Loading...
Deep learning-based Models for Distributed Damage Detection and Quantification in Concrete Using Sinusoidal Ultrasonic Response Signals
Ranjbar, Iman | 2022
385
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 55166 (09)
- University: Sharif University of Technolog
- Department: Civil Engineering
- Advisor(s): Toufigh, Vahab
- Abstract:
- In this thesis, supervised and unsupervised deep learning-based frameworks were proposed for distributed damage detection and quantification in concrete using sinusoidal ultrasonic response signals. Before the main study on ultrasonic-based concrete damage assessment, a preliminary study was performed on deep learning-based concrete compressive strength prediction. In this study, convolutional neural networks were utilized to predict the compressive strength of concrete through its mix proportions. The Genetic algorithm was employed to find the optimum number of filters in each convolutional layer of the convolutional neural networks. The proposed framework demonstrated high accuracy in modeling a benchmark dataset and outperformed previous studies on the same dataset.After the preliminary study, a comprehensive experimental study was designed and performed. This study aimed to obtain rich datasets for training and evaluation of the proposed deep learning based frameworks. This study provided 1920 ultrasonic test results from five different geopolymer concrete specimens.The proposed supervised framework used LSTM networks as the base predictive model. Two different LSTM architectures were proposed: (1) to predict the damage stage of concrete, and (2) to predict the concrete specimen's absorbed energy ratio. The proposed LSTM-based supervised models were able to identify the health condition of concrete specimens with high accuracy.An unsupervised framework was proposed for distributed damage detection in concrete. This framework used a deep auto-encoder to reconstruct the ultrasonic response signals of the specimen. This model was trained on the intact specimen's responses.The auto-encoder's reconstruction error, measured by three parameters, was considered an effective damage-sensitive feature and was utilized to collect an Isolation Forest's binary decision trees. This Isolation Forest was then utilized for anomaly (damage) detection in the specimen.The supervised models reached an accuracy of 98.0% in predicting the concrete's damage stage and an R2 of 0.977 in predicting the absorbed energy ratio. The unsupervised models showed damage detection accuracy of 95.4% for damaged specimens and 92.0% for intact specimens.
- Keywords:
- Damage Detection ; Damage Assessment ; Concrete ; Deep Learning ; Sinusoidal Ultrasonic Response Signals