Aspects on probabilistic approach to design From uncertainties in pre-investigation to final design

University dissertation from Stockholm : KTH Royal Institute of Technology

Abstract: Geotechnical engineering is strongly associated with large uncertainties. Exploring a medium (soil) that is almost entirely and completely hidden from us is no easy task. Investigations can be made only at discrete points, and the majority of a specific soil volume is never tested. All soils experience inherent spatial variability, which contributes to some uncertainty in the design process of a geotechnical structure. Furthermore, uncertainties also arise during testing and when design properties are inferred from these tests. To master the art of making decisions in the presence of uncertainties, probabilistic description of soil properties and reliability-based design play vital roles. Historically, the observational method (sometimes referred to as the “learn-as-you-go-approach”), sprung from ideas by Karl Terzaghi and later formulated by Ralph Peck, has been used in projects where the uncertainties are large and difficult to assess. The design approach is still highly suitable for numerous situations and is defined in Eurocode 7 for geotechnical design. In paper I, the Eurocode definition of the observational method is discussed. This paper concluded that further work in the probabilistic description of soil properties is highly needed, and, by extension, reliability-based design should be used in conjunction with the observational method. Although great progress has been made in the field of reliability-based design during the past decade, few geotechnical engineers are familiar with probabilistic approaches to design. In papers II and III, aspects of probabilistic descriptions of soil properties and reliability-based design are discussed. The connection between performing qualitative investigations and potential design savings is discussed in paper III. In the paper, uncertainties are assessed for two sets of investigations, one consisting of more qualitative investigations and hence with less uncertainty. A simplified Bayesian updating technique, referred to as “the multivariate approach”, is used to cross-validate data to reduce the evaluated total uncertainty. Furthermore, reliability-based design was used to compare the two sets of investigations with the calculated penetration depth for a sheet-pile wall. The study is a great example of how a small amount of both time and money (in the pre-investigation phase) can potentially lead to greater savings in the final design.