Beruflich Dokumente
Kultur Dokumente
Abstract
The paper presents a procedure for correlating soil physical and mechanical
parameters from CPT data, illustrated by an example of Young modulus E
evaluation from the cone tip and sleeve resistance test values. A statistically valid
technique to group soils by the analogy of their correlations between Young
modulus 𝐸 and CPT resistance is proposed. The correlation equations are
generated. A method for developing a single nonlinear regression model for a wide
range of soils, improving the accuracy of prediction, is presented.
Key words: cone penetration test, correlation equations, deformation
modulus
1. Introduction
Aсcording to ii. 5.3.5 and 5.3.10 СП (Construction Code) 22.13330.2011
[19] it is allowed to determine E, c and φ values for structures of importance class
II from valid regional data, given in regional construction codes, only from CPT
data without comparison with parallel tests of the same soil. The correlation
equations and tables (Appendix И, СП 47.13330 [20]) for such indirect evaluation
of soil strength and stiffness parameters shall be elaborated by means of parallel
comparison with laboratory test data (tri-axial, one-plane shear, compression tests,
etc.) and field tests (plates of different areas, pressure meter), related to their
geological formation history, age and genetic type [4,5].
Major differences of physical and mechanical parameters of clay and sand
soils of different genesis and age have been identified by Ryzhkov and Isaev
(2010) [17], Ignatova (2009, 2014) [10, 11], Ziangirov and Kashirsky (2005) [9],
Sanglerat (1972) [35]; Lunne и Christoffersen (1983) [30]; Senneset et al. (1989)
[36], were also referred to in regulations SP(СП) 47.13330-2012 [20], ТСН 50-
302-2004 [22], ТСН 50-304-2001 [23].
The current State Standard (GOST) ГОСТ 25100-2011 [7] recommends to
identify disperse soils by their types and subtypes, varieties and sub-varieties along
with their genesis and material composition. Varieties of disperse soils are defined
by their quantitative material composition indices, structure, conditions and
properties in accordance with Appendices Б.2 and В.2. Table 2 of this GOST
relates rubble, sand, clay and organic mineral soil to subtypes of disperse soils.
When defining varieties of disperse soils there are used indices of their properties:
grain size, grain size composition, grain size heterogeneity rate, water saturation
rate, porosity of loose soils and plasticity index, liquidity index, etc. of clay soils.
All these indices characterize soil physical properties. ASTM D 2487 [25] and ISO
14688-2 [28] classifications are compiled on similar principles, while Appendix Е
ГОСТ (GOST) 25100 [7] gives tables of correspondence of disperse soils names in
Russian and international standards.
If it is not possible to directly determine soil physical parameters, their
classification is difficult, and unambiguous determination of their subclass
(cohesive, non-cohesive) and subtype (sand, clay, silt, peat) is only possible by
sampling from borehole. Nevertheless, both abroad (Begemann, 1965 [26]), and
then in Russia (СН-448-72 [18]) in order to classify soils the following parameter
came into use as calculated by equation 𝑅𝑓 = (𝑓𝑠 ⁄𝑞𝑐 ) ∗ 100 %, with 𝑓𝑠 as specific
resistance along the probe side surface and 𝑞𝑐 as unit tip resistance. Later there
appeared diagrams (Cf. Boldyrev, 2013 [3]; Ryzhkov and Isaev, 2010 [17]),
plotted on the basis of several rather than one measured and calculated CPT
parameters: Olsen и Douglas (1981) [27], Robertson (1986 и 1990) [33, 34],
Jefferies и Davies (1991) [29], Olsen and Mitchell (1995) [32], et. al. Varieties of
disperse soils are singled out from CPT data as per certain zones on the
classification diagram. The zones boundaries (up to 12) on the diagrams, were
obtained by comparison of CPT data in parallel with in-lab research. Fig.1 shows
the classification diagram, elaborated by Robertson in 1990 [33]), which is broadly
applied for geotechnical investigations worldwide. The numbers of the zones
correspond to soil types, as is shown in Table 1.
Table 1 Compares Russian and ASTM D 2487 classification; therein
parameters 𝑃𝐼 and 𝐿𝐿 (soil plasticity and liquidity indices as per ASTM D 2487 are
recalculated with the help of equations, recommended by GOST 25100, with 𝑃𝐿 =
𝑊𝑃 , 𝐿𝐿 ≈ 1,48 ∗ 𝑊𝐿 − 8,3, and 𝑃𝐼 = 𝐿𝐿 − 𝑃𝐿.
Table.1 Classification of soils by Robertson (1990)
Data pre-processing
Correlation analysis
Linearization
Factor analysis
Statistical
analysis
Development of correlation equations
𝑛 𝑛𝑗
2
1 2
𝜎вн = 𝑛 ∑ ∑(𝑦𝑗𝑙 − 𝑦̅𝑗 ) (1)
∑𝑗=1 𝑛𝑗 − 𝑛
𝑗=1 𝑙=1
with 𝑛𝑗 as the number of measurements of the value in j-th EGE; 𝑛 is the number
of EGEs; 𝑦𝑗𝑙 as particular values of the variable within j-th EGE; 𝑦̅𝑗 as average
value of a variable within j-th EGE.
In order to correct the sample representativeness deficit there was introduced
a weight function with values of each EGE equal to the number of measurements
of parameters in the EGE equal to the number of measurements of parameters in
this EGE (Наследов, 2005) [13].
2.2. Correlation analysis
In the analysis there is established the direction and the shape of the link
between variables, its tightness is measured and confidence intervals are assessed
for correlation coefficients and statistical significance of their difference from zero
[12]; multiple collinearity diagnostics is performed [15], and there are excluded
variables, having no statistically significant correlation with Young modulus as
well as those closely correlated between themselves [2].
2.3. Linearization of relationships
In order to linearize final regression equations there are modified the
variables, having non-linear relationship with Young modulus [12].
2.4. Factor analysis
In the course of factor analysis there is performed classification of statistical
analysis variables and reduction of their number along with latent variables
identification [2].
2.5. Determination of correlation relationships
An important element of statistical analysis is soils grouping by the character
of the objective parameter correlation links (as per E) with CPT parameters. At this
stage group elements are selected, which are measured or calculated in CPT (see
Table 2), whose values are the best to delineate soil groups with similar correlation
links behavior.
Then the optimal number of regression equations is selected, which, on one
hand, shall ensure their highest accuracy, “geared” for a particular group of soils,
and, on the other hand, excludes excessive number of regression equations. With
the limited number of statistical data it is necessary to check that for the generation
of each regression equation there should be sufficient amount of experimental data.
Grouping soils as per the character of correlation links is based on the
following assumptions:
1. Each group of soils, having similar type of correlation links between CPT
parameters and the objective function of soil properties, is assigned a separate
correlation equation. Also such grouping may not coincide with GОSТ 25100,
ASTM D 2487, ISO 14688-2 classifications, namely one correlation equation
could be applied to different soil types (clay, clay loam, sand) or different
correlation equations can be used for one soil (Fig. 3).
2. In order to select the type of the correlation equation there are used CPT
parameters rather than physical parameters.
3. The ranges of parameter values for grouping soils in accordance with the
character of correlation links is assigned on the basis of sensitivity of the
correlation links character to variation of these very parameters. In the process of
soil grouping versus their correlation links there could be applied several levels of
the classification tree (groupings).
4. Grouping soils by the character of CPT parameters correlation links with
soil parameters (𝐸, 𝜑, 𝑐 и др.) can differ for various parameters.
Algorithm of soils grouping by the character of their links
Fig. 3. Robertson classification diagram Robertson (1990) and application zones of regression
equations.
(8)
𝐸 = 𝑓1 (𝐹𝑟 , 𝐼𝑐 ) + 𝑓3 (𝐹𝑟 , 𝐼𝑐 ) ∗ 𝑞𝑐 + 𝑓3 (𝐹𝑟 , 𝐼𝑐 ) ∗ 𝑓𝑠 .
Coefficients of non-linear regression model (8) were computed with the help
of SPSS Statistics software [15]. There were checked linear, quadratic, cubic and
exponential functions 𝑓𝑖 (𝐹𝑟 , 𝐼𝑐 ). After removing low-significance coefficients
from equation (8) there was obtained the following unified non-linear regression
equation for evaluating Young moduli of cohesive and non-cohesive soils:
2 E evaluation
Δ, МПа 𝑅𝑎𝑑𝑗
Name of
Soil type 𝑞𝑐 𝑓𝑠 𝐼𝐿 𝑄𝑡 𝐼𝑐 accuracy
classification
✓ 4,53 0,714 E
Cohesive ✓ ✓ 4,23 0,734 ↓ evaluation
accuracy
✓ ✓ 4,01 0,749 growth
GOST 25100
✓ 2,53 0,817 E
Non-cohesive ✓ ✓ 1,86 0,840 ↓ evaluation
accuracy
✓ ✓ ✓ 1,76 0,849 growth
тип 3 ✓ ✓ 4,84 0,704 E
Robertson тип 4
↑
Clay and
dust fractions
✓ ✓ 5,37 0,656 ↓ evaluation
(1990) тип 5 content ✓ ✓ 1,84 0,843 accuracy
тип 6 ✓ ✓ 1,62 0,851 growth
зона I ✓ ✓ 4,43 0,719
зона II ✓ ✓ 3,12 0,795
Grouping by E
the character зона ↓
Clay and
✓ ✓
of correlation III ↑ dust fractions 2,45 0,821 evaluation
accuracy
зона content
links ✓ ✓ 1,40 0,864 growth
IV
зона V ✓ ✓ 1,13 0,883
Non-linear Cohesive and non-
cohesive ✓ ✓ 1,67 0,872
regression
Table 4 shows that inclusion of parameter 𝑓𝑠 in regression equation in
addition to 𝑞𝑐 reduces error ∆ by 6,6 % for cohesive soils and by 26,5 % for non-
cohesive soils. As to cohesive soils, the inclusion of parameter 𝐼𝐿 in the regression
equation in addition to 𝑞𝑐 reduces error ∆ by 11,5 % as compared to the equation
that includes 𝑞𝑐 alone. As to cohesive soils, there may be included an additional
variable Qt in the regression equation if the lateral soil pressure 𝜎𝑣0 , at rest is
known that reduces error ∆ by 5,4 % as compared to equation that included only 𝑞𝑐
and 𝑓𝑠 .
Рис. 5 Young modulus, calculated with the help of two different linear equations for
cohesive and non-cohesive soils (10) and (11) (diagram 2), general non-linеar equation (diagram
1) and the average test-plate Young modulus for the GEG ( )
The increase of clay and dust fractions content in soil lowers the accuracy of
Young modulus evaluation as per CPT data: the regression equation error ∆ is
79,1 % greater than that for non-cohesive soils (GOST 25100 classification); as to
Robertson classification (1990) for clays (type 3), the error is 198,8 % greater than
for sands (type 6). The increase of ∆ results in reduced value of the design Young
modulus, as compared to the derived one.
In order to determine the design value of Young modulus from CPT data
(the regression equation includes 𝑞𝑐 , 𝑓𝑠 and the constant term) application of the
proposed technique for grouping soils in this paper reduces error ∆ for cohesive
soils by 27,0 % on the average for GOST 25100 classification and by 53.5% for
Robertson classification. As for non-cohesive soils the error ∆ reduction is 46,5 %
GOST 25100-2011 classification and 36,2 % for Robertson classification.
Application of the non-linear regression model with parameter 𝐼𝑐 inclusive
enabled reduction of ∆ error for non-cohesive soils by 10,2 %, for cohesive soils
and by 60,5 % as compared to the conventional method of linear regression with
one parameter 𝑞𝑐 .
2
Generation of a high-quality (𝑅𝑎𝑑𝑗 = 0,872) regression equation is an
attractive option for a broad range of soils. It enables Young modulus assessment
both for “perfect” clays and sands with mean values from a training sample of
physical mechanical parameters and for many transition soil types (e.g. sand
loams) that results in prediction improvement. An example, comparing Young
modulus assessment with application of linear regression equation for non-
cohesive soils
𝐸 = 14,429 + 1,254 ∗ 𝑞𝑐 (10)
and cohesive
(11)
𝐸 = −3,652 + 9,687 ∗ 𝑞𝑐
soils and the general non-linear regression equation (9) is shown on Fig. 5.
Conclusions
1. Incorporation of parameter fs in addition to qc in the regression equation
for evaluation of Young modulus of soils reduces the error ∆ by 6,6 % for the
subclass of cohesive soils and by 26,5 % for non-cohesive soils.
2. If soil physical parameters are known then it is possible to include
parameter 𝐼𝐿 in addition to 𝑞𝑐 that gives reduction of error ∆ by 11,5 % for
cohesive soils; inclusion of parameter 𝑄𝑡 in addition to 𝑞𝑐 and 𝑓𝑠 reduces error ∆
by 5,4 % for non-cohesive soils.
3. If the content of dust and clay fractions grows in soils then lowers the
accuracy of Young modulus evaluation from СPT data, the error ∆ for the subclass
of cohesive soils is by 79,1 % greater than that for non-cohesive soils (GOST
25100); as for Robertson classification (1990) for clays the error ∆ is 198,8 %
greater than that for sands..
4. Classification of soils by the character of correlation dependences
between Young modulus and CPT parameters (regression equations includes 𝑞𝑐 , 𝑓𝑠
and the intercept term) enables higher precision in evaluation of the design value of
Young modulus for cohesive soil i.e., 27,0 % as per GOST 25100 classification
and 53,5 % as per Robetrson classification, for non-cohesive soils by 46,5 % and
36,2 % respectively.
5. Non-linear regression model application, with parameter 𝐼𝑐 included,
enables compilation of general regression equations for a broader range of soils,
with Young modulus assessment both for “pefect” clays and sands and for soils
and for a multitude of intermediate soil types, e.g. sand loams. In this very example
it was possible to reduce error ∆ for non-cohesive soil by 10,2 % and for cohesive
soils by 60,5 % as compared to the classical method of linear parameter q
regression.
REFEREnCES
1. Айвазян, С.А. Прикладная статистика. Исследование зависимостей:
справочное издание / С.А. Айвазян, И.С. Енюков, Л.Д. Мешалкин; под ред.
С.А. Айвазяна. – М.: Финансы и статистика, 1985. – 487 с.
2. Айвазян С.А. Прикладная статистика: Классификация и снижение
размерности: Справочное издание / С.А. Айвазян, В.М. Бухштабер, И.С.
Енюков, Л.Д. Мешалкин; под ред. С.А. Айвазяна. – М.: Финансы и
статистика, 1989. – С. 607.
3. Болдырев, Г.Г. Полевые методы испытаний грунтов (в вопросах и
ответах) / Г.Г. Болдырев – Саратов: Издательский центр «РАТА», 2013. – 356
с.
4. Болдырев Г.Г., Мельников А.В., Новичков Г.А. Интерпретация
результатов полевых и лабораторных испытаний с целью определения
прочностных и деформационных характеристик грунтов. Часть I.
Интерпретация результатов полевых испытаний с целью определения
прочностных характеристик грунтов. Инженерные изыскания, №5-6, 2014. –
С. 68-77.
5. Болдырев Г.Г., Мельников А.В., Новичков Г.А. Интерпретация
результатов полевых и лабораторных испытаний с целью определения
прочностных и деформационных характеристик грунтов. Часть III.
Интерпретация результатов полевых испытаний с целью определения
деформационных характеристик грунтов // Инженерные изыскания. – 2014. –
№5-6. – С. 86-97.
6. ГОСТ (GOST) 20522-2012. Грунты. Методы статистической
обработки результатов испытаний. – 2013. – 19 с.
7. ГОСТ (GOST) 25100-2011. Грунты. Классификация. – 2011. – 63 с.
8. Елисеева, И.И. Эконометрика: учебник / И. И. Елисеева; под ред. И.
И. Елисеевой. – М.: Финансы и статистика, 2004. – 344 с.
9. Зиангиров Р.С., Каширский В.И. Оценка деформационных свойств
дисперсных грунтов по данным статического зондирования // Основания,
фундаменты и механика грунтов. – 2005. – № 1. – С. 12-16.
10. Игнатова О.И. Деформационные характеристики юрских глинистых
грунтов Москвы // Основания, фундаменты и механика грунтов. – 2009. – №
5. – С. 24-28.
11. Игнатова О.И. Исследование корреляционных связей модуля
деформации четвертичных глинистых грунтов разного генезиса с удельным
сопротивлением при статическом зондировании // Основания, фундаменты и
механика грунтов. – 2014. – № 2. – С.15-19.
12. Львовский, Е.Н. Статистические методы построения эмпирических
формул: учебное пособ. для вузов / Е.Н. Львовский. – 2-е изд. – М.: Высш.
школа, 1988. – 239 с.
13. Наследов А.Д. SPSS 19. Профессиональный статистический анализ
данных – СПб.: Питер, 2011. – 400 с.
14. ООО «Строй-тех: [Электронный ресурс]. – URL: www.s-teh.com/
(дата обращения: 18.08.2014).
15. Программное обеспечение SPSS [Электронный ресурс]. – URL:
www.ibm.com/software/ru/analytics/spss/ (дата обращения: 18.08.2014).
16. Руководство по составлению региональных таблиц нормативных и
расчетных показателей свойств грунтов / ПНИИИС Госстроя СССР. – М:
Стройиздат, 1981. – 55 с.
17. Рыжков, И.Б Статическое зондирование грунтов: монография / И.Б.
Рыжков, О.Н. Исаев. – М.: Издательство Ассоциации строительных вузов,
2010. – 496 с.
18. СН-448-72 Указания по зондированию грунтов для строительства. –
1972. – 32 с.
19. СП 22.13330.2011 Основания зданий и сооружений. – 2011. – 162 с.
20. СП 47.13330-2012 Инженерные изыскания для строительства.
Основные положения. – 2012 – 110 с.
21. Технический отчет об инженерно-геологических изысканиях по
объекту «Жилой комплекс с торгово-административными помещениями» /
ООО «Строй-тех». – Том I, Шифр 01-01и/2-2013. – Пенза, 2013. – 32 с.
22. ТСН 50-302-2004 Проектирование фундаментов зданий и
сооружений в Санкт-Петербурге. – 2004. – 63 с.
23. ТСН 50-304-2001 Основания, фундаменты и подземные
сооружения. г. Москва. – 2003. – 56 с.
24. Anatolyev S. Durbin–Watson statistic and random individual effects //
Econometric Theory (Problems and Solutions) // Econometric Theory. – 2003. –
Vol. 19. – No. 5. – P. 882–883.
25. ASTM D2487-2006 Standard practice for classification of soils for
engineering purposes (Unified soil classification system). – 2006. – 12 p.
26. Begemann H.K.S. The friction jacket cone as an aid in determining the
soil profile // Proc. of the 6-th Int. Conf. Soil Mech. and Found. – Montreal, 1965.
– Vol. 1. – P. 17-20.
27. Douglas B. J., Olsen R. S. Soil classification using electric cone
penetrometer. Cone penetration testing and experience // In Proceedings of the
ASCE National Convention. – New York, NY, USA, 1981. – P. 209–227.
28. ISO 14688-2:2004 Geotechnical investigation and testing –
Identification and classification of soil – Part 2: Principles for a classification. –
2013. – 13 p.
29. Jefferies M. G., Davies M. P. Soil classification by the cone penetration
test: Discussion // Canadian Geotechnical Journal. – 1991. – 28(1). – P. 173-176.
30. Lunne, T., Christoffersen H.P. Interpretation of cone penetrometer data
for offshore sands // 15-th Annual OTC in Houston. – TX, 1983. – P. 181-192.
31. MATLAB [Электронный ресурс]. URL:
www.mathworks.com/products/matlab/ (Дата обращения: 18.08.2014).
32. Olsen R. S., Mitchell J. K. CPT stress normalization and prediction of
soil classification // Proceedings of the International Symposium on Cone
Penetration Testing, CPT’95. – Linköping, Sweden, 1995. – Vol. 2 – P. 257-262.
33. Robertson P.K. Soil classification using the cone penetration test //
Canadian Geotechnical Journal. – 1990. – No. 27 (1). – P. 151-158.
34. Robertson P.K., Campanella R.G., Gillespie D., Greig J. Use of
piezometer cone data // In-Situ’86 Use of In-situ testing in Geotechnical
Engineering, GSP 6 , ASCE, Specialty Publication. – Reston, VA, 1986. – P.
1263-1280.
35. Sanglerat, G. The penetrometer and soil exploration / G. Sanglerat. –
Amsterdam: Elsevier Publishing Company, 1972. – 488 p.
36. Senneset, K., Sandven, R., Janbu, N. The Evaluation of soil parameters
from piezocone tests // Transportation Research Record. – 1989. – Vol. 1. – P. 24-
37.