Bayesian analysis for learning and updating geotechnical parameters and models with measurements - Risk and Reliability in Geotechnical Engineering - page 217

bayesian analysis for learning and
updating geotechnical parameters
and models with measurements
Daniel Straub and Iason Papaioannou
5.1 IntroDuCtIon
Geotechnical planning and construction is typically associated with large uncertainties
and limited data on site conditions. To describe the geotechnical performance as accu-
rately as possible, it is thus necessary to combine information from different sources (site
measurements, expert knowledge, and data from literature). The engineers collect a few
hypotheses about site conditions and then gather field observations (e.g., measurements
of deformations, stresses, or other relevant data) to identify the correct hypothesis. As we
show in this chapter, this process can be formalized through Bayesian updating as part
of a probabilistic reliability and risk assessment. Thereby, a prior probabilistic model is
updated with the new data to a posterior probabilistic model, which is then the basis for
further reliability and risk assessments. Bayesian updating has significant advantages over
other methods for learning geotechnical models, due to its flexibility and the possibility to
consistently combine data and observations from various sources with mechanical models
and expert estimates.
In this chapter, we present the basic concepts and theory of Bayesian updating, together
with simple and advanced computational methods and algorithms. Following an introduc-
tion to the theory in Section 5.2, a hands-on presentation of the method is provided in
Section 5.3. By means of didactical examples, it will be demonstrated how the procedure is
implemented through the following steps:
1. Establishing an initial prior model
2. Computing the reliability and risk based on the prior model
3. Describing new observations and data
4. Updating the model
5. Updating the reliability and risk
6. Communicating the results
In the general case, Bayesian updating is performed numerically on computationally
demanding models, and efficient algorithms are thus required. Section 5.4 provides an intro-
duction to the state-of-the-art algorithms, starting out with conceptually easy-to-understand
algorithms and closing with an overview on computationally more efficient algorithms. In
Sections 5.5 and 5.6, we present two applications of Bayesian updating, which highlight
different aspects of the theory. In Section 5.5, we consider a transmission tower foundation,
where the probabilistic model and the reliability estimate are updated with site measure-
ments and with observed performance under past loadings. In Section 5.6, we demonstrate
Bayesian updating in the context of finite-element models; deformation measurements are
used to update a spatial random field model of soil parameters as well as the reliability. We
221