Bayesian Learning Methods for Geotechnical Data
Publication: ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering
Volume 7, Issue 1
The special collection on Bayesian Learning Methods for Geotechnical Data is available in the ASCE Library (https://ascelibrary.org/ajrua6/bayesian_methods_geotechnical).
Probabilistic methods are frequently criticized as inapplicable to geotechnical data because sample size is small and uncertainties are significant. Because geotechnical data emerge from naturally occurring materials, every site is to some extent unique. Multivariate data (which are commonly encountered in site investigation programs) can be substantially incomplete and typically indirect (that is, a transformation model is needed to convert the measurements to the desired design parameters). One notable feature of big data is veracity. The indicative percentage of outliers in commercial site investigation reports is unknown at present. In short, the characteristics of geotechnical data, which can vary in space and time, can be succinctly described as MUSIC-X: multivariate, uncertain and unique, sparse, incomplete, and potentially corrupted, with X denoting the time and/or spatial dimension (Phoon et al. 2019).
The frequentist approach that remains the dominant paradigm for dealing with uncertainties defines probabilities in terms of the occurrence frequencies of repeated events. Its application to geotechnical engineering has been problematic—the most common limitation is that sample size is too small to characterize statistics with any confidence and to identify the correct probability distribution function. Some engineers also argue that different sites cannot be treated as repeated events because each site is unique in its own ways. This issue is denoted by the U in MUSIC. It is not well recognized among engineers that these objections do not apply to the Bayesian approach. The Bayesian approach is, at heart, a logic for reasoning in the presence of uncertainty in a principled way. The benefits of Bayesian reasoning include natural and unified modeling of many difficult data-driven problems, the ability to accommodate unstructured data, and powerful algorithms for data fitting and analysis under uncertainty.
Although Bayesian methods have gained increasing attention in the last decade, there are still some critical and important issues for further investigation:
•
Computational issues: The computational cost of Bayesian methods is a critical issue that hampers the use by engineers. This issue becomes even more important as big data have to contend with volume, variety, and velocity.
•
Problem dimension: Many existing Bayesian methods struggle with a large number of uncertain parameters. This issue is denoted by the M in MUSIC.
•
Model selection: Although model selection is available in the Bayesian framework, computational issues remain for the large number of possible model classes. For example, in the selection of the empirical formula, there are a large number of possible combinations with only 5 or 10 variables because these variables can appear in different functional forms and there can be cross terms.
•
Sparse learning: In geotechnical applications (such as site investigation), sparse data are often encountered. Although some studies on sparse learning have already been conducted in geotechnical engineering, there is still a lot of room for development.
•
Data quality: Due to the difficulty and disturbance in soil and rock testing, the data quality in geotechnical engineering is usually not as high as in other disciplines. It will be crucial to screen out erroneous data, especially that the number of data points is usually small in this field. This issue is denoted by the C in MUSIC.
In this special collection, the papers address the preceding issues to certain extent. In addition to the theoretical developments, they also cover a variety of physical problems in geotechnical engineering, including estimation of clay parameters, estimation of compressive strength of rock, slope stability, liquefaction, and site investigation. Research has shifted from models that are motivated by theoretical tractability to models that are motivated by features of real data. The guest editors believe that this increasing emphasis on real data and developing fast algorithms to maximize value for decision-making is critical for the digitalization of geotechnical engineering. We hope to attract further attention and interest in this area.
References
Phoon, K. K., J. Ching, and Y. Wang. 2019. “Managing risk in geotechnical engineering: From data to digitalization.” In Proc., 7th Int. Symp. on Geotechnical Safety and Risk (ISGSR 2019), 13–34. Singapore: Research Publishing.
Information & Authors
Information
Published In
History
Received: Aug 7, 2020
Accepted: Aug 19, 2020
Published online: Oct 16, 2020
Published in print: Mar 1, 2021
Discussion open until: Mar 16, 2021
Authors
Metrics & Citations
Metrics
Citations
Download citation
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.