Minor Project report

On
Design and Probabilistic Analysis of Pile Raft
Foundation

Department of Civil Engineering

National Institute of Technology Patna
Ashok Rajpath, Patna, Bihar-800005
 DECLARATION

We want to declare that we have done our minor project on the design and analysis of
Pile and Raft foundation under the guidance of Prof. Pijush Samui (Professor at NIT
Patna, Dean planning and development) from 12/08/2024 to 25/11/2024 in 7th semester,
while pursuing the degree of bachelor of technology in civil engineering at National
Institute of Technology Patna. The work, which is being presented in the Minor project
report is an authentic record and true to the best of our knowledge.

Signature of Student
 ACKNOWLEDGEMENT

It gives us immense pleasure to have successfully completed our Minor Project on the Design and
analysis of Pile raft foundation under the guidance of Dr. Pijush Samui (Professor NIT Patna, Dean
planning and development). Minor Project is a golden opportunity for learning and enhancing the
practical knowledge of the transformation of the infrastructural professional world.
We owe special debt of gratitude to Dr. Pijush Samui Sir for providing us this excellent hands-on
experience of the actual field work of geotechnical engineering and for guiding us throughout the
project work.
All the practical knowledge that we have gained in this minor project is no less than a treasure for us.
We hope that we can build upon the experience and knowledge that we have gained and make
valuable contributions towards the development of society in the coming future.

We extend our sincere thanks to our director Prof. Pradeep Kumar Jain for reminding us always what
professional ethics is that has made us a better individual, for the advice he has given us and his
support for us to be competitive students worldwide. We are thankful to Prof. Anshuman Singh
(H.O.D Department of Civil Eng.) for giving us this opportunity and for providing us with the
requisite equipment and facilities during this research work. We would like to thank our parents and
friends for their endless support in our daily lives and for having them at our side to guide us always.
 CERTIFICATE

The undersigned certify that Shailendra Yadav, Roll No. 2103108, Kunal Raj Roll No. 2103022,
Mehul, Roll No. 2103113 and Anjali Kumari, Roll No. 2103098 are registered for the B. Tech
program in department of Civil Engineering under my supervision to fulfill the requirement of
CE73131 Minor Project. I hereby recommend that the dissertation entitled “Design and
Probabilistic Analysis of Pile raft Foundation be accepted as Minor Project Report.

Under The Guidance of :

Prof. Pijush Samui

Professor & Dean (P&D),
Dept. of Civil Engineering
NIT Patna-800005
 CONTENTS

❖ Declaration ……………………………………………………..2

❖ Acknowledgement………………………………………………3

❖ Certificate…… …………………………………………………4

❖ Contents…………………………………………….…………5-6

❖ List of figure and table…… ..…………………………………..7

1. Introduction 8-10
1.1. The Project Area
1.2. General Overview of the Pile Raft foundation
1.3. Project Need and objectives

2. Literature review 11-16

2.1 General
2.2 Variability and Uncertainty
2.3 Reliability Analysis and Methods
2.3.1 FOSM
2.3.2 MARS
2.3.3 GP
2.4 Gaps in Literature Review

3. Methodology 17--34
3.2 Multivariate Adaptive Regression Splines
3.2.1 Introduction to MARS
3.2.2 MARS Mechanics
 3.2.3 Details of MARS
3.2.4 Reliability Analysis
3.2.5 First Order Second Moment Method
3.2.6 Methodologies
3.3 Genetic Programming
3.3.1 Introduction
3.3.2 Crossover
3.3.3 Mutation
3.3.4 Methodologies

4. Result and Discussions 35- 41

4 .1 MARS Model
4.2 Genetic Programming
\
5. Conclusions 42

6. References 43
 LIST OF FIGURES

Figure 3.1: SOIL PROFILE……………………………...………………………….…..PAGE NO. 18

Figure 3.2: SOIL PROFILE……………………………………………………….…….PAGE NO. 15
Figure 3.1.1: PLAN OF PROPOSED BUILDING ...…………………………….…….PAGE NO. 18
Figure 3.1.2: SECTIONAL VIEW…………………………………………..…….…….PAGE NO. 21
Figure 3.3.1: GENETIC PROGRAMMING CYCLE ……………..……………….…..PAGE NO. 29
Figure 3.3.2: CROSSOVER .…………………….………………………………….…..PAGE NO. 29
Figure 4.1:MARS MODEL OUTPUT TRAINING DATA..………………...………….PAGE NO. 30
Figure 4.2: MARS MODEL OUTPUT TESTING DATA ……..….…………………....PAGE NO. 31
Figure 4.3: RELIABILITY PERFORMANCE MARS MODEL………………..…........PAGE NO. 31
Figure 4.4: GP MODEL OUTPUT TRAINING, TESTING DATA.………..…………..PAGE NO. 32
Figure 4.5: RELIABILITY PERFORMANCE GP MODEL.………..……………….....PAGE NO. 32

LIST OF TABLES

TABLE 3.1: LOAD CALCULATION…………………………………………………..PAGE NO. 23

TABLE 3.2: …………………………………………………………………………….PAGE NO. 25
TABLE 3.3,3.4: DESIGN OF PILE …………………………………………………….PAGE NO. 25
TABLE 3.5: NORMALISED TRAINING DATA ………………………………………PAGE NO. 26
TABLE 3.6: PREDICTED TRAINING AND TESTING ……………………………....PAGE NO. 27
TABLE 3.7: GENETIC PROGRAMMING PARAMETERS …………………………..PAGE NO. 34
TABLE 4.1: COEFFICIENT OF BASIS FUNCTION …………………………….…..PAGE NO. 35
TABLE 4.2:RELIABILITY INDEX MARS MODEL ………………………………....PAGE NO. 37
TABLE 4.3: PROBABILITY OF FAILURE BY MARS MODEL ………………........PAGE NO. 38
TABLE 4.4:RELIABILITY INDEX GP MODEL ……………………………………..PAGE NO. 40
TABLE 4.5:PROBABILITY OF FAILURE BY GP MODEL ………………………...PAGE NO. 41
 Chapter 1

INTRODUCTION

1.1 Project Area

The project encompasses a total area of 300 square meters, with dimensions of 15 meters by 20
meters. The goal of this project is to design a pile-raft foundation system, taking into consideration
various geotechnical and structural aspects to ensure a stable and reliable foundation for the
structure. Given the project's location and soil conditions, the need for a robust foundation system is
crucial. The design will focus on optimizing the load-bearing capacity while minimizing settlement
and ensuring safety under diverse loading conditions.

1.2 General Overview of Pile Foundation

A Pile Raft Foundation is a type of deep foundation system that combines the benefits of both pile
foundations and raft foundations. It is commonly used for large buildings or structures where the
ground conditions are not suitable for conventional shallow foundations. This foundation type is
especially effective when the soil has weak bearing capacity, or when the structure’s load exceeds
what can be supported by a simple raft or shallow foundation alone.

Components of Pile Raft Foundation

1. Raft (Slab): The raft is a large, thick concrete slab that spreads the load from the
superstructure across a large area. It functions as a rigid base that helps distribute the load
evenly. The raft acts like a shallow foundation, carrying the majority of the load.

2. Piles: Piles are long, slender columns made of concrete, steel, or timber, driven or bored deep
into the ground to provide vertical support. The piles transfer the load to deeper, more stable
layers of soil or rock. Piles are used when the upper soil layers cannot support the load of the
structure.

3. Interaction between Piles and Raft: In a pile raft foundation, the piles and the raft work
together to support the structure. The raft serves to distribute the load, while the piles take a
portion of the load by transferring it to deeper, stronger soil layers. This interaction helps
reduce the overall settlement of the structure and can mitigate risks associated with uneven
soil conditions.
Design Considerations

1. Load Distribution: One of the key advantages of a pile raft foundation is the ability to
distribute the load effectively. The raft carries part of the load directly to the soil, while the
piles carry loads that the soil itself cannot support. The design ensures that the loads are
shared efficiently between the raft and the piles.

2. Settlement Control: Pile raft foundations are designed to minimize differential settlement. In
soft soil conditions, where traditional foundations may lead to excessive settlement, the piles
provide additional support, ensuring the structure remains stable and level over time.

3. Soil-Structure Interaction: The interaction between the soil, raft, and piles is a critical
factor in the design of pile raft foundations. The foundation must be analyzed carefully to
predict how the piles will interact with the surrounding soil and how the raft will distribute
loads across the site.

4. Economical Use of Piles: Pile raft foundations can also be more economical than using either
piles or a raft alone. By using piles in areas where the soil is weaker and relying on the raft
where the soil is stronger, this system optimizes material use and reduces construction costs.

Applications

Pile raft foundations are typically used in projects with the following characteristics:

1. Heavy Loads: Large buildings, such as high-rise structures, industrial plants, or bridges,
where the loads are too high for a conventional foundation to support.

2. Weak or Variable Soil: Locations where the surface soils have low bearing capacity, or the
soil conditions vary significantly across the site.

3. Limited Depth for Piles: In some cases, the design of a pile raft foundation can be
advantageous in areas where deep foundations are necessary but the depth of competent soil
is relatively shallow.

Thus, pile raft foundations are a highly effective solution for buildings in areas with challenging soil
conditions. They offer a balanced approach to distributing loads and minimizing settlement while
making efficient use of materials.
1.3 Project Need and Objectives

The primary need for this project arises from the challenging soil conditions at the site, where
traditional shallow foundations might not offer the necessary stability or load-bearing capacity. By
using a pile-raft foundation, the project ensures that both the vertical and lateral loads are
effectively transferred to deeper, more stable soil layers, minimizing the risks of excessive settlement
or foundation failure.

The project objectives are as follows:

1. To Design Pile-Raft Foundation Based on Standard Penetration Test (SPT): The

foundation design will be based on the results of SPT, which provides critical data about soil
strength and behavior at various depths. This will ensure that the piles and raft are optimized
to handle the site-specific conditions.

2. To do Reliability Analysis of Pile Foundation Using MARS and Genetic Programming

(GP): The reliability of the pile foundation will be assessed using advanced techniques like
MARS (Machine-Assisted Reliability Study) and GP. MARS will aid in evaluating the
uncertainty in the pile design, while GP will be used to investigate the subsurface conditions,
ensuring that the piles are installed in the optimal locations and depths.

3. To Determine Probability of Failure: The probability of failure of the pile foundation will
be calculated based on various parameters, such as load-bearing capacity, soil conditions, and
the integrity of the piles. This will allow for a thorough understanding of the potential risks
involved and guide in designing a system that minimizes failure likelihood.

By achieving these objectives, the project aims to deliver a safe, cost-effective, and reliable
foundation solution for the structure, addressing both engineering challenges and environmental
conditions.
 Chapter 2

LITERATURE REVIEW

2.1 General

Estimating risk and analysis of probability of failure and reliability are indispensable part of any
project. Uncertainty is natural in engineering problems and it may either be ignored, addressed by
using conservative ideas, by applying the observational method, or by using reliability analysis to
quantify it. As early as in mid 1970s, there was huge attention of researchers in this field. However,
much early, Casagrande (Casagrande 1965b) has floated the term “calculated risk” in 1965. He
described it as “the use of imperfect knowledge to estimate the possible ranges of all pertinent
quantities that enter into a solution” and “the decision on an appropriate margin of safety, or degree
of risk taking into consideration the losses that would result from failure”. He divided it into two
categories: human risk and unknown risk. Observational methods were first conceptualized by
Terzaghi as “learn-as-we-go” method (Terzaghi 1943) and significantly proposed by Peck in his
Rankine papers (Peck 1969). Peck quoted about the limitations of the method, “The method is not
without its pitfalls and limitations. It should not be used unless the designer has in mind a plan of
action for every unfavorable situation that might be disclosed by the observations. The observations
must be reliable, must reveal the significant phenomena, and must be so reported as to encourage
prompt action. The possibility of progressive failure may introduce a serious element of uncertainty”.

Although the Reliability based designs (RBD) had become quite popular in other fields, it’s
application in geotechnical field was not much exciting. A committee was formed which examined
the issue and gave future recommendations. Among the early applications of RBD in geotechnical
field were (Folayan et al. 1970) and (Hoeg, K. Murarka 1974). (Folayan et al. 1970) applied
probabilistic methods to determine the factor of safety of settlement of marshland. (Hoeg, K.
Murarka 1974) applied probabilistic method for the reliability analysis of retaining wall and
compared it with the results of deterministic and conventional methods. He argued for the use of
coefficient of variation instead of ‘safety margin’. As per him, the factor of safety approach and
deterministic approach gives misleading results. Among other early works are (Athanasiou-Grivas et
al. 1977; EINSTEIN 1988)

2.2 Variability and Uncertainty

Geotechnical problems are naturally associated with a variety of uncertainties. Various sources of
uncertainties make soil highly variable in nature. There has been series of studies in this field
(Kulhawy 1993; Phoon and Kulhawy 1999b, c; Christian 2004). (Terzaghi 1929) warned against,
“blindly trusting in purely statistical relations with an extraordinary wide range of variations to both
sides of the average as a consequence, progress in this field came practically to a standstill”.
(Kulhawy 1993) characterised geotechnical uncertainties arising from three category of sources:
inherent variability, transformation uncertainty and measurement error.

1. Inherent Variability: Natural geological processes of erosion, deposition, weathering etc. keeps
modifying the soil properties. Soil properties keep varying both horizontally and vertically.

2. Measurement error: Sampling error or statistical error can be minimized by taking a greater
number of samples since it arises out of limited observations. Errors may also come while lab testing
or field testing.

3. Transformation uncertainty: This form of uncertainty is associated with empirical relations

which is designed after taking number of assumptions for ideal conditions.

(Phoon and Kulhawy 1999d) evaluated COV of inherent variability and measurement error and
COV of transformation uncertainty is discussed in companion paper (Phoon and Kulhawy 1999a).

(Christian 2004) listed several strategies to deal with geotechnical uncertainties:

1. Ignoring it
2. Being conservative
3. Using the observational method
4. Quantifying uncertainty

2.3 Reliability Analysis Methods

Among the popular methods of reliability analysis are First-Order Reliability Method (FORM),
Second-Qrder Reliability Method (SORM), Monte-Carlo simulation (MCS) etc. FORM includes two
models (Haldar and Mahadevan 2000): First-Order Second-Moment method (FOSM) and Advanced
First-Order Second-Moment method (AFOSM).

2.3.1 FOSM

(Cornell CA 1969) proposed FOSM formulation. For structure to be reliable, the resistance or load
capacity must be higher than applied loading. The term performance function (Z) is introduced as the
difference between resistance and applied loading (Thoft-Christensen et al. 1982).

AFOSM was introduced by (Hasofer and Lind 1974) as an improved version of FOSM. This is a
simplified method in which coordinates are so shifted that mean becomes one and variance becomes
zero. The normal variables get transformed into standard normal variables in the standardized space
of coordinates. FOSM can be easily programmed to calculate mean and coefficient of variation in
reliability analysis problems (Ayyub and McCuen 2016)

(Ruiz 1984) applied FORM in reliability analysis of vertical piles subjected to lateral loads. The
bending strength is taken as the output parameter. He concludes, “Very wide variations are found for
the reliability index (β) of piles designed in accordance with ordinarily accepted regulations”.

(Folse 1989) performed reliability analysis of a vertical pile subjected to vertical and lateral loads,
moments or combination of these using FORTRAN and design point method.

(Phoon et al. 1990a) applied stochastic finite element method and FOSM based reliability analysis to
pile head settlements. The coefficient of variation and mean is calculated using FOSM and reliability
index is calculated by carrying out reliability analysis. For simpler computation of the reliability
index, the outcomes of the reliability analysis are presented in form of concise design charts.

Other early works on the application of FOSM on pile foundation includes (Tang et al. 1990; Alberto
and Catalina 1993; Lacasse and Nadim 1996; Ayyub et al. 1997; Eloseily 1998).

(Tandjiria et al. 2000) took the case of laterally loaded piles and applied Response surface method.
Maximum bending moment and pile head displacements are taken as output parameters. Monte-carlo
simulation is carried out and the results of both the methods were compared. The reliability index is
calculated by Hasofer-Lind method. The results of both the methods are in agreement with each
other. He concluded, “Compared to Monte Carlo simulation, the response surface methods are more
efficient as they need only a small number of iterations to perform the reliability analysis”.

(Low et al.) applied reliability analysis to piles subjected to lateral loadings and concluded that
reliability analysis methods are advantageous in design than factor of safety method.

(Eloseily et al. 2002) applied reliability analysis for single pile and group piles subjected to lateral
loads in sandy soil. He concluded, “The proposed reliability-based assessment methodology is
practical for evaluating pile foundations. The methodology considers uncertainties involved in loads,
strength variables, and prediction models. The proposed methodology can be an alternate
replacement to the currently used safety margin method”.

(Honjo et al. 2002) applied FORM to vertically loaded piles. The paper proposed a method to
determine partial factors for the design of pile foundation.

(Afolayan and Opeyemi 2008) performed FORM based reliability analysis on load capacity of
pre-cast concrete pile in cohesive and cohesionless soil. The research is aimed at studying the
relationship of pile length in cohesive and cohesionless soil with safety index. He concluded, “safety
index of piling denigrates with increasing length of piles in both cohesive and cohesionless soil….
From the results, it can be concluded that concrete piling should be discouraged in cohesionless soil
and if it must be used at all, the pile length must not exceed 20 m. Also, even in cohesive soils where
piling capacity is grossly conservative, the static pile capacity equations are very expensive”.
(Chan and Low 2009a) carried reliability analysis of piles subjected to vertical loads using
conventional FORM, Response surface method and Monte-Carlo simulation. The outcomes of all the
models are in agreement to each other.

(Kim et al. 2011) performed FORM based reliability analysis axially loaded piles in cohesionless
soil.

(Kwak et al. 2010) performed reliability analysis based on FORM and Monte Carlo simulation for
steel pipe piles. He concluded, “Reliability indices and resistance factors computed by the MCS are
statistically identical to those computed by FORM”.

(Stuedlein and Reddy 2013) studied the relation between sample size of augured cast-in-place
(ACIP) piles and reliability index. He concluded, “Sample sizes greater than about 40 provided
relatively consistent estimates of the reliability index; however, its uncertainty continued to decrease
with increasing sample sizes”.

(Ng and Sritharan 2016) performed reliability analysis of H-piles in clayey soil. 19 piles were
analysed using LRFD, a closed-form FOSM based procedure. The results are in agreement with
FORM though having less computational efforts than the latter.

(JIANG et al. 2016) used response surface method to piles applied with lateral loads on sloping
grounds. Results are in agreement with that of Monte Carlo method.

(Haldar 2019) summarised the concepts of uncertainty, variability and reliability analysis methods. It
addresses reliability of pile foundation under dynamic as well as static loading. The work proposes
foundation design to incorporate the reliability analysis.

(Heidarie Golafzani et al. 2019) applied FOSM and AFOSM to assess the reliability analysis of
bearing capacity of axial pile. The study compared the results of CPT-based analysis, SPT-based
analysis and different static methods. In results, CPT-based methods outperformed SPT-based
methods and static methods performed poorly.

2.3.2 MARS

To build over the shortcomings of traditional regression models, various soft-computing methods
have been introduced in the past. Artificial Neural Networks (ANNs) have become increasingly
popular and successfully used in various fields of geotechnical engineering (Shahin and Jaksa 2008).
Most popular among the various ANN models is back-propagation (BP) algorithm (Rumelhart et al.
1986) but it’s shortcomings are long computational time and trial-and-error approach to ascertain the
optimal number of hidden neurons. To overcome these shortcomings, multivariate adaptive
regression spline (MARS) was introduced (Friedman 1991b). (Zhang and Goh 2013) gave short
glimpse on the MARS algorithm, its development and usage in various fields of geotechnical
engineering.
(Samui and Kurup 2012) performed reliability analysis of undrained shear strength of clay using
MARS and LSSVM. The performance of both the models are robust and equally likely.

(Samui 2012a) applied MARS in reliability analysis of ultimate load capacity of piles embedded in
cohesionless soil. The results of the model are compared with that of GRNN. MARS outperforms
GRNN and proves to be a robust model to be applied for reliability analysis of ultimate capacity of
piles.

(Lashkari 2013) studied application of MARS for mobilised shaft resistance piles embedded in
granular soil. The performance of the model is concluded to be excellent.

(Zhang et al. 2015) performed reliability analysis of liquefaction of soil using MARS and other
methods. MARS model outperforms other models and presents simple and easily understandable
model for reliability analysis of soil liquefaction.

2.3.3 GP

(Koza 1992) introduced pioneering, revolutionary machine learning method which outperforms other
artificial intelligence methods including Neural networks. The method is applied in wide variety of
applications across all the disciplines of the engineering.

(Alavi and Sadrossadat 2016) applied GP in Rock engineering. Linear genetic programming model is
used to examine the bearing capacity of foundations lying on rock-mass.

(Fatehnia and Amirinia 2018) applied GP to predict the load capacity of pile foundation and
compared the performance of GP with that of ANN. Both the models are applied in several studied
applications and found to be having excellent performance.

(Yong et al. 2020) developed a GP model, ANFIS model and simulated annealing–GP model to
predict the ultimate bearing capacity of the pile. 50 datasets are collected and divided into training
and testing datasets.4

2.4 Gaps in Literature Review

After careful study of different literature on the topic of slope analysis, the following gaps have been
pointed out. These are:

1. FOSM has been found to be very time consuming and requires additional efforts due
to the large number of calculation performance functions and/or its partial derivatives
with respect to the random basic variables.

2. Other applied methods also possess serious drawbacks.

3. Till now, ELM, MARS, ENN and MPMR have not been applied to reliability analysis
of load capacity of pile foundation.

4. Till now, LSSVM, GMDH, GPR, RVM, GP, ANFIS and GRNN have not been
appliedto reliability analysis of load capacity of pile foundation.
Chapter 3

Methodology

3.1 GEOTECHNICAL ANALYSIS

Soil data :

Fig:3.1

17 | P a g e
 Fig:3.2

3.1.1 Methods used for calculations : Ultimate bearing capacity as per shear criteria

● Method 1(Terzaghi Bearing Capacity)

qnu=(1+0.3 B/L)CNc+[γt.y +(Df-y) γsub](Nq-1) + (1-0.2 B/L)0.5B.γsub Nγ
For Pure Clay: Nc= 5.7
qnu= 205.495 kn/m2 eq-3.1
● Method 2(Meyerhoff's Approach)
qnu=CNc[ic .Sc. dc] +qNq[iq. Sq. dq] + 1/2 B.γ.Nγ [iγ.Sγ.dγ]
qnu= 856.955 kn/m2 eq-3.2
● Method 3 (IS Code Approach)
qnu =CNc[ic .Sc. dc] + q(Nq-1)[iq .Sq .dq] 1/2 B.γ.Nγ [iγ. Sγ. dγ].W'
qnu = 661.853 kn/m2 eq-3.3
18 | P a g e
3.1.2 Method used for calculations : Allowable Bearing Capacity for Raft
qnet(safe)=0.88.N.Sa Cw = 599.785 kn/m2 eq-3.4
Here we have taken Sa (Maximum allowable settlement) = 75mm as per IS code.

(NOTE: Average corrected SPT ‘N’ value = 17.05839)

PILE (D = 500mm)

Length of of Pile = 16

Number of Pile = 12

COLUMN (500mm X 500mm)

Raft plan = 20000mm X 15000mm

Raft thickness = 500mm

All Dimensions are in Millimetre

Fig: 3.1.1 ( Plan of proposed building)

19 | P a g e
 Fig: 3.1.2 ( sectional drawing)

Bearing Capacity Under Shear: Indicates the maximum load the soil can withstand without
experiencing shear failure.
Bearing Capacity Under Settlement: Reflects the load at which the soil's settlement exceeds
permissible limits.

20 | P a g e
Since the settlement capacity (599.78 kN/m²) is less than the shear capacity (661.85 kN/m²), the
soil will fail due to excessive settlement before reaching its shear capacity. This means the structure
might experience uneven or excessive sinking, compromising its stability and usability, even though
the soil's shear strength has not yet been reached.

Now, our assumed maximum load on a column is 1200 KN. (1200/0.5^2 = 4800 KN/M^2)
So, when the shear capacity of raft is fully utilised then it undergoes settlement and we do our
assessment to introduce pile as a settlement reducer as follow-
now we have total 12 columns so, total load on raft is 12 x 1200 = 14400 KN
Total pressure = 14400/(15x20)= 48 KN/m2 (external pressure)

Here the pile is provided as a settlement reducer.

3.1.3 Assessment carried out for pile raft foundation:

The use of pile to reduce the raft settlements and differential settlements can lead to considerable
economy without compromising the safety of and performance of the foundation. Such a foundation
makes the use of both pile and raft and is referred to here as a pile enhanced raft or a piled raft.

Three stage design process is proposed :

1. preliminary stage to assess feasibility.
2. To assess the location where the piles are required.
3. to obtain the detailed design information.

An assessment of the maximum column loadings that may be supported by the raft without a pile
below the column.

A typical column on a raft is shown in Fig. (3.1.1). There are at least four circumstances in which
a pile may be needed below the column:

(a) if the maximum moment in the raft below the column exceeds the allowable value for the raft
(b) if the maximum shear in the raft below the column exceeds the allowable value for the raft
(c) if the maximum contact pressure below the raft exceeds the allowable design value for the soil
(d) if the local settlement below the column exceeds the allowable value.

21 | P a g e
Table:3.1
DESCRIPTION VALUE UNIT

p Design Load 1200 KN

a Characteristics Length of Raft 4.3 M

a= 〖t[E_r (1-γ_s^2 ")/6" E_s (1-γ_r^2)]" " 〗^(1/3)

q Contact Pressure Below Raft 599.78 KN/M2

c column radius 0.25 M

cq Shear factor 1.6

qd Design allowable bearing pressure 599.78 KN/M

qu Ultimate bearing capacity of raft 661.853 KN/M

fp Factor of Safety for Piles 2

fs Factor of Safety for Contact pr. 3

vs Poisson's Ratio of Soil 0.35

vr Poisson's Ratio of Raft 0.2

Es Young's Modulus of Soil 6.9 Mpa

Er Young's Modulus of Raft 30000 Mpa

1. Maximum Moment Criterion

Mx= Ax.P Ax = A-0.0928ln⁡(c/a) 0.539007591

My= By.P By = B-0.0928ln⁡(c/a) 0.599007591

Mx= 646.809 KN-M

My= 718.809 KN-M

Md= Design Moment Capacity of Raft max of Mx and My = 718.809 KN-M

Pc1=Md/(larger of Ax and By) = 1333.5788 KN

22 | P a g e
2. Maximum Shear Criterion

Vmax= Maximum Shear below a column

Pc2= Maximum Column Load

3. Maximum Contact Pressure Criterion

q(max )=Maximum contact Pressure on the base of raft= 661.853 KN

q ̅=contact Pressure factor= 1

4. Local Settlement Criterion

S=Settlement below column

ω=Settlement Factor = 0.64
Sa=Allowable Settlement =20MM

23 | P a g e
 Table: 3.2
Pc1 Pc2 Pc3 Pc4

in ( KN ) 1333.5788 1200 4079.2206 1056.623

Assessment of Pile Requirements for a Column Location

3.1.4 Design of Pile:

Table:3.3
Description Value Unit

D Diameter of 0.5
pile

Area of base 0.1962

29.43 3.94

15 Degree

27.47 4.924 190 180.603

17 Degree

26.49 5.416 285 302.923

18 Degree

24 | P a g e
 43 Degree

F.O.S 3

End bearing resistance= 302.924 KN

Table:3.4

Depth Height Surface Area

8 8 12.57 345.023
9 1 1.57 43.1279
10 1 1.57 43.1279
11 1 1.57 41.5893
12 1 1.57 41.5893
13 1 1.57 41.5893
14 1 1.57 41.5893
15 1 1.57 41.5893
16 1 1.57 41.5893
pile length 16m Total 680.814 KN
=

Total skin friction = 983.738 KN

Pile Capacity 327.913 KN

25 | P a g e
 Table:3.5 (Training data given below : 10 out of 100 )

N- C1 N-C2 N-C3 N-Qus

0.094736842 0.505263157 0.536842105263158 0.25884016

0.621052631 0.505263157 0.431578947368421 0.820988128068601

0.505263157 0.694736842 0.263157894736842 0.695597914404833

0.705263157 0.578947368 0.71578947368421 0.844256761348575

0.610526315 0.410526315 0.4 0.821019791294788

1 0.684210526 0.105263157894737 0.950544454705098

0.431578947 0.042105263 1 0.504914290378516

0.905263157 0.536842105 0.989473684210526 0.878360934426113

0.926315789 0.505263157 0.231578947368421 0.975355950642867

0.315789473 0.642105263157895 0.8631578947 0.472682085612768

3.2 MULTIVARIATE ADAPTIVE REGRESSION SPLINES

3.2.1. Introduction to MARS

What is MARS ?

In statistics, Multivariate adaptive regression splines (MARS) is a form o f regression analysis.

It is a non-parametric regression technique and can be seen as an extension of linear models that
automatically models non linearities and interactions between variables.

26 | P a g e
 Why use MARS Model

• MARS is ideal for users who prefer results in a form similar to traditional regression
while capturing essential nonlinearities and interactions.

• MARS builds its model by piecing together a series of straight lines with each allowed its
own slope as shown in Fig. 3.

• This permits MARS to trace out any pattern detected in the data.

Fig. 3.2.1

3.2.2 MARS Mechanics

Fig.3.2.3 If we are trying to create a line with the smallest square error it may work something like
this.

27 | P a g e
 Fig.3.2.4 This follows some error,
The basic error can be seen in the chart below

Fig.3.2.5.This follows some error,

The basic error can be seen in the chart below.

Fig.3.2.6 We then create a pivot point and then

add a second and third linear line connecting to
this line to create a ‘knot’ in the model.

Fig.3.2.7 The resulting model will have a lower

overall error term and can match the pattern of the
data in a much more effective manner.

28 | P a g e
3.2.3 Details of MARS
MARS is a form of regression analysis introduced by Jerome H. Friedman in 1991.

To develop a model of input and output, MARS uses a basis function.

The relationship of input(x) and output (y) is given below:

Where, Bm(x) is the basis function, M is the number of functions, a0 is constant and Cm is the
coefficient of the mth basis function.

Two steps are adopted to develop the MARS model.

Forward step In this step, basis functions are used to define input and output value.

Backward step Basis functions have been removed as per generalized cross validation (GCV)
value.

Where N =the number of data and C (B) = a complexity penalty. yi is the output xi is the input
parameters.

The value of C (B) is determined by

using the following equation.

Where d = penalty for each basis

function (BF) included into the model. Graph: 3.1

3.2.4 Reliability analysis:supply vs. demand

29 | P a g e
Failure takes place when demand exceeds supply.
For an engineering system:
Available resistance is the supply, R
Load is the demand, Q
Margin of safety, M=R-Q
The reliability of a system can be defined as the probability that R>Q .

3.2.5. First Order Second Moment Method (FOSM)

Reliability index =

Where σd and Qus are defined as standard deviations and mean values of the predicted value of
ultimate pile capacity respectively.

Drawbacks of FOSM:

It is a time consuming method for implicit performance function.

3.2.6 Methodologies

•Data has been collected from the location of existing premises of M/s CPCL at Manali,Chennai.
•To develop the MARS Model, the data have been branched into the following 2 groups. Training
dataset, Testing database

Datasets : Table:3.5

Processing of Dataset

● Training Dataset: we have used 100 dataset out of 70 as training dataset. This is used to
develop the model.
• Testing Dataset: This study uses the remaining 30 dataset as testing dataset. This is
used to verify the developed model.
• Input variables are normalized between 0 and 1.

Developed Equation is given below:

30 | P a g e
3.3 GENETIC PROGRAMING

3.3.1 Introduction

● GP (Koza, 1992) is an extension of more elementary method GA (Holland, 1975)

● GP determines the best model by artificial evolution.

The following steps are involved for developing

● Create population of random equation
● Determine the fitness value of each equation
● Select parents from equations through tournament
Generate offspring's from the parents through the process of reproduction, crossover and
mutation

Fig. 3.3.1

3.3.2 Crossover

31 | P a g e
•Crossover is one of the genetic operators used to recombine the population
•Crossover point has been randomly chosen, portions of the parent's chromosome (strings) combined
to produce the new offspring.

Fig:3.3.2

3.3.3 Mutation
The mutation operator introduces new genetic structures in the population by randomly changing
some of its building blocks, helping the algorithm escape local minima traps

32 | P a g e
3.3.4 Methodology
•Data has been collected from the location of existing premises of M/s CPCL at Manali,Chennai.
•To develop the GP, the data have been branched into the following 2 groups

Datasets Used

Table: 3.6
Predicted Training Predicted Testing

0.2571 0.9048

0.8224 0.6748

0.6939 0.5725

0.844 0.6611

0.8218 0.0072

0.9448 0.4908

0.507 0.7926

0.8817 0.8681

0.9765 0.8681

0.4718 0.4952

Training dataset: 70 data has been used as training dataset. This is used to develop the model.
Testing dataset: 30 data has been used as testing dataset. This is used to verify the developed model.
The data have been normalized between 0 and 1

33 | P a g e
 Table:3.7 (Calculated Parameters)

Parameters Training Data GP Testing Data Ideal Value

GP

WMAPE -0.0340031221157 -0.04211258257 0.086

546 31096

NS 0.7677971587963 0.99846824658 1
8563157 4875

RMSE 19.15577 1.01512657594 0.019

04

VAF 99.99999 100 87.88

0.767797 0.99846824658 1
4875

0.58918 0.99728997472 1
7087

PI -17.5666 0.98216339597 87.88

9258

RSR 0.481874 0.03913762148 1.08

01647

Bias Factor 1.038452 0.42756003944 0

8553

NMBE 79.33171 0.51484958570 8.13

4011

34 | P a g e
 Chapter 4

RESULTS & DISCUSSIONS

4.1 MARS Model

1.1 Equation developed by Mars Model is given below:

Table: 4.1
BFm BFm Equation Cm

BF1 max(0,x1-0.452631578947) 0.548677

where
x1= c1 BF2 max(0,0.452631578947-x1) 1.330649
x3=c3 BF3 max(0,x3-0.452631578947) 0.303422
BF = Basis Function
Cm = Coefficient of BF4 max(0,0.4526315789473-x) 0.269837
Basis Function

This output graph, showing a high correlation between predicted and actual values, indicates that
the model has successfully learned the relationship between the inputs and the target variable.
The model was able to generate accurate predictions, as evidenced by the nearly perfect linear fit
and correlation coefficient of 0.99979.

35 | P a g e
 Fig: 4.1(Output of MARS model - training data)

36 | P a g e
 Fig:4.2(Output of MARS model - testing data)

Table:4.2 (Reliability index given below):

Reliability Index Training Data Testing Data

3.6026 3.6907

3.5257 3.7316

Graph showing the Reliability Performance of Pile Foundation.

37 | P a g e
 Fig: 4.3

Table:4.3 (Probability of failure by MARS Model):

Training Data Testing Data

4.2 GENETIC PROGRAMING

•The best equation have been produced by the GP for the determination of Qu

4.2.1 Equation developed by GP Model is given below:

where
x1 = C1
x2 = C2
x3= C3
x4 = C4

38 | P a g e
The two plots depict the relationship between actual and predicted values for the reliability of
pile foundations, likely obtained from a genetic programming model.

Both plots exhibit a strong linear correlation, suggesting that the model is performing well in
capturing the underlying trends in the data.

39 | P a g e
 Fig: 4.4

Graph showing the Reliability Performance of Pile Foundation.

Graph no: 4.5

Table:4.4 (Reliability index by GP model given below):

Reliability Index Training Data Testing Data

3.6026 3.6907

3.5239 3.7141

40 | P a g e
Table:4.5 (Probability of failure byGP Model):

Training Data Testing Data

41 | P a g e
Chapter 5

CONCLUSIONS

The primary objectives of this project were successfully achieved, culminating in significant
contributions to the design and reliability analysis of pile raft foundations. By leveraging the
Standard Penetration Test (SPT) dataset, a robust pile raft foundation design was developed,
effectively addressing key geotechnical challenges and providing a stable foundation solution
tailored to the project requirements.

The reliability analysis was conducted using two advanced models: Multivariate Adaptive
Regression Splines (MARS) and Genetic Programming (GP) models. A comparative evaluation of
these methods revealed that the GP model outperformed the MARS model in terms of predictive
accuracy and reliability assessment. This finding underscores the suitability of GP models for
complex geotechnical reliability analyses, offering enhanced precision in predicting performance
under varying conditions.

1.Design of Pile Raft Foundation using SPT data has been performed.

2.The reliability index (β) value for the actual datasets based on MARS and GP lies between 3.00
and 4.00.

3. A comparative evaluation of these methods revealed that the GP model outperformed the MARS
model in terms of predictive accuracy and reliability assessment.

4. Developed equations can be used for practical purposes.

In addition, the project quantified the probability of failure for the designed foundation system,
providing critical insights into its safety and performance margins. This probabilistic assessment
highlights the importance of incorporating advanced statistical and machine learning techniques into
geotechnical engineering practices to ensure the reliability and robustness of foundation designs.

Chapter 6
42 | P a g e
 REFERENCES

1. Piled raft foundations: Design and applications by Harry G. Poulos.

2. Civil Engineering Beyond Limits ( www.acapublishing.com/journals/1/cebel) .
3. Introduction to Design Engineering By NPTEL IIT Madras.
4. home: www.acapublishing.com/journals/1/cebel.
5. https://ndcbe.github.io/cbe67701-uncertainty-quantification/08.01-First-Order-Secon
d-Moment-Method.html .
6. Pile design as per shear design criteria IS Code 6403 - 1981.
7. Covers Design and construction of Shallow Foundation - IS 1080:1985.
8. Covers Design and construction of Foundation in soil.
9. Reliability analysis of allowable pressure on shallow foundation using response
surface method : G.L. Sivakumar Babu *, Amit Srivastava
10. Testing Multivariate Adaptive Regression Splines (MARS) as a Method of Land
Cover Classification of TERRA-ASTER Satellite Images Elia Quirós, Ángel M.
Felicísimo and Aurora Cuartero.
11. Performance of Multivariate Adaptive Regression Splines (MARS) in predicting
runoff in mid-Himalayan micro watersheds with limited data V. N. SHARDA1, S. O.
PRASHER2, R. M. PATEL2, P. R. OJASVI1 & CHANDRA PRAKASH1.
12. GENETIC PROGRAMMING-BASED APPROACH FOR STRUCTURAL
OPTIMIZATION By Chee Kiong Soh1 and Yaowen Yang2.
13. Handbook of Research on Advanced Hybrid Intelligent Techniques and Applications
[Siddhartha Bhattacharyya, Pinaki Banerjee, Dipankar Majumdar, Paramartha Dutta]
14.GENETIC PROGRAMMING-BASED APPROACH FOR STRUCTURAL OPTIMIZATION
by Chee kiong.

43 | P a g e