Urban underground stormwater pipeline networks are complex systems that can be affected by such external factors as boundary soil information, storm and flooding histories and network factors such as pipe material integrities. Because the pipeline system is underground, accurately locating aging pipe locations has been historically challenging (almost as challenging as mineral prospecting) and can involve the use of nonintrusive monitoring techniques, which is time- and cost-consuming.
In this project, we propose to develop a spatially explicit network modeling framework and software package (DeepPipe) based on deep learning, a state-of-the-art artificial intelligence approach, for automated characterization and anomaly detection of NCDOT’s existing underground storm drainage pipe network. DeepPipe will focus on the prediction of pipe location, features, and service life using deep learning-based graph neural network techniques as pipe networks are fundamentally graphs. The DeepPipe system will be trained and validated using existing storm management system data (pipe material, pipe location, pipe quality, flow quantity), surrounding subgrade data, and storm water history, collected through various sampling and monitoring devices and data sources to provide comprehensive information that can be used to rectify the limitations of existing the limitations of existing drainage network management approaches. To enhance underground stormwater pipeline network management, robust spatially explicit deep learning algorithms and other machine learning techniques will be developed as a core component of DeepPipe to resolve the challenge facing the auto-recognition, extraction/migration and transfer of pipe network data. Web- and mobile app-based implementations will be provided to facilitate the use of the DeepPipe system within in-situ environments.
The DeepPipe system can be used by several NCDOT divisions and other government entities working on veracious aspects of urban flooding managements. The proposed deep learning-based model and software product will provide substantial support for the rapid and automated network data manipulation, which will bring significant benefits (e.g., cost and time savings) to NCDOT asset management. The results of this research will also assist NCDOT in establishing policies and decision-making pertaining to extreme climate adaptation strategies.
