Transportation infrastructure system is a key component of smart and sustainable cities. To support the development of future smart cities, the efficiency and sustainability of transportation infrastructure systems must be improved. In this context, there is an increasing demand for efficient and effective monitoring and sensing of such infrastructures. This thesis proposes a novel framework to monitor and sense the transportation infrastructures using crowdsensing data in vehicles. With the participation of passengers and drivers, the framework has the potential to monitor a large number of transportation infrastructures with reduced costs in a timely manner. Under this framework, this thesis first develops methods to identify damage and mode shapes of bridges using the vibration data collected from vehicles crossing these bridges. Numerical analysis and laboratory experiments are conducted to verify the proposed methods. Second, a deep learning-based algorithm is developed to automatically identify cracks in pavements using in-vehicle camera. At the end of this thesis, the conclusions and recommendations for future research are presented.

The general contribution of this thesis is the development and investigation of a crowdsensing-based framework for infrastructure monitoring using vehicles. The detailed contributions can be summarized as below:​ 1) A methodology is proposed for bridge condition screening using a large number of vehicles. This method identifies the damage in the bridge by introducing Mel-frequency cepstral coefficients and comparing the distributions of the extracted features. The method is verified by numerical analysis and laboratory experiments with professional sensors and smartphones. 2) A methodology is developed to identify bridge mode shapes from moving vehicles. In this method, the problem is first converted into matrix completion problem through a mapping process. Then, a “soft-imputing” algorithm is implemented to fill the matrix for system identification. The numerical results demonstrate that the methodology can find mode shapes accurately with multiple mobile sensors moving at traffic speed. 3) A methodology is proposed for pavement crack detection using in-vehicle camera and deep learning algorithm. With the help of connectivity maps and generative adversarial networks, the proposed method can achieve better performance than traditional image processing methods, and also has the potential to scan the roads quickly with reduced costs and high efficiency.