is beneficial for short term mechanical recovery but is not conducive to the recovery of mechanical properties in the long term. The addition of 10% LP is most conducive to improvement in the recovery of mechanical properties, especially for samples undergoing 60 W/D cycles. When SF and LP were simultaneously added to ECC mixtures, the addition of SF and LP at the same time shows a better recovery effect than adding only one mineral alone. FA55-SF5-LP10 shows the best recovery rate after 60 W/D cycles. In the case of repeated load application, FA55-SF15 shows the highest healing efficiency in the first round of self-healing, FA60-LP10 shows the highest splitting tensile strength recovery rate after three loads, and the recovery rate can still reach 66.43%. When SF and LP are simultaneously added, after three rounds of self-healing the splitting strength recovery rate of FA55-SF5-LP10 is significantly higher than other mixes, reaching 81.56%. The SEM and XRD analysis of healed cracks in different mineral-incorporated ECC mixtures showed that a mixture of CaCO3 and C-S-H are found as main self-healing products, however, the dominant healing product depends on the type and the proportion of minerals in ECC. The recovery in the mechanical and durability performance of the mixtures due to self-healing proposed in this research is anticipated to positively affect life cycle costs and lead to increased civil infrastructure sustainability. In order to develop a robust prediction tool for modelling the self-healing capability of ECC, a comparative study was conducted to evaluate the efficacy of machine learning models in predicting the self-healing ability of ECC using empirical data from the experimental study. In this study, four individual models, including Linear Regression (LR), Support Vector Regression (SVR), Back-propagation Neural Network (BPNN), and Classification and Regression Tree (CRAT), and three ensemble models namely bagging, AdaBoost and stacking, are adopted to develop 13 prediction models in total. The comparative analysis showed all 13 models came with expected accuracy and predictability. The BPNN model was prominent in the individual models in terms of forecast error, Root Mean Square Error (RMSE) and accuracy, R2. Stacking model was superior to all other individual or ensemble models on the basis of all three performance measures. The efficiency analysis of different machine learning methods for self-healing prediction in the comparative study provides a reference for choosing machine learning methods to predict the repeatability of crack self-healing in ECC. Therefore, the BPNN model is further optimized by the Evolutionary Algorithm (EA) in structured tree and list to construct two predictive tools to model the self-healing repeatability of ECC for improving the prediction performance. The proposed EA-based BPNN models overcame the drawback of BPNN with slow convergence and getting trapped in local minima. Especially, the EA-based BPNN in structured tree model ensures genetic diversity and keep fit solutions guaranteeing quality children of following generations, and is more space efficient leading to quick searching and convergence. Computational results reveal that the EA-based BPNN in structured tree model is superior to EA-based BPNN in structured list model which improves the performance of BPNN on all three statistical measurements on all three datasets including training, validation and testing.
- Subject
- ECC; self-healing; repeatability; mineral; machine learning; evolutionary algorithm
- Identifier
- http://hdl.handle.net/1959.13/1501381
- Identifier
- uon:55133
- Rights
- Copyright 2021 Guangwei Chen
- Language
- eng
- Full Text
- Hits: 92
- Visitors: 104
- Downloads: 24
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | ATTACHMENT01 | Thesis | 10 MB | Adobe Acrobat PDF | View Details Download | ||
| View Details Download | ATTACHMENT02 | Abstract | 259 KB | Adobe Acrobat PDF | View Details Download |