Reliability-Based Design in Geostructural Engineering

Abstract

ENGLISH ABSTRACT: Current standards used in South African geotechnical design practice follow the partial factor limit states design approach. Although partial factors used in this approach are calibrated for standardised target reliability levels, the approach does not take direct account of the probability of failure. Probabilistic reliability analysis provides additional insight to practitioners, making for potentially more optimal geotechnical structures. The technical committee responsible for drafting the South African geotechnical design standard has been requested by representatives of the geotechnical community to consider including standardised guidelines to reliability based geostructural design. As background research towards compiling the pre-normative report for such a set of guidelines, the use of reliability analysis in geostructural design needs to be considered in the context of the following problems. Firstly, the reliability analysis techniques appropriate to different geostructural design problems – a design standard can only be successful if the analysis methods are sufficiently accurate, robust, and practical to apply. Secondly, the appropriate statistical descriptions of the various model parameters – reliability analysis of civil structures is extrapolatory by nature, and so is very sensitive to the choice of functional form (distribution) and the values used to constrain its parameters. Thirdly, the minimum requirements on sample quality – sophisticated reliability analysis techniques have little value if parameter values are based on biased samples that are not representative of the material upon which the structure is to be founded. This thesis presents a number of studies associated with the issues listed above, from which it is concluded that geostructural design to a target reliability is possible, provided that the resistance model and the statistics describing its parameters are accurate and unbiased. However, it is shown that for this to be achieved, samples consisting of a greater number of specimens than currently used in routine geotechnical practice would be required, while the distribution types used to represent the various geotechnical material parameters should be standardised. Of the range of reliability analysis techniques available, the variants of the first order reliability method (FORM), in combination with an analytical surrogate performance function (response surface) where required, are shown to provide the best balance of transparency, economy, and accuracy.

AFRIKAANS OPSOMMING: Geostrukturele ontwerpstandaarde wat tans in die Suid Afrikaanse praktyk gebruik word, volg die parsiële faktor limietstaatontwerp benadering. Hoewel die faktore in dié benadering teen gestandaardiseerde teikenbetroubaarheidsvlakke gekalibreer is, word die faalwaarskynlikheid nie direk in ag geneem nie. Betroubaarheidsanalise en -ontwerp met ’n direkte waarskynlikheidsgrondslag bied die praktisyn dieper stogastiese insig, wat potensieel meer optimale geotegniese struktuurontwerp kan bewerkstellig. Die tegniese komitee verantwoordelik vir die opstel van die Suid Afrikaanse geotegniese ontwerpstandaard, is deur verteenwoordigers van die geotegniese ingenieursprofessie versoek om die insluiting van gestandaardiseerde riglyne vir betroubaarheidsontwerp te oorweeg. As grondslag vir die verwante konsepverslag is dit nodig om geostrukturele betroubaarheidsontwerp in die konteks van die volgende probleme te oorweeg. Eerstens, die metodes vir betroubaarheidsanalise van verskeie geostrukturele ontwerpsprobleme – ’n ontwerpstandaard kan slegs suksesvol wees met akkurate, veelsydige, toepasbare analise metodes. Tweedens, die gepaste statistiese beskrywings vir die verskeie modelparameters – betroubaarheidsanalise van siviele strukture is uiteraard ekstrapolerend, en dus baie sensitief vir die keuse van verdelingstipe en parameterwaardes. Derdens, die vereiste steekproefgrootte en -kwaliteit – gevorderde betroubaarheidsanalise is van min nut wanneer parameterwaardes sydig en nie verteenwoordigend van die struktuur se onderliggende grondlaag is nie. Hierdie tesis bevat ’n stel navorsingstudies gemik op dié vrae, en bevind dat geostrukturele ontwerp teen ’n teikenbetroubaarheidsvlak uitgevoer kan word, mits die weerstandsmodel én die statistiese beskrywing van parameters, akkuraat en onsydig is. Dít is egter slegs moontlik met meer steekproefwaardes as wat tans die norm in die praktyk is, terwyl verdelingstipes vir die verskeie geotegniese parameters gestandaardiseer behoort te word. Uit die verskeidenheid metodes vir betroubaarheidsanalise is bevind dat die verskillende vorme van die lineêre betroubaarheidsmetode (‘first order reliability method’, FORM), waar nodig gekombineer met ’n vervangde limietstaatfunksie (responsoppervlak), die optimale balans van duidelikheid, doeltreffendheid, en akkuraatheid bied.