Full length articleEstimation of the degree of hydration of concrete through automated machine learning based microstructure analysis – A study on effect of image magnification
Introduction
It is estimated that the concrete is consumed at a rate of 11 billion metric tons per year, and it is the most widely used construction material for buildings, bridges, roads, and dams because of the exceptional properties such as workability, durability, and strength [1]. The physical and mechanical properties of the concrete are governed by the microstructure of the concrete [2], and they can be modified by making suitable changes to the microstructure of the concrete. The microstructure of concrete is a complex system with a high degree of heterogeneity, composed of solid phases, pores, and water. At the simplest heterogeneity level, concrete consists of aggregates, hydrated cement paste, anhydrous cement and pores [3]. The quantitative microstructure characterization of concrete is useful to determine the relationship to macrostructure properties [4]. Understanding the microstructure and hydration of the cement will ensure the strength and durability of concrete, helps in recognizing and mitigating the stresses to prevent cracking, and assists in determining the curing conditions and construction practices. Therefore, it is essential to understand the relationship between microstructure and the properties of concrete because the inhomogeneity in the microstructure of the concrete can lead to severe effects on strength and other related physical and mechanical properties. Among these properties, the strength and durability of the concrete can be highly influenced by the cement hydration rate. The analysis and understanding of the microstructure of concrete, more specifically the degree of hydration, can help researchers and practitioners to improve the performance of the concrete, modify the bond strength and make durable concrete repairs. In addition, it can also play a vital role in determining the cause of failure or damage to infrastructure during concrete forensics studies.
The microstructure analysis of concrete has a wide range of applications; some of them include concrete forensics, bridge condition monitoring, etc. There exist various methods for concrete microstructure investigation such as light optical microscopy with associated digital image analysis, scanning electron microscopy (SEM), X-ray computed tomography with image processing [5]. With the advancement in the image processing techniques, the scanning electron microscopy (SEM) image analysis has been widely adopted for the microstructure characterization. This SEM image analysis technique usually involves specimen preparation, SEM image collection, and image segmentation. For an accurate quantitative estimation of properties such as the degree of hydration, porosity, or water-cement ratio, the selected image segmentation algorithm should be precise and reproducible [6]. Various researchers [7], [8], [9] presented different SEM image segmentation methods using edge detection, particle swam optimization, histogram intensity method, and etc. Some of the limitations in using these techniques are that these algorithms are not adaptable, non-reproducible, semi-automated, and most importantly rely on the specific magnification image. The effect of magnification on microstructure analysis is one of the significant challenges because researchers or practitioners capture SEM images at different magnification to better understand the sample, but the existing methods are limited on processing images with certain preset magnification.
Therefore, this study presents a machine learning based SEM image segmentation methodology for the concrete microstructure analysis, more specifically it focuses on the estimation of the degree of cement hydration through the proposed approach. The proposed machine learning based SEM image segmentation method adopts pixel-based image segmentation which generates highly accurate results for microscopy images. This method can produce consistent results provided if the model is trained on same features and precise training samples. Some most commonly used machine learning classifier algorithms were tested and evaluated in the experiment to identify the best fit for the proposed machine learning based concrete SEM image segmentation. This study also further investigates the effect of magnification on classifier training, and model accuracy for microstructure characterization and degree of hydration measurement considering two scenarios. In the first scenario, a classifier is trained using images of specific magnification. Whereas in the second scenario, a common classifier is trained using images of all the magnification. Understanding the effect of magnification on classifier performance helps researchers and practitioners to process the images of different magnification using single classifier which reduces the intensive manual effort and long computation time.
The rest of the paper is structured as follows. First, the literature review section reviews the literature on the significance of microstructure analysis and degree of hydration measurement, existing methods for microstructure analysis and machine learning based segmentation. Then, the methodology section illustrates the proposed machine learning based microstructure analysis using SEM images. Followed by the experimental validation section, in which it evaluates the performance of the proposed methodology. In the end, this study is concluded with a summary of all results and findings, limitations to this research, and recommendations for future research.
Section snippets
Current methods for concrete SEM image analysis
SEM image analysis and X-ray Diffraction are the most commonly used techniques to quantify microstructure phases of the concrete. Due to the presence of amorphous phases in the concrete, SEM image analysis technique is more accessible to apply compared to the X-ray diffraction method [10]. SEM images of concrete are segmented into various components such as aggregates, hydrated cement, anhydrous cement, and pores using image-processing techniques. The segmented images allow measurement of areas
Automated machine learning based microstructure analysis
The framework of the proposed machine learning based microstructure analysis is shown in Fig. 1. The first step of the methodology is the preparation of samples for SEM imaging. Multiple SEM images are then collected from each specimen. The collected SEM images are used as a training and testing data for machine learning classifier algorithms. The trained classifier is used for the segmentation of images. The segmented images are used for microstructure characterization and degree of hydration
Experiment setup
The SEM images of four specimens were used in this study to achieve the proposed objective. The SEM images of the specimens were obtained from a study of determining the crack-repair efficiency of self-healing concrete through microencapsulated calcium nitrate (microcapsule) [34] Table 2 provides the details of cement concentration, agitation rate, and average microcapsule size used to prepare the specimens. For the specimen preparation, Type I cement with no supplementary cementitious
Conclusion
Researchers or practitioners capture multiple SEM images of concrete at different magnifications to better understand the sample. The SEM imaging at different magnification involves high cost and time. The existing methods for determining the degree of hydration requires SEM images of specific magnification, where the other images are not utilized. Moreover, the existing methods for microstructure characterization and degree of hydration measurement are highly influenced by image magnification,
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (34)
Backscattered electron imaging of cementitious microstructures: understanding and quantification
Cement Concrete Compos.
(2004)- et al.
Pore segmentation of cement-based materials from backscattered electron images
Cement Concrete Compos.
(2006) - et al.
Determining the water–cement ratio, cement content, water content and degree of hydration of hardened cement paste: Method development and validation on paste samples
Cement Concrete Res.
(2009) - et al.
Binary segmentation of aggregate in SEM image analysis of concrete
Cement Concrete Res.
(2001) - et al.
Determination of the proportion of anhydrous cement using SEM image analysis
Constr. Build. Mater.
(2016) - P.K. Mehta, P.J. Monteiro, Concrete Microstructure, Properties and Materials,...
- D.M. Roy, G. Idorn, Concrete microstructure,...
- et al.
A Practical Guide to Microstructural Analysis of Cementitious Materials
(2016) Microstructure of Concrete, High Performance Concrete Technology and Applications
(2016)- et al.
An effective image segmentation technique for the SEM image
Segmentation-assisted edge extraction algorithms for SEM images
SEM-backscattered electron imaging and image processing for evaluation of unhydrated cement volume fraction in slag blended Portland cement pastes
J. Wuhan Univ. Technol.-Mater. Sci. Ed.
A comparative analysis of various image segmentation techniques
Automated color model–based concrete detection in construction-site images by using machine learning
ASCE J. Comput. Civ. Eng.
Automated crack detection on concrete bridges
IEEE Trans. Automat. Sci. Eng.
Image processing-based recognition of wall defects using machine learning approaches and steerable filters
Comput. Intel. Neurosci.
Cited by (36)
Image enhancement and microstructure characterization of energy dispersive X-ray spectroscopy images of blended cement pastes
2024, Expert Systems with ApplicationsBenchmarking AutoML solutions for concrete strength prediction: Reliability, uncertainty, and dilemma
2024, Construction and Building MaterialsDeep learning based automated fracture identification in material characterization experiments
2024, Advanced Engineering InformaticsIdentification of experienced temperature in mortar and concrete using microstructural image and deep learning
2023, Construction and Building Materials