You are on page 1of 4

AUTOMATED

MODEL AND
PARAMETER
SELECTION
Incorporating Expert Input into
Geotechnical Analyses
II By Ronald B J. Brinkgreve. PhD

Numerical analysis has major advantages determination could narrow the range of
over traditional methods regarding the level results and increase confidence in using
of detail and insight that can be obtained and accepting the results from numerical
in geotechnical stability analysis, limit-state analysis for geotechnical design.
Excavation for construction of MIT Stata Center in Cambridge. design, and soil-structure interaction
MA. This complex soil-structure interaction problem required From Traditional Methods to Numerical Analysis
advanced numerical methods and soil constitutive models A) problems. A key success factor in numerical Traditional geotechnical analysis and design methods, imple-
nnin computational PLAXIS 3D model of intermediate excavation stage. B) Model mented in easy-to-use computer software. have been around
2U19 GEOTECHNICS settlements at intermediate stage. C) Photo showing intermediate analysis is an appropriate determination of for more than 35 years. They are well accepted in geotechnical
excavation stage; D) Model settlements at near-final state
Graphics courtesy of Zhandos Orazalin and Andrew Whittle. design. In this respect, it is good to realize that easy-to-use
constitutive model parameters based on the numerical analysis software, based o n the finite element or the
available, and too often limited, soil data, finite difference methods, has also been available for 30 years,
but is far less accepted in geotechnical design. It s worth noting
and on engineering judgement. Prediction that in those 30 years, numerical methods and constitutive
models have evolved and improved dramatically, whereas
competitions demonstrate a large variety traditional methods are essentially still the same.
So. given the advantages offered In numerical analyses* why
of results when different engineers solve is our geotechnical profession using old and less sophisticated
the same problem. This gives rise to the methods, whereas other disciplines seem to be innovating
and adopting new technolog)- much faster? Here are four key-
question of whether automated parameter reasons, although there may be others:

3 8 GEOSTRATA JANUARY/FEBRUARY 2019 www geomstitute org

0
Limited confidence and under- Hence, when performing numerical analysis, i t must be based o n experience and engineering judgement. This makes
standing is one of the main reasons that done thoroughly, using constitutive models that capture the parameter determination a difficult task.
geotechnical engineering consultants essential features o f soil behavior required for the applica-
are reluctant to recommend and tion considered. A Paradox
conduct numerical analyses as part of The technical ability to properly execute numerical mod*
the scope of geotechnical engineering Model Parameter Determination elling, with tlx* most appropriate constitutive models and
and design services. Ideally, determination of constitutive model parameters corresponding parameters selection, is a true sign of modem
It s a common perception that is based upon comprehensive site investigation data and geotechnical -engineering expertise and experience. A few
numerical geotechnical analysis is laboratory testing on high-quality soil samples. Engineers practicing geotechnical engineers who have used numerical
difficult, mainly because numerical need to convince project owners that site investigation and lab methods and advanced constitutive models for several years
models require that more parameters be testing are not just costs; they are investments that, provided have developed their own workflow process for parameter
determined than traditional methods. they are thoughtfully planned and properly conducted, pay off determination. They' haw developed databases with local
In numerical analysis, the soils stress- by reducing project uncertainty and risk. data, collected correlations from literature, created their own
strain response is simulated by means In my paper in ASCE GSP 128. 1 offered an overview of elaborate formulas, and implemented them in spreadsheets
of constitutive models. The models methods of parameter determination for various constitutive and tailored software tools. They are tlx* experts. Combined
describe the soil behavior qualitatively, models. Proper parameter determination should be performed with their engineering judgement, it gives them a competitive
while model parameters quantify* in conjunction with the application (e.g.. stress levels, stress edge in numerical modelling, and they sit on it; its their bread
specific aspects of that behavior, such as paths, strain levels, and loading rale), at least when simplified and butter.
stiffness and strength. Note that model soil models arc used. In this way, soil features lacking i n the New or infrequent numerical model users would low it
parameters (e.g.. a reference stiffness constitutive model can. to some extent, be "compensated" if they could somehow benefit from the expert ’s wealth of
parameter in a formula for strcss- by adaptation of the model parameters, but this can be risky. experience and data they' possess. Such data, as well as sets of
dependency of stiffness) are principally Thus, another reason why more advanced constitutive models correlations and their valid domain o f application, and other
Fi gure 1 Example model and results of a 3D FE analysis of a complex barette not the same as soil properties (e.g.. are preferred is that their parameters are generally less depen* sources of information, could lx? made available in public soft-
foundation system. partide composition and pores, relative dent on interpretations by the engineer. ware to help determine model parameters. Some commercial
density, waler content, plasticity index) Parameter determination can lx* based upon: software contains collections of correlations to generate soil
or soil parameters (e.g., friction angle, parameters, but there's no publicly available software that
undrained shear strength), although the ■ Correlations with field test data (from SPT, CPT, or DMT ) automatically determines model parameters for advanced
1. Limited availability' of soil data, variability in the methods of latter may also serve as model parameters. Also note that soil ■ Correlations with soil-index properties like liquid and plastic constitutive models based o n site investigation data.
obtaining and interpreting this data, and the (mispercep- properties may change over time, or, as a result of evolution of limits, plasticity and liquidity indices, and relative density Most of the experts reject the idea o f automatic parameter
tion that limits the benefits of advanced numerical analysis stress and strain during loading (e.g.. grain crushing), whereas ■ Results from laboratory testing (triaxial, oedometer, direct determination, even though, if properly implemented, auto-
2. limited knowledge about the fundamental mechanics model parameters are supposed to be constant. In principle, simple shear) mated parameter generation tools could provide a significant
incorporated in numerical methods and constitutive proper selection of the constitutive model and the correspond- ■ Tables, charts, databases benefit to the industry.
models ing model parameters can properly account for the associated ■ Engineering judgement
3. The (mis)perception that numerical analysis is time- changes in soil behavior. Quality and Reliability of Numerical Analysis
consuming and less productive Once an engineer has crossed the threshold and started Each method provides different parameter values. Their The quality o f results from numerical analysis highly
4. limited confidence and understanding in the results of lierforming numerical analysis, it takes another step to selection requires knowledge of the methods and, more depends o n the ability of the user to select the right
numerical analyses from a "black box" software tool advance from using simplified, linear-elastic perfectly- plastic importantly, their validity, restrictions, and limitations. The constitutive model for the application considered and the
constitutive models to more complex, non-linear elastoplaslic outcomes are not unique and must be properly interpreted, model parameters needed for them. Data from prediction
The too-oftcn-limited availability of soil data and limited models. In general, more complex models include more
knowledge are the main reasons why geotechnical engineers features of soil behavior, but also require more model param-
are reluctant to perform numerical analysis. The (mis)percep- eters to be determined. Nevertheless, for many engineering
tion that numerical analysis is time-consuming and expensive issues it's essential to lake this next step. as illustrated by the
is a reason why managers persist in using traditional methods, following dialogue:
thereby overlooking the benefits that numerical analysis will neighbour: "You're digging a big hole right next
bring (if performed correctly), namely: to my house; it will settle and get damagedr
Mr. Wiseguy: “Don't itorry. My numerical analysis
■ Solving multiple engineering and design Issues tn one analysis
has determined that your house will heme rather
■ Providing more insight in lire soil response and soil -structure
than settle"
Interaction, and the sensitivities therein
■ Allowing the ability to realistically model complex situations Using too -simple constitutive models may lead to the
and coupled processes (Figure I) wrong conclusion. As shown in Figure 2. models o f the same
■ Providing the opportunity to further optimise the overall geometry can produce dramatically different deformation
design behavior by using different soil constitutive parameters. F igure J Deformation results of FE model of same geometry with different constitutive models and parameters.

40 GEOSTRATA JANUARY/FEBRUARY 2019 www geomstitute org

0
soil layers in the profile. The idea is that correlations, and possibly different
automation makes parameter deter- path lengths, that lead to the same
mination less dependent on the user model parameter. The system must be
and reduces the variability of results. able to cope with multiple parameter
So I property 2
It's meant to help mitigate the human values and check the outcomes. After
factor. eliminating the outliers, a mathemati-
interpreting soil-behavior type (SBT) cal procedure (i.c., a weighted average
and soil layering from field tests is or more sophisticated procedure) can
common practice. All geotechnical engi- be applied o n the remaining parameter
User input
neers ought to be familiar with the SPT values to calculate the final param-
interpretation methods of Meyerhof and eter value to be used i n the analysis.
the CPT interpretation diagrams from Parameters determined in this way an*
Robertson and (Campanella. These data supposed to be representative for ser-
provide geometric information about viceability limit state (SLS) situations,
the thickness and lateral distribution whereas for ultimate limit states (ULS)
r jure 3 Schematic representation of the automatic parameter determination process of soil layers and the corresponding the applicable design approach with
SBTs, which can be automated straight- partial factors can be applied. Doing
forwardly. For numerical analysis, a this for all model parameters and for
certain minimum layer thickness should all soil layers provides a collection o f
competitions and benchmarks show a large bandwidth of be used to avoid the generation of very thin soil layers. This automatically generated material data
results, even if the application and corresponding soil data means that some manipulation and “averaging' of thin layers sets (constitutive model ♦ parameters, Figure 4 PLAXIS SodTest module and traditional triaxial test apparatus (inset).
are clearly defined. This uncertainty does not contribute to is needed while maintaining the influence of weak layers. representing each type of soil) that can
confidence in using numerical analysis for geotechnical- Automated selection of the appropriate soil constitutive
design purposes. To build confidence, numerical methods model must recognize and capture the essential features of
and constitutive models used must be transparent, results soil behavior for the soil type as identified from field testing
must be validated, and the uncertainty bandwidth should and. ideally, good -quality laboratory testing. In the majority'
SAVE THE DATE
be narrowed. of cases, the selected model will be an advanced, nonlinear
The quality and reliability of numerical results can be constitutive model that takes stress level-, stress path - and
improved by combining the following elements: strain -dependency of soil behavior into account. Different INTERNATIONAL AIRFIELD & HIGHWAY
models can lx? used. de|x*rxling on the soil types encountered DI
1. Education - Traching/leaming about the capabilities and or the different types of analysis to be conducted. For example, PAVEMENTS CONFERENCE
limitations of constitutive models and numerical methods, a dynamic liquefaction analysis of silty soil layers may require Chicago, Illinois | July 21-24, 2019
and how to properly determine the parameters used i n the that a specific constitutive model be used, but it may not be
models the best model for a static deformation analysis. The selected
2. Experience - Practicing model parameter determination on constitutive model implicitly defines which model parameters
published benchmarks and applying it in real applications need to be determined.
to build the necessary experience To automate selection of the model parameters, all com-
3. Exchange - Exchanging and sharing knowledge* and monly available knowledge should to be implemented in an
experience among geotechnical engineers in industry and expert system. The system must "know" the validity and limita-
academia tions of the correlations, formulas, charts, and tables in it. The
system must also be flexible and extendable to new informa-
Despite decades of attempts to accomplish these three tion as it becomes available and accepted in practice. It should
things via conferences, workshops, communication channels, haw an algorithm to automatically generate valid correlation
online platforms, etc., it seems that the (mis) use of numerical paths (i.e., chains of correlations) that link measured field-test
analysis in geotechnical design is growing at a faster rate than quantities to model parameters via intermediate soil proper-
the quality' of the results. ties (e.g.. relative density*, plasticity index) and soil parameters.
Figure 3 provides a conceptual illustration of model parameter
Automatic Parameter Determination selection based on chains of correlations. Each box represents
An effective way to go forward is to develop an online software a correlation having one or mon* input parameters, and one or
tool for automatic characterization and interpretation of the more conditions. The resulting parameter value may be used in
soil stratigraphy based on standardized site- investigation another correlation.
ASCE's longstanding specialty conference for
data, and the subsequent automatic selection of constitutive Because in many cases more than one correlation will airfield and highway pavement professionals.
models and corresponding model parameters for the different be valid, there are generally different paths, with different

42 GEOSTRATA JANUARY/FEBRUARY 2019 www geomstitute o r g 4 3

0
be automatically assigned to the then be compared with high quality laboratory' test data, ►RONALD B.J. BRINKGREVE, PhD $ manager of the
corresponding soil layers in the which is. ideally, part of a standardized site investigation Competence Centre Geo -Engineering of Plaxis in the
numerical model. report. Such a comparison gives a feel for the accuracy at Netherlands He is responsible for all geo-engineenng related
In general more complex which real soil behavior is approximated by the model, research and development of the PLAXIS Finite Element soft-
Liability and Validation and i t offers insight into the capabilities and limitations ware He rs also an associate professor at Delft University of
Although automated constitutive model of the model. Technology (TUDdft. NL) His mam research interests include
models include more features and parameter selection should lead so«l constitutive models and numerical methods He can be
to more consistent and more reliable The Next Step reached at r.bj brmkgreve&tudelft nl
of soil behavior, but also require results, the ultimate responsibility and Today, geotechnical design is still largely' based o n traditional
liability for geotechnical design remains methods that are decades old. Numerical analysis continues
with the engineer. This means that the to show great potential in providing more insight in under-
more model parameters to be engineer must verify and validate the standing overall ground behavior and helping optimise the
material data sets that are automatically overall design, and the use of numerical modelling continues
determined. created. Therefore, the procedure of to expand within the industry. However, with increased use
automatic parameter determination comes increased risk for misuse and misinterpretation.
must be transparent and verifiable. Automated soil constitutive model and parameter selection
The validation of the behavior may provide a tool to help mitigate the human factor, reduce
represented by the constitutive model tire variability of residls. and increase confidence in numerical
and its parameters can be selected by analysis. I f we join forces, we can develop such a tool together,
simulating laboratory tests (Figure 4). for the benefit of our profession. Anyone who may be inter-
The resulting stress-strain curves can ested is welcome to participate.

CALL FOR ABSTRACTS


ASCE Member Benefit

Access ►Engineering. I PIPELINES 2019


KEY DATES
Coll for Submissions Open: June 20, 2 0 1 8
1 Abstracts Due: August 26, 2 0 1 8

-err-'' CONFERENCE Draft Papers Due: December 1, 2 0 1 8


Registration Opens: January 16, 2 0 1 9
AccessEngineering delivers Nashville, TN | July 21 - 24 Final Papers Due: February 2 1 , 2 0 1 9

multidisciplinary engineering
information through these
dynamic interactive features:

Data Vis

Tutorial Videos

For up-to-date information, visit


www.pipelinesconference.org

44 GEOSTRATA JANUARY/FEBRUARY 2019 www geo«nstitute o r g

You might also like