A Current Challenge within the Industry
Construction processes are always very complex, and many variables must be considered. Regarding tunneling, the building process is even more challenging as the risks evolve all the time due to inherent uncertainties such as unforeseen variations of the geological conditions and unpredicted tunneling-induced ground movements. To overcome these challenges, an enormous amount of engineering data is recorded during tunneling. Tunnel boring machines (TBM) are manufactured to record thousands of sensor measurements every few seconds, while real-time instrumentation and monitoring systems measure ground and building movements within a large area using various sensing technologies. However, due to its large volume and inherent complexity, TBM operators and geotechnical engineers find it’s not easy to interpret such big data intuitively and efficiently using traditional data analysis techniques.
Collaborative Research Work: AI in Tunnel Monitoring
The advancement of artificial intelligence (AI) and machine learning (ML) techniques in the past few years have opened extensive opportunities for managing and interpreting ‘big data’. These technologies enable us to exploit hidden patterns within complex and high-dimensional data, develop data-driven prediction models, and move toward more automated processes.
Ongoing research on using machine learning to connect Tunnel Boring Machine (TBM) sensing to ground monitoring, and to develop a systematic framework for a data-driven autonomous tunneling system is conducted by the Soga Research Group based at the University of California, Berkeley, in collaboration with industry partners of Sixense Inc., Enzan Koubou Co. Ltd., and Shimizu Corp. The tunneling data sets were provided by the Washington State Department of Transportation and MRTJ which includes the ground deformation data of the Seattle State Route-99 tunnel project collected by Sixense between 2012-2017.
The outcome of the 1st stage research shows that TBM data can be used to characterize the geology along the tunnel alignment using both supervised and unsupervised learning algorithms. The research is approaching its second stage, which connects the underground (TBM) data with the aboveground (ground surface and structures monitoring) data, expecting more collaborative research work from the Sixense team. It is expected that the proposed system may improve the tunneling reliability, risk profile, and safety, and eventually reduce the cost of tunneling. Furthermore, it is envisioned that the proposed system will advance the development of autonomous TBM technology.