Summary and Conclusions
This chapter summarizes all foundation load test databases presented in Chapters 4-7. Those databases with names prefixed by NUS are available upon request. It also offers a comprehensive survey of performance databases for other geostructures beyond foundations and their associated ULS model factor statistics. The geostructures include mechanically stabilized earth (MSE) walls, soil nail walls (SNW), pipes and anchors, slopes and braced excavations. The indicative model statistics in the Joint Committee on Structural Safety (JCSS) Probabilistic Model Code are updated using the model factor statistics presented in Chapters 4-8. A practical three-tier scheme for classifying model uncertainty according to the model factor mean and coefficient of variation (COV) is proposed. This classification scheme is deemed reasonable based on the extensive statistical analyses covering numerous geotechnical structures and soil types. It will provide engineers/designers with an empirically grounded framework for developing resistance factors as a function of the degree of site/model understanding - a concept already adopted in design codes, such as the CHBDC (CSA 2019) and being considered in the new draft for Eurocode 7 Part 1 (EN 1997— l:202x) (e.g. CEN 2018; Franzen et al. 2019).
Generic Foundation Load Test Database
Phoon and Kulhawy (2005) stated that comprehensive databases with well- documented field and laboratory tests are a good tool to assess geotechnical model uncertainty in the form of a model factor defined as the ratio of measured to calculated value of a design quantity. Lesny (2017a) outlined the use of databases for model uncertainty assessment. The authors opined that model uncertainty can only be reliably quantified if sufficient information on the actual behaviour of a geotechnical structure is available. Such information may be obtained from measurements on previous construction or preferably from load tests performed for the purpose of calibrating existing design methods or verifying new design methods. The usefulness of load test data is attested by the development of pile design methods. Although a significant number of load tests exist, as they are mandated by building regulations worldwide, the number of properly documented load tests that are publicly available for the characterization of model uncertainty is relatively limited, as presented in Section 6.5. As noted by Phoon (2020), we already have a lot of data, but the vast majority is shelved after a project is completed - “dark data.”
The generic foundation load test database compiled by the authors is summarized in Figure 8.1. This database has been used to characterize the model factor of foundation capacity in Chapters 4-7. It covers shallow foundations, offshore spudcans and various pile foundations. Three test types are involved. The first is a 1-g laboratory test at a small model scale, such as the 103 case records for helical anchors in clay (Tang and Phoon
2016). The resulting stress level significantly differs from the actual stress level. This stress level effect is particularly important and significant for footing in sand, as discussed in Chapter 4. This type of test was commonly carried out in the early stage of development of footing design methods, such as most research studies by Meyerhof. The second is also a scaled model test but performed in centrifuge facility under ng conditions (more expensive than 1-g laboratory model tests) to create the stress level similar to that caused by actual foundation sizes, including 31 tests for shallow foundations in sand-over-clay, 212 tests for offshore spudcan in sand-over- clay and layered clay with sand and 141 tests for shallow foundations in
Figure 8.1 Distribution of load tests in the generic foundation load test database according to foundation type sand. In 1-g laboratory and centrifuge model tests, the test conditions are controlled, in which soil samples are well-prepared to meet a prescribed stratification profile and soil properties in each stratum. Centrifuge testing in rocks appears to be limited (e.g. Leung and Ко 1993; Dykeman and Valsangkar 1996). The third is a large-scale field or prototype test conducted under natural ground conditions, including 778 load tests on shallow foundations and 3,619 load tests on various pile types (300 on steel H pile, 943 on concrete/steel pile, 542 on drilled shaft, 721 on rock socket and 1,113 on helical pile). This type of test will correctly account for the actual stress level and ground conditions but requires the most efforts and cost. The model statistics are expected to be influenced by extraneous factors, such as the transformation error inherent in the estimation of geotechnical design parameters based on in situ test data (e.g. SPT or CPT) and the bias arising from the interpretation of load test results. Therefore, the following aspects are of special importance in model uncertainty assessment (Lesny 2017a): (1) soil conditions (natural or laboratory controlled), (2) test scale (prototype or scaled model) and (3) interpretation of load test results (e.g. estimation of geotechnical design parameters and definition of foundation capacity).
Model Factor Statistics for Other Geostructures
The performance of common foundation types is well studied because of the availability of load test databases. The field performance of other geotechnical structures is less well documented in the literature. A comprehensive review covering the resistance of mechanically stabilized earth (MSE) retaining structures, embedment depth of cantilever retaining walls, FS for slope and base heave stability and wall and ground movement was conducted by Phoon and Tang (2019). It is reproduced in this chapter for completeness. It is worth mentioning that liquefaction databases are available (e.g. Andrus et al. 1999; Cetin et al. 2004; Moss et al. 2006; Idriss and Boulanger 2010; Juang et al. 2013; Ku et al. 2012; Kayen et al. 2013), but they are only relevant to some seismic regions. Tunnelling-induced ground movements are widely monitored, but no comprehensive databases have been compiled thus far to the best of the authors’ knowledge.