Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 56662 (09)
- University: Sharif University of Technology
- Department: Civil Engineering
- Advisor(s): Toufigh, Vahab
- Abstract:
- Concrete, a versatile construction material, combines cement, aggregates, and water to form a composite known for its durability and adaptability in various architectural and engineering applications. Its widespread use stems from its ability to efficiently bear heavy loads while maintaining longevity. However, producing concrete's key ingredient, cement, contributes significantly to environmental harm by emitting substantial carbon dioxide during its manufacturing process. This carbon footprint underscores the urgent need for sustainable alternatives and innovative practices in the construction industry to mitigate the detrimental effects on our planet. On the other hand, understanding the impact of fire on concrete structures is crucial for both safety and economic considerations, as it directly influences structural integrity during emergencies, potentially preventing fatalities. Additionally, this knowledge informs effective design and materials choices, minimizing post-fire repair costs and enhancing the overall resilience of built environments. During the experimental phase, a total of 972 cubic specimens, each with dimensions of 10 centimeters, were meticulously fabricated employing activating solutions of sodium hydroxide and sodium silicate. To this end, a meticulously manufactured set of 81 distinct mixed designs was formulated, encompassing four pivotal variables: 1) the activator-to-binder ratio (0.35, 0.40, and 0.45), 2) the water-to-binder ratio (0.35, 0.40, and 0.45), 3) the molarity of sodium hydroxide (8, 11, and 14), and 4) the sodium silicate-to-sodium hydroxide ratio (1.5, 2, and 2.5). Each mixed design included two representative specimens, and the mean of their results was considered the output. These prepared specimens were subsequently exposed to six target temperatures, including 25, 100, 250, 500, 700, and 900 degrees Celsius. Upon completion of the thermal exposure, a comprehensive observation of the specimens' visual attributes, encompassing alterations in color and surface crack formation, was conducted. Moreover, a quantification of the mass reduction was performed. After this preliminary investigation, a three-step testing procedure was conceived for the heated specimens in the electric furnace. This ensemble of tests encompassed non-destructive, destructive, and microstructural tests and analyses. The non-destructive assessment was implemented in two primary constituents: the ultrasonic pulse velocity (UPV) technique and the ultrasonic signal extraction method. On the other hand, the destructive assessment was conducted by harnessing a universal testing machine to acquire stress-strain curves, elasticity modulus, and strain energy. Eventually, microstructural analyses were undertaken using scanning electron microscopy (SEM), energy dispersive X-ray spectroscopy (EDS), X-ray diffraction (XRD), and thermogravimetric analysis (TGA). In the analytical phase, the objective was to employ new methodologies for effectively training networks and optimal utilization of the data generated in this study. To this end, data were initially collected from various technical literature references in the field of construction materials. Three distinct datasets were examined using several machine learning algorithms, including regression, genetic modeling, artificial neural networks, and support vector machines. Considering the characteristics of each algorithm, decisions were made on the extracted data from the present research study based on the algorithms' performance. Three datasets were collected from the experimental phase of the project: 1) data from the universal testing (compressive strength, elastic modulus, energy absorption) and ultrasonic pulse velocity tests, 2) data from compressive strength tests of specimens, and 3) digital signal data from ultrasonic signal tests. For the first dataset, regression was employed, genetic modeling for the second dataset, and artificial neural networks and support vector machines for the classification of the third dataset. Utilizing the first network, the results of non-destructive and destructive tests were correlated. The second network facilitated predicting the compressive strength of heat-treated slag-based concrete specimens. Leveraging the third dataset, prediction of the level of damage imposed on the specimen was achieved through non-destructive testing. The experimental-phase results revealed that the presence of air voids led to the development of cracks at elevated thermal levels beyond 500 degrees Celsius, reducing ultrasonic pulse velocity. Meanwhile, the elasticity modulus of the specimens exhibited a decreasing trend beyond 250 degrees Celsius, with a noticeable alteration primarily observed at 100 degrees Celsius compared to ambient conditions. SEM images depicted a denser microstructure and increased hardness for specimens subjected to thermal damage at 100 and 250 degrees Celsius. In line with this, the compressive strength of the specimens generally exhibited enhancements at the specified thermal levels compared to similar specimens under ambient conditions. A reduction in compressive strength occurred post-250 degrees Celsius, continuing up to 900 degrees Celsius. Depending on the mixed design, considerable strength deterioration was predominantly observed at 700 and 900 degrees Celsius. The most influential parameters in enhancing the thermal performance of slag-based concrete specimens with alkali activators, compared to ordinary Portland cement concrete, were found to be the maximum activator-to-slag ratio, molarity ratio, sodium silicate-to-sodium hydroxide ratio, and minimum water-to-slag ratio. The findings of the analytical phase demonstrated that, for the purpose of establishing empirical relationships, the utilization of regression and genetic modeling, despite the high precision exhibited by support vector regression networks, is preferable. Conversely, for classification purposes and considering the size of datasets within the realm of this project, artificial neural networks and support vector machines emerged as notably effective algorithms. Moreover, the correlation between the outputs of destructive tests and a relatively simple nondestructive test such as ultrasonic pulse velocity led to the introduction of some practical regression models with acceptable accuracies. Consequently, employing these models without the need for destructive tests makes it feasible to predict the decreased mechanical properties of fire-exposed specimens. Additionally, the presented genetic modeling accurately predicted slag-based concrete's thermal and fire-induced resilience, employing the concrete mixed design as input parameters. Furthermore, by leveraging artificial neural networks and support vector machine-based models for the digital signals received from damaged specimens, the damage level or the degree of thermal exposure to which a specimen has been subjected can be predicted with a fair accuracy of over 95%
- Keywords:
- Fire ; Elevated Temperature ; Geopolymer Concrete ; Slags ; Nondestructive Test ; Machine Learning ; Destructive Testing
Digital Object List
-
محتواي کتاب
- view
Bookmark