Loading...
Damage Detection and Health Monitoring of Geopolymer Concrete Using Ultrasound Waves and Machine Learning
Rahmati, Mohammad | 2021
235
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 54814 (09)
- University: Sharif University of Technology
- Department: Civil Engineering
- Advisor(s): Toufigh, Vahab
- Abstract:
- The behavior of construction materials at high temperatures has been an important aspect of researches in recent years. The first part of the current research aims to non-destructively monitor the formation and growth of damage at geopolymer concrete (GPC) after being exposed to high temperatures using linear and nonlinear ultrasonic techniques. Ultrasonic waveform and pulse velocity (UPV) tests were conducted on the specimens before and after exposure to the temperatures. Nonlinear wave signals were processed in phase-space domain for qualitative health monitoring of GPC. Furthermore, feature extraction was applied to phase-plane attractors using fractal dimension for quantitative assessment. As the prediction of concrete’s compressive strength after being exposed to high temperatures can be an important step in the damage assessment of buildings and fire safety applications, the second part of this research was assigned to investigate this problem. Two machine learning (ML) approaches, namely Support Vector Regression (SVR) and Artificial Neural Network (ANN), were used to predict the compressive strength of GPC containing the different compositions of fly ash and ground granulated blast furnace slag as binder materials at high temperatures range between 100⁰C to 1000⁰C. Different criteria such as coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE) were used for the evaluation of models’ performance. To have a better comparison, reliability curves such as the probability density function (PDF) and cumulative density function (CDF) of evaluation parameters were calculated. The results show that the phase-space analysis of ultrasound can successfully represent the severity of thermal damage at GPC. At the same time, the fractal dimension can be used as a supplementary damage index to UPV. The results also indicate that both SVR and ANN models have powerful capabilities in the estimation of GPC’s compressive strength after exposure to high temperatures. However, the testing phase of the models shows that the SVR (R² =0.7988, RMSE= 7.066, MAE= 4.3431) outperforms the ANN with two hidden layers (R² = 0.7868 ,RMSE= 8.2042, MAE= 5.1784) and ANN with one hidden layer (R² = 0.7702 , RMSE= 8.6869 , MAE= 5.6916)
- Keywords:
- Geopolymer Concrete ; High Temperature ; Ultrasonic Test ; Health Monitoring ; Artificial Neural Network ; Support Vector Regression ; Damage Identification
- محتواي کتاب
- view