|Internet of things and deep learning for construction-induced vibration monitoring
|Zhu, Songye (CEE)
|FCE Awards for Outstanding PhD Theses (2022/23)
|Vibration -- Measurement
Hong Kong Polytechnic University -- Dissertations
|Department of Civil and Environmental Engineering
|xxvii, 274 pages : color illustrations
|Construction activities frequently generate excessive vibrations that adversely affect nearby structures, facilities, and human beings. Consequently, vibration monitoring is typically adopted to assess the impact of vibrations during construction. Common monitoring solutions include fixed monitoring stations and portable seismographs, which have limitations, such as high costs, specialized operation, difficult maintenance, and limited functions. Moreover, huge amounts of monitoring data have not been efficiently used in other studies due to missing relevant information, such as corresponding construction activities and anomalous markings. To solve these problems, this thesis is divided into two parts. The first part focuses on developing an innovative construction-induced vibration monitoring system by integrating the Internet of Things (IoT) and a cloud platform. The second part proposes advanced deep learning (DL)-based approaches for construction activity classification and data anomaly detection.
The development of IoT-based monitoring systems is further divided into two branches. First, a wireless IoT sensing system was developed by combining an open-source single-board microcontroller unit (MCU), a microelectromechanical systems (MEMS) accelerometer, and a long-term evolution (LTE) network. Real-time vibration impact assessment was achieved through the sufficient computing capability of the MCU. A cloud platform was built for data storage, data visualization, and autonomous alarming. The developed system was verified through laboratory shaking table tests and then applied to a real construction site with ongoing sheet piling work.
Second, a smartphone is selected as an IoT monitoring device. As a device equipped with a variety of sensors, powerful processors for computing, and multiple network infrastructures, a smartphone is an ideal alternative in an IoT environment. By leveraging the advantages of smartphones, the first iOS application (app) was designed and developed for construction-induced vibration monitoring. The developed app features diverse functions, such as data measurement, specialized index analysis, convenient data sharing, and other functions that cannot be provided by traditional monitoring devices. Comprehensive calibration tests were performed in a laboratory to verify the feasibility of using iPhone models. In these tests, the sensitivity of the iPhone is compared among different model types (including four iPhones and two Android smartphones) and a high-fidelity accelerometer. Thereafter, a series of field measurements were conducted to validate the performance of the smartphone app on real construction sites with different construction activities. A cloud platform is integrated into the system to provide scalable data storage, convenient data sharing, and technical support for functions, such as user login and identification authentication.
In addition, a smartphone has the unique ability to measure noise and geographic information along with vibrations. Using a smartphone to measure the interior vibration and noise of trains has been explored, providing valuable experiences for the application of IoT monitoring technology in rail transit systems and enabling the realization of citizen sensing in the future.
The second part of the thesis comprises a supervised method and a semi-supervised method for construction activity classification. Given the problems of data imbalance and scarcity, a state-of-the-art time-series data augmentation method (i.e., RandAugment) was applied to synthesize additional training data with similar intrinsic properties as the original data. The accurate classification of the construction activity was realized by training a convolutional neural network on the augmented data in conjunction with the original data.
The supervised learning method requires training data coupled with labels. However, labels are generally missing for a vast amount of monitoring data, while manual labeling is usually laborious, error-prone, and impractical. Two semi-supervised learning methods, called Pseudo-Label and Ladder network, were applied to overcome the difficulty. Both methods were trained on a real-world database that consists of vibration data generated by various construction activities. The results indicate that semi-supervised methods can learn from unlabeled data, and the Ladder network is markedly more effective, given a highly limited number of labeled data.
The last part of this thesis proposes an unsupervised data anomaly detection method for online construction-induced vibration monitoring systems, in which labels of anomalous data are not required. The proposed method identifies anomalies through the difference between the original data and the data reconstructed by DL algorithms. An adaptive threshold with an adjustment coefficient was applied to set a reliable anomaly threshold. Furthermore, a novel cloud computing technology (i.e., distributed training) was utilized to accelerate the training process and realize real-time data anomaly detection in an operating monitoring system. Experiments were conducted on a public cloud platform composed of various numbers of graphics processing units, and the performance of different distributed frameworks was compared.
|All rights reserved
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: