Challenges in data-driven site characterization

ABSTRACT

Site characterisation is a cornerstone of geotechnical and rock engineering. “Data-driven site characterisation” refers to any site characterisation methodology that relies solely on measured data, both site-specific data collected for the current project and existing data of any type collected from past stages of the same project or past projects at the same site, neighbouring sites, or beyond. It is an open question what data-driven site characterisation (DDSC) can achieve and how useful are the outcomes for practice, but this “value of data” question is of major interest given the rapid pace of digital transformation in many industries. The scientific aspects of this question are presented as three challenges in this paper: (1) ugly data, (2) site recognition, and (3) stratification. The practical aspect that cannot be ignored is how to scale any solution to a realistic 3D setting in terms of size and complexity at reasonable cost. No deployment in practice is possible otherwise. At this point, the practicing community at large has yet to be convinced what data, big or small, could do to transform current practice. The authors believe that we need a more purposeful agenda to hasten research in this direction that would include articulating clearer statements for the challenges, developing benchmarks to compare solutions, and bringing research to practice through software.

KEYWORDS:

• Data-driven site characterisation (DDSC)
• ugly data
• MUSIC-3X
• site recognition
• stratification

Acknowledgments

The first author is grateful to Professor Zhao-Hui Lu, Symposium Chair for the 7th International Symposium on Reliability Engineering and Risk Management, 12–14 November 2020, Beijing, China, for inviting him to deliver the second Alfredo Ang Lecture. This paper summarises the key observations presented in this lecture.

Disclosure statement

No potential conflict of interest was reported by the author(s).