Application of Physics-Informed Neural Networks for Forward and Inverse

Analysis of Pile-Soil Interaction

M. Vahaba , B. Shahbodagha , E. Haghighatb , N. Khalilia

a School of Civil and Environmental Engineering, The University of New South Wales, Sydney 2052, Australia
b Massachusetts Institute of Technology, Cambridge, MA, USA

Abstract
arXiv:2212.08306v1 [cs.CE] 16 Dec 2022

The application of the Physics-Informed Neural Networks (PINNs) to forward and inverse analysis of pile-soil
interaction problems is presented. The main challenge encountered in the Artificial Neural Network (ANN)
modelling of pile-soil interaction is the presence of abrupt changes in material properties, which results in
large discontinuities in the gradient of the displacement solution. Therefore, a domain-decomposition multi-
network model is proposed to deal with the discontinuities in the strain fields at common boundaries of
pile-soil regions and soil layers. The application of the model to the analysis and parametric study of single
piles embedded in both homogeneous and layered formations is demonstrated under axisymmetric and plane
strain conditions. The performance of the model in parameter identification (inverse analysis) of pile-soil
interaction is particularly investigated. It is shown that by using PINNs, the localized data acquired along
the pile length –possibly obtained via fiber optic strain sensing–can be successfully used for the inversion of
soil parameters in layered formations.
Keywords: Physics-Informed Neural Networks (PINNs); Deep learning; Pile-soil interaction; SciANN

1. Introduction

1.1. Background
The Finite Element Method (FEM) is ubiquitously used for the computational analysis of various geotech-
nical engineering problems [1–4]. Consistently, in many classical studies, FEM has been the natural choice
for the design of inversion or back-analysis algorithms for inferring important model parameters, like me-
chanical properties of soils [5]. The state-of-the-art techniques in this context mainly rely on minimization of
the deviations between the simulation responses (due to an adjustable set of model parameters) with respect
to field and laboratory measurements. For this purpose, a range of direct or gradient-based optimization
techniques are extended which exploit successive numerical solutions iteratively to yield optimal precision for
the sought-after parameters [6]. Nonetheless, both approaches can be prohibitively computationally expen-
sive in particular for large-scale problems. Further challenges encountered during numerical back analysis in
engineering applications have been due to non-uniqueness, material model limitations, and disparate data
sources [7, 8].
Deep Learning (DL) has proven to be a rigorous approach for the forward and back analysis of geotech-
nical engineering problems [9–12]. In the analysis of piles, which is the focus of this study, DL has been

∗ Corresponding author:
Email address: [email protected] (B. Shahbodagh)

Preprint submitted to Elsevier December 19, 2022

applied for the prediction of the shaft and tip resistance of concrete piles by Momeni et al. [13], for esti-
mation of the uplift resistance of screw piles by Mosallanezhad and Moayedi [14], and for evaluation of the
lateral load bearing capacity of piles by Das and Basudhar [15] and Armaghan et al. [16]. It is noteworthy
that the enumerated studies emphasize on forward prediction of mechanical response of piles itself. Based
on the Universal approximation theorem (a.k.a., Cybenko theorem [17]), neural networks comprising of at
least a single hidden layer can uniformly reconstruct any function with arbitrary continuous non-linearity.
Evidently, the composite system of soil-piles involves a strong material discontinuity, which in turn leads to
a discontinuous strain field (i.e., the gradient of displacement). This study aims to develop a DL framework
capable of handling the discontinuities induced by material interfaces within soil-pile systems.

1.2. Physics-Informed Deep Learning

Early Artificial Intelligence (AI) approaches were developed in the mid-20th century to tackle the intel-
lectually difficult and complex-to-articulate problems, which pertain to straight-forward solution algorithms
within relatively sterile environments [18]. DL aims to mitigate major sources of difficulty with conventional
real-world Machine Learning (ML) practice, in terms of data sensitivity, noises, and other representation
learning issues, through a nested hierarchy of simpler representations/concepts [19]. Nowadays DL algo-
rithms are predominantly employed in an increasing number of areas in engineering and science, including
geotechnics [20], structural engineering [21, 22], engineering mechanics [23], reservoir engineering [24], ma-
terial science [25], physical modelling [26], and earth sciences [27], to name a few.
Most conventional DL algorithms rely on the availability of data for training to enable the development
of reliable predictive models. However, in many engineering circumstances, the data acquisition costs could
be prohibitive. Partial information threatens the robustness of machine learning to the extent that draws
decision-making cumbersome, if not impossible. In a diversity of engineering and scientific applications,
there exists a vast prior physical knowledge that can be exercised as a regularization agent rendering the
admissible solution space into a manageable size [28]. These approaches could be considered as a subset of
Reinforcement Learning (RL) [29] in a sense that a range of incentives are optimized rather than relying on
the training dataset alone [30, 31]. Concurrent efforts in recent years have been made to incorporate prior
physical information in DL by Owhadi [32], Han et al. [30], Bar-Sinai et al. [33], Rudy et al. [34], and
Raissi et al. [35].
Physics-Informed Neural Networks (PINNs) are a class of deep learning that incorporate a series of
physical laws, frequently described in the form of partial differential equations (PDEs), to steer the learning
towards the solution for sparse training data-sets, which could not be plausible with classic DL algorithms.
The prosperity of PINNs is attributed to substantial algorithmic advances (e.g., graph-based automated
differentiation [36]) and major software developments (e.g., TensorFlow [37], Keras [38]). PINNs have been
utilized with great success in a broad range of engineering disciplines, including solid mechanics [31, 39–42],
fluid mechanics [43–45], and thermo-mechanics [46]. Conventional data-driven inversion models frequently
fail in generalization, owing to extrapolation or observational biases [28, 35]. A beneficial remedy can be
offered by physics-informed deep learning, which provides the network model along with the laws of physics
governing the system. This can rectify the issues associated with the missing ingredients induced by the
sparsity of data, uncertainties, or other less understood factors. Inverse physics-informed solutions have
been hitherto investigated in nano optics by Chen et al. [47], in solid mechanics by Haghighat et al. [31],
for conservation laws by Jagtap et al. [48], and for flow problems by Lou et al. [49].

2
1.3. Our contribution
In this study, we emphasize on a novel application of PINNs to the solution and inverse analysis of soil-
pile interaction problems. For this sake, we use the longitudinal strain profile along the entire length of piles,
which could be obtained through the optical fiber strain-sensing technique in practice (e.g., see Mohamad
et al. [50, 51]). Differential strains (i.e., at either side of piles) may also be extracted by the installation of a
group of fibers so as to enable the monitoring of both axial movements and bending [50]. Here, PINNs are
first employed to construct efficient neural networks for the solution of soil-pile interactions in the absence of
any data. Salient features of these systems, namely strain discontinuity, contact constraints, compatibility of
stresses, and inhomogeneities, are incorporated through introducing a domain-decomposition multi-network
model [45, 46, 48, 52]. The localized data acquired along the pile length, in turn, is utilized for the inversion
of key mechanical properties of soil in layered formations.
This paper is organized as follows: In section 2, the equations governing the pile-soil interactions are
explained in detail. Section 3 is devoted to a brief introduction to the fundamentals of PINNs and its
application to the solution of pile-soil systems. The forward solution of single piles –in both cylindrical and
Cartesian coordinate systems –is explored in section 4. In addition, the inverse analysis of soil mechanical
properties is conducted for homogeneous and layered formations. Concluding remarks are presented in
section 5.

2. Problem Statement and Governing Equations

Fig. 1 depicts the pile-soil system considered in this study. The system is modelled in both two-
dimensional (2D) and three-dimensional (3D) settings, representing a sheet-pile wall and a single cylindrical
pile embedded in soil media, respectively. The Cartesian coordinate system is used for the 2D plane strain
analysis, whereas the cylindrical coordinate system is employed for the 3D analysis of pile-soil interaction.
In the figure, ΩP and ΩSk (k = 1, ..., N ) denote the regions of the space occupied by the pile and the k th
soil layer, respectively, where N is the number of the soil layers. The boundaries of the pile and soil regions
are designated by Γp and ΓSk , respectively. The pile is assumed to be continuously bonded to and fully
embedded in the soil medium. We leverage a generic representation of the pile deformation which enables

S1 ,  S 1 Layer # 1

P
 S2 ,  S2 P Layer # 2

 Sk ,  Sk Layer # k
2a

SN 1 ,  SN-1 Layer # N-1

Semi-infinite
Medium # N
 S N ,  SN

Fig. 1: Schematic representation of a single pile in a layered soil medium.

3
the model to capture the non-uniform deformation along the cross-section of the pile, essential for the proper
simulation of short pile response and interactions at low pile-soil stiffness ratios. The equilibrium equations
governing the pile-soil system are expressed in the cylindrical coordinate system as
α α
1 ∂ α 1 ∂σrθ ∂σrz σα
(rσrr )+ + − θθ + frα = 0 ,
r ∂r r ∂θ ∂z r

α α
1 ∂ α 1 ∂σθθ ∂σθz σα
(rσrθ )+ + + rθ + fθα = 0 , on Ωα (α = P, Sk ) (1)
r ∂r r ∂θ ∂z r

α α
1 ∂ α 1 ∂σθz ∂σzz
(rσrz )+ + + fzα = 0 ,
r ∂r r ∂θ ∂z
α
where σij (i, j = r, θ, z) is the Cauchy stress tensor , fiα is the body-force density vector, and P and
Sk (k = 1, ..., N ) denote the corresponding quantities of the pile region and the soil medium, respectively.
The strain-displacement relations are given by

∂uα ∂uα ∂uα

 
r 1 z
εα
rr = , εα
θθ = θ
+ uα
r , εα
zz = ,
∂r r ∂θ ∂z

1 ∂uα uα 1 ∂uα ∂uα

    
1 r ∂ 1 z
εα
rθ = +r θ
, εα
zθ = + θ
, on Ωα (α = P, Sk ) (2)
2 r ∂θ ∂r r 2 r ∂θ ∂z

∂uα ∂uα
 
1 r z
εα
rz = + ,
2 ∂z ∂r
where εα α
ij (i, j = r, θ, z) is the strain tensor and ui is the displacement vector. Using the indicial notation,
Eq.s (1) and (2) can be written in the Cartesian coordinate system (i, j = x, y, z) as
α
σji,j + fiα = 0 ,
1 α on Ωα (α = P, Sk ) (3)
εα ui,j + uα


ij = j,i ,
2
The elastic constitutive relation employed for the pile-soil system is given by
α
σij = λ α εα α
kk δij + 2µα εij , on Ωα (α = P, Sk ) (4)

where λα and µα are the Lamé constants and δij is the Kronecker delta. The Lamé constants can be
expressed in terms of the Young’s Modulus Eα and Poisson’s ratio να as:
Eα να Eα
λα = , µα = . (5)
(1 + να )(1 − 2να ) 2(1 + να )
These field equations are accompanied by the boundary conditions at the interfaces of pile and soil and
soil layers, i.e.
 
α β
σji (x) − σji (x) nα
j =0 ,
∀x ∈ (Γα ∩ Γβ ), where α 6= β (6)
β
uα
i (x) − ui (x) = 0 ,
in which nα
j is the unit outward normal vector of Γα (α, β = P, Sk ). On the top surface,
S1
σji (r > a, θ, z = 0) nSj 1 = 0 ,
P
(7)
σji (0 ≤ r ≤ a, θ, z = 0) nP P
j = ti ,

4
 1 2 LL
Σ Σ ΣΣ

Σ Σ ΣΣ

Input Σ Σ ΣΣ Output
Output

x Σ Σ ΣΣ u

≡ z0 Σ Σ ΣΣ ≡≡ zzLL
Σ Σ ΣΣ

l l l l-1 l Σ Σ ΣΣ
Σ = σ (W z + b )

Fig. 2: Standard PINN architecture, defining the mapping u : x 7→ Nu (x; W, b).

where tPi is the surface traction
applied to the pile head and a is the radius of the pile. Under vertical
2
loading condition, tP
i = Q/ πa 0 0 , where Q is the applied vertical load. For a single pile embedded
in a half-space, the regularity conditions at infinity are specified as
p
Sk
σij (r, θ, z) → 0 , as r2 + z 2 → ∞ (8)

For the case with underlying bedrock, the boundary condition at the bedrock level can be expressed as

uSi N (r, θ, z = H) = 0 , (9)

where H is the depth to bedrock.

3. PINN Forward and Inverse Solvers

In this section, we briefly review the construction and training process of Physics-Informed Neural
Network (PINN) solvers. Next, we implement the PINNs for the analysis of piles in homogeneous and
layered formations.

3.1. Physics-Informed Neural Networks

Suppose Nu (x; W, b) is an L-layer neural network with x and u being the input and output vectors,
and W and b being weights and biases of the transformation, respectively. As depicted in Fig. 2, a feed-
forward network is employed to approximate the solution variable u for given inputs u through the following
transformations:
u = ΣL ◦ ΣL−1 ◦ · · · ◦ Σ1 (x) , (10)
where
Σl (ẑ l−1 ) := ẑ l = σ l (Wl · ẑ l−1 + bl ) , l = 1, ..., L (11)
In the above relation, z0 ≡ x and zL ≡ u are the inputs and outputs of the model, with σ l s being
the nonlinear activation functions and ◦ being the compositional construction of the network. Notably, the
hyperbolic-tangent is frequently employed as the activation function.
Consider a quasi-static partial differential operator P acting on the solution variable u, as Pu(x) = f (x).
It is noteworthy that u is the vector form of the displacement field described as u = (ur , uθ , uz ) or u =
5
(ux , uy , uz ) in cylindrical or Cartesian coordinate systems, respectively. The boundary conditions associated
with the solution field can also be expressed versus partial differential operator B as Bu(∂x) = g(∂x),
for x ∈ Rd , where d is the spatial dimension of the problem. Suppose the network parameters are all
SL
collected in θ = i=0 (Wi , bi ). In the context of PINNs, the network parameters are determined through
the minimization of a loss function constructed versus the equations governing the problem on the domain
and boundary conditions as
X
L(x; θ) = λi Li = λ1 kPu − f kΩ + λ2 kBu − gk∂Ω + . . . , (12)

with L being a loss function, and λi s being a selection of weights associated with each loss term which are
determined adoptively throughout the solution process.
The mean squared error norm is elaborated for the evaluation of the loss function, i.e., k◦k = MSE(◦).
In this fashion, the network parameters are determined by means of an optimization problem represented
by

θ ∗ = arg min L(X; θ) , (13)

θ∈RD

in which D is the total number of trainable parameters, with X ∈ Rn×d being the set of n collocation
points used for the optimization of the loss function. In this study, the construction and training of PINNs
is performed by taking advantage of the open-source python API SciANN [39], which is implemented on
reputed deep-learning packages TensorFlow [37] and Keras [38].

3.2. Domain-decomposed PINNs for Layered Soils

The primary partial differential equation that governs the deformation of pile-soil system is the equilib-
rium equation described in section 2. Provided that the material properties of the pile as well as each layer of
the surrounding soil could be significantly different, we are dealing with inherent discontinuities within the
strain field. Indeed, the variations in material stiffness, that could differ by several orders of magnitude, is
responsible for the discontinuity in the derivatives of the displacement field. As such, a domain-decomposed
PINN solver is a reasonable choice [45, 46, 48, 52]. We, therefore, introduce a series of domain factors to
distinguish one material domain from another as

1 , x ∈ Ωα
Π̄α (x) = 0 , x ∈ Γα (14)
−1 , x∈
/ (Ωα ∪ Γα )


in which α = P, S for homogeneous domains, and α = P, Sk (k = 1, ..., N ) in case of layered formations. In

line with the definition of domain factors, it is required to define a distinct neural network for each domain
involved. This facilitates the presence of a weak discontinuity in the displacement field across the material
interfaces. Hence, the displacement field can be represented over the entire solution domain as

N
X
ui ' NuPi (x).ΠP (x) + NuSik (x).ΠSk (x) , (15)
k=1

where Πα = (Π̄α + 1)/2.

As mentioned in section 2, in order to avoid non-physical overlap/separation of the displacement fields
associated with different solution domains, displacement constraints need to be imposed along the material
interfaces. For the sake of simplicity, it is also stipulated that no-slip may occur along any material interface.

6
As such, the compatibility constraints corresponding to the displacement field across the whole domain can
be expressed by
 α
 Nui (x) − Nuβi (x) = 0 ,
(16)
∀x ∈ (Γα ∩ Γβ ) , where α 6= β


where i = (r, θ, z) or (x, y, z) for the Cylindrical and Cartesian coordinates, respectively.

4. PINNs Solution and Parametric Study

In this section, application of the proposed framework is demonstrated for the solution and parametric
study of piles in both homogeneous and layered formations. For this sake, the governing equations presented
in section 2 are employed for the solution of pile-soil systems under axisymmetric and plane strain conditions.
The performance of the framework for parameter identification in layered soils is explored in the final
example.

4.1. Forward Solution of Cylindrical Piles in Homogeneous Soils

In this example, the PINNs solution of an axisymmetric cylindrical pile under vertical loading is in-
vestigated. The general form of the PDEs governing the deformation of the pile-soil system is given by
the Equilibrium Eq.s (1). Considering the axisymmetricity of the problem, all derivatives with respect to
θ and the terms related to shear stresses exerted in direction θ are vanished. To avoid the singularity of
the governing equations, the equilibrium equations are multiplied by r. Thus, the simplified form of the
equilibrium equations is expressed as

∂ ∂σ α
α
(rσrr ) + r rz − σθθ
α α α
≡ Prr u =0 ,
∂r ∂z
(17)
α
∂ α ∂σzz
α α
(rσrz )+r ≡ Pzz u =0 .
∂r ∂z

Q
ΓS

r
z ΩP
l0
d0
ΩS ΓP EP , νP lT

ES , νS

Fig. 3: The cylindrical pile embedded in homogeneous soil; problem definition and boundary conditions.

7
 Fig. 4: The network training history for the cylindrical pile in homogeneous domains.

The strain field is described by Eq. (2), except that all terms involving uθ or derivatives with respect to
θ are vanished. In addition, the singularities due to the presence of the term 1/r at the origin (i.e., r = 0)
are relieved through the addition of an small amount ¯, as 1/r ' 1/(r + ¯) (here, ¯ = 0.001). Therefore, it

0.0 0.00 0.0 0.00 0.0 0.00

0.2 0.05 0.2 0.05 0.2

0.05
0.4 0.10 0.4 0.4
0.10
0.6 0.15 0.6 0.6 0.10
0.15
0.8 0.20 0.8 0.8 0.15
1.0 0.25 1.0 0.20 1.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) PINN solution for η = 25 (b) PINN solution for η = 50 (c) PINN solution for η = 100
0 0 0

0.05 0.04 0.04

0.08
0.10 0.08

0.12
0.15
0.12
0.16
0.20
0.16
0.20
0.25

(d) Reference solution for η = 25 (e) Reference solution for η = 50 (f) Reference solution for η = 100

Fig. 5: Contours of the normalized vertical displacement uz /`T (10−2 ) for the cylindrical pile in homogeneous domains.

8
follows that

∂uαr uα
r ∂uαz
εα
rr = , εα
θθ = , εα
zz = ,
∂r (r + ¯) ∂z
(18)
∂uα ∂uα
 
1 r z
εα
rz = + ,
2 ∂z ∂r
where α = P, Sk . The above equations are accompanied by the constitutive relation given by Eq. (4).
The compatibility constraint required for the interface formed at the intersection of pile and soil is
expressed by  P S
 ur (x) − ur (x) = 0 ,




uP S
z (x) − uz (x) = 0 , (19)




∀x ∈ (ΓP ∩ ΓS )


As depicted in Fig. 3, suppose a cylindrical pile with the slenderness ratio of `0 /d0 = 5, which is subject
to the vertical loading of Q = 100 kN. The domain consists of a homogeneous soil layer that is extended
for the normalized radius of rT /d0 = 10 and length of `T /`0 = 2. The material properties for the soil
are assumed as: Young’s Modulus of Elasticity, ES = 100 MPa; Poisson’s ratio, νS = 0.25. The material
properties of the pile are: Young’s Modulus of Elasticity, EP = 2.5, 5, 10 GPa; Poisson’s ratio, νP = 0.25.
In this fashion, the problem is studied for the stiffness ratios η = EP /ES = 25, 50 100.
In order to perform the PINNs solution, as explained in section 3, we need to use multiple neural networks
proportional to the number of materials existing throughout the entire domain. As such, two distinct neural
networks are introduced as per component of the displacement field as

uP P
r ' Nur (r, z) , uP P
z ' Nuz (r, z) ,
(20)
uSr ' NuSr (r, z) , uSz ' NuSz (r, z) .

4 hidden layers with 20 neurons in each layer are considered in all the cases. Hyperbolic-tangent is also
used as the activation function. The physics-informed loss terms of the total cost function are expressed as

LT = LΩ + LΓB.C. + LΓCont ,

P P S S
LΩ = λ1 Prr u on Ω + λ2 Prr u
P P P
S S on ΩS
+ λ3 Pzz u on Ω + λ4 Pzz u on Ω ,
P S

(21)
P P P
S S S

LΓB.C. = λ5 Brr u − grr on ΓP \ΓS
+ λ6 Brr u − grr on ΓS \ΓP
P P P
S S S

+ λ7 Bzz u − gzz on Γ \Γ + λ8 Bzz u − gzz on Γ \Γ ,

P S S P

LΓCont = λ9 uP − uS on Γ + λ10 tP − tS on Γ

,
P ∩ΓS P ∩ΓS

where P α , B α and g α are differential operators associated with the equilibrium equations, boundary condi-
tions, and preassigned boundary values of the problem, respectively, in the directions r and z.

9
 Q

l1
x
ΩP
l0 y
d0
l2
ES , νS ΩS
1 1 1

ES , ν S
2 2
ΩS 2

Fig. 6: The sheet-pile wall in layered formation; problem definition and boundary conditions.

The training is performed by using 3000 sampling points over each of domains ΩP and ΩS (i.e., 6000 in
total), whereas at least 50% of the sampling points are clustered over the boundaries ΓP and ΓS . The use
of non-uniform sampling grid ensures that the complex boundary conditions introduced along the pile-soil
interface as well as the external boundaries are properly satisfied. Furthermore, Neural Tangent Kernel
(NTK) adaptive weighting is employed to determine λs, which guarantees all loss terms are calibrated
proportionally throughout the training process [53]. Here, the Adam optimization scheme is adopted for the
training with the learning rate of 0.003.
In Fig. 4, the evolution of normalized loss versus epochs and time is illustrated. Evidently, in all
cases, the designated architecture converges rapidly to the relative error of 10−6 within less than 1000
epochs. Meanwhile, the training process is accomplished relatively fast (total duration ≈ 3000 s). Contours
of the normalized vertical displacement field uz /`T is shown for all the cases in Fig. 5. Furthermore, a
reference solution obtained using the FEM software COMSOL Multiphysics [54] is presented for the sake of
comparison. An excellent agreement is observed between the PINNs solution and the results of the FEM
analysis. This demonstrates the validity of the proposed framework for the analysis of pile-soil systems
under the axisymmetric condition.

4.2. Forward Solution of Piles in Homogeneous Soils Under Plane Strain Condition
This example investigates the PINN solution of a sheet-pile wall subject to vertical loading in homo-
geneous soils. The response of the wall is governed by the equilibrium Eq. (3) in Cartesian coordinate
systems. However, as sheet-piles are relatively long structural members in nature, simplified plane strain
description of the equilibrium equation is typically applied for the analysis of their response in soil medium.
In this respect, all derivatives with respect to y are vanished. In the absence of body forces, the equilibrium
equation under the plane-strain condition can be described as
α α
∂σxx ∂σzx α α
+ ≡ Pxx u =0 ,
∂x ∂z
(22)
α α
∂σxz ∂σzz α α
+ ≡ Pzz u =0 .
∂x ∂z

10
 In the plane-strain regime, the strain terms manifest in Eq. (3) are further simplified as

∂uα ∂uα ∂uα ∂uα

 
x z 1 x z
εα
xx = , εα
zz = , εα α
xz = εzx = + , (23)
∂x ∂z 2 ∂z ∂x
where α = P, Sk . Here, the compatibility constraint for the material interface of pile-soil system is expressed
as  P S
 ux (x) − ux (x) = 0 ,




uP S
z (x) − uz (x) = 0 , (24)




∀x ∈ (ΓP ∩ ΓS )


Consider a sheet-pile wall with the slenderness ratio of `0 /d0 = 5 subject to a vertical line load of Q =
10, 000 kN/m, as shown in Fig. 6. In this example, it is assumed that the surrounding soil is homogeneous
with the depth ratio of `1 /`0 = 2. The stiffness ratios η = EP /ES = 10, 25, 50 with EP = 5 GPa and
Poisson’s ratio νS = νP = 0.25 are considered in this problem. The PINNs solution of this problem is
performed by using the below set of neural networks

uP P
x ' Nux (x, z) , uP P
z ' Nuz (x, z) ,
(25)
uSx ' NuSx (x, z) , uSz ' NuSz (x, z) .

The architecture of all neural networks consists of 4 hidden layers with 20 neurons each, where hyperbolic-
tangent is used as the activation function. The loss terms of the total cost function in here are defined
identically to the previous example (see Eq. (21)), except that the indices r are now replaced by x. The
training is performed by means of 6000 sampling points, with 3000 points assumed for each of ΩP and ΩS
domains. NTK adaptive weighting is applied for the training in conjunction with a learning rate of 0.003.
The training history of the normalized loss versus epochs and time is reported in Fig. 7. As can be
seen, the loss function has immediately reached below the relative error norm of 10−5 within 500 epochs.
The improved performance in terms of convergence rate in comparison to the previous example is attributed
to the increased simplicity of the governing equations in the Cartesian system of coordinates. Finally, in
Fig. 8, contours of the normalized vertical displacement field uz /`1 is presented for all the cases considered
based on the PINNs solution and a reference FEM using COMSOL Muiltiphysics. The excellent agreement
between the PINNs results and the reference solution indicates the robustness of the extended framework
in the study of pile-soil systems under the plane strain condition.

4.3. Identification of Material Parameters by Inverse Analysis

In this example, we evaluate the performance of PINNs for the identification of model parameters involved
in pile-soil systems. For this sake, the interaction of a single pile with the surrounding homogeneous soil
medium under the plane strain condition is considered first. The material properties considered in here
are: Young’s Modulus of Elasticity, EP = 5 GPa, ES = 0.5, 0.1, 0.02 GPa; Poisson’s ratio, νS = νP = 0.25,
stiffness ratios η = EP /ES = 10, 50, 250. The input data-set is due to the longitudinal strain (or equivalently
stress) profile along the pile, which is obtained by means of optical fiber strain-sensing technique in practice
(e.g., see [50, 51]). In lieu of field data, a synthetic data-set is generated through the high-fidelity FEM
solution of the same problem using COMSOL software. To emulate the limitations encountered in practice,
merely a 1D profile along the center line of the pile is extracted from the FEM solution. This input data,
in turn, is employed for the inversion of the Young’s Modulus of the surrounding soil (i.e., ES ). Sampling

11
 Fig. 7: The network training history for the sheet-pile wall in homogeneous soils.

grid involves 3000 sampling points over each solution domain ΩP and ΩS , in conjunction with an input
data consisting of 2000 points (8000 points in total). It is noteworthy that a similar study is not feasible by
the use of conventional deep learning considering the sparsity of the input data in this problem. Here, we
demonstrate the versatility of PINNs in handling such study with extremely limited data-set.
The same governing equations presented in the previous example are applied in here for the inverse
analysis of piles in homogeneous formations (i.e., Eqs. (22)-(24)). However, the loss function is now updated
by the inclusion of an extra term corresponding to the available input data as

LT = LΩ + LΓB.C. + LΓCont + LΩσP ,

zz
P P
(26)
LΩσP = λ11 σzz − σ̄zz on Ω
,
zz P

0.0 0.00 0.0 0.0 0.0 0.0

0.2
0.2
0.05
0.2
0.1 0.2
0.4 0.4 0.4
0.10 0.2 0.4
0.6 0.6 0.6
0.3 0.6
0.8 0.15 0.8 0.8
1.0
1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00
1.0
1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00
0.4 1.0
1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00
(a) PINN Vsolution for η = 10 (b) PINN Vsolution for η = 25 V
(c) PINN solution for η = 50
0 0 0

0.04 0.1 0.2

0.08
0.2 0.4
0.12
0.3
0.6
0.16
0.4

(d) Reference solution for η = 10 (e) Reference solution for η = 25 (f) Reference solution for η = 50

Fig. 8: Contours of the normalized vertical displacement uz /`1 (10−2 ) for the sheet-pile wall in homogeneous domains (forward
solution).

12
 Fig. 9: The network training history for the inverse analysis of soil-pile interaction in homogeneous domains.

in which σ̄zP is the input data-set.

In Fig. 9, the training history of the normalized loss versus epochs and time is presented for the
identification study. As can be seen, the relative error norm has reached below 10−4 within 1000 epochs.
This indicates a substantial reduction in convergence rate in comparison to the forward study, which is
expected in inverse solutions. In Fig. 10, contours of the normalized vertical displacement field uz /`T are
presented for the backward solution and compared to the reference FEM solution using COMSOL. Evidently,
the PINNs results are in excellent agreement with the reference solution. In table 1 the exact values of the
Young’s modulus of the soil inverted in each case are reported. Notably, the deduced values in all cases lie
within 5% variation from the exact values.
In the last example, the application of PINNs to identify the material parameters of layered soils based
on the inverse analysis of the axial strain profile measured/obtained along the pile is demonstrated. Consider
a two-layered soil with the thickness ratios of `1 /`t = 0.25 and `2 /`t = 0.75. The material properties of

0.0 0.00 0.0 0.0 0.0 0.0

0.2 0.2 0.2
0.05 0.2 1.0
0.4 0.4 0.4
0.10 0.4 2.0
0.6 0.6 0.6

0.8 0.15 0.8 0.6 0.8 3.0

1.0 1.0 1.0
1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00 1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00 1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00
V V V
(a) PINN solution for η = 10 (b) PINN solution for η = 50 (c) PINN solution for η = 250

0 0 0
0.02 0.1 0.5
0.04
0.2 1
0.06
0.08 0.3 1.5
0.10 0.4 2
0.12 0.5
2.5
0.14
0.6
0.16 3
0.18 0.7
3.5

(d) Reference solution for η = 10 (e) Reference solution for η = 50 (f) Reference solution for η = 250

Fig. 10: Contours of the normalized vertical displacement uz /`T (10−2 ) the sheet-pile wall in homogeneous domains (inverse
analysis).

13
 Fig. 11: Training data history for the pile in layered soil.

the domain are: Young’s Moduli of Elasticity, EP = 5 GPa, ES1 = 0.1 GPa, ES2 = 0.02 GPa; Poisson’s
ratio, νP = νS1 = νS2 = 0.25. Here, the equations governing the response of homogeneous soils (i.e., Eq.s
(22)-(24)) need to be extended by inclusion of both soils layers (i.e., α = S1 , S2 ). In particular, the contact
constraints (i.e., Eq. (24)) need to be imposed across the interface of the pile with each soil layer as well
as between the soil layers itself (see Eq. (16)). Such derivations are straightforward task, which is not
presented here for the sake of brevity.
PINNs are employed to extract the Young’s moduli of both soils (i.e., ES1 and ES2 ). This inversion is
again carried out by using a 1D data-set involving the strain (stress) profile across the center line of the
pile. Six neural networks are elaborated for the PINNs solution in this example as

uP P
x ' Nux (x, z) , uP P
z ' Nuz (x, z) ,

uSx 1 ' NuSx1 (x, z) , uSz 1 ' NuSz1 (x, z) , (27)

uSx 2 ' NuSx2 (x, z) , uSz 2 ' NuSz2 (x, z) ,

which consist of the same architecture as the previous case. The loss function is extended to incorporate

0.0 0.0 0
0.2 0.2 0.2
0.4 0.4 0.4
0.6 0.6 0.6
0.8
0.8 0.8
1.0
1.00 0.75 0.50 0.25 0.00 0.25 0.50 0.75 1.00
V
(a) PINN solution (b) Reference solution

Fig. 12: Contours of the normalized vertical displacement uz /`T (10−2 ) for the pile in layered soil.

14
both layers of the soil in conjunction with the additional contact constraints as

LT = LΩ + LΓB.C. + LΓCont + LΩσP ,

zz

P P S S S S
LΩ = λ1 Pxx u on Ω + λ2 Pxx1
u 1 on Ω + λ3 Pxx2
u 2 on Ω
P S1 S2
P P S S S S
+ λ4 Pzz u on Ω + λ5 Pzz u on Ω + λ6 Pzz u on Ω
1 1 2 2
,
P S1 S2

P P P
S S S1

LΓB.C. = λ7 Bxx u − gxx on ΓP \(ΓS1 ∪ΓS2 )
+ λ8 Bxx1
u 1 − gxx onΓS1 \(ΓP ∪ΓS2 )
S S S2
P P P

+ λ9 Bxx u − gxx onΓ \(Γ ∪Γ ) + λ10 Bzz u − gzz on Γ \(Γ ∪Γ )
2 2
(28)
S P S1 P S1 S2
S S 2
S1
S S S2

+ λ11 Bzz1
u 1 − gzz + λ12 Bzz
on ΓS1 \(ΓP ∪ΓS2 )
2
u 2 − gzz on ΓS2 \(ΓP ∪ΓS1 )
,

LΓCont = λ13 uP − uS1 on + λ14 uP − uS2 on + λ15 uS1 − uS2 on

ΓP ∩ΓS ΓP ∩ΓS ΓS ∩ΓS
1 2 1 2
P S1
P S2
S S2

+ λ16 t − t on
+ λ17 t − t on
+ λ18 t − t on
1
,
ΓP ∩ΓS ΓP ∩ΓS ΓS ∩ΓS
1 2 1 2

P P

LΩσP = λ19 σzz − σ̄zz on Ω
,
zz P

Sampling grid involves 3000 sampling points over ΩP , 2000 points in each of the soil layers ΩS1 and
ΩS2 , and 2000 points due to the input data-set, which is summed at 9000 points in total. In Fig. 11, the
training history of the normalized loss is presented. As can be seen, the convergence rate versus epochs has
been relatively fast. Still, given the number of losses has expanded dramatically, the training is conducted
significantly slower with respect to the case of inversion in homogeneous soils. Contours of the normalized
vertical displacement field uz /`T are depicted in Fig. 12 for both PINNs and reference solution by COMSOL.
Evidently, excellent agreement is observed between the PINNs results and FE simulation. Finally, in Table
2 the inverted Young’s modulus for each soil layer is reported. The inverted values are in very good
agreement with the precise amounts. This further demonstrates the excellent performance of PINNs in
complex parametric studies pertaining to extremely limited input data.

5. Conclusions

A physics-informed deep learning framework is presented for the analysis of pile-soil interaction under
vertical loading. In the framework, a domain-decomposition multi-network model is introduced to deal with

Table 1: Inversion of soil Young’s Modulus for the pile embedded in homogeneous domains (unit: GPa).

pre exact
Analysis EP ES ES
Analysis 1. 5.0 0.0213 0.02
Analysis 2. 5.0 0.105 0.10
Analysis 3. 5.0 0.525 0.5

Table 2: Inversion of soil Young’s Modulus for the pile embedded in layered formation (unit: GPa).

pre excat pre excat

Analysis EP ES 1
ES1
ES 2
ES 2
5.0 0.095 0.1 0.023 0.02

15
the sharp discontinuities in the strain field at the interfaces of pile-soil regions and soil layers. The framework
is trained by minimizing the loss function defined in terms of the equilibrium equations and the boundary
conditions governing the pile-soil interaction problem. Several examples are provided to demonstrate the
performance of the framework in the analysis of single piles embedded in homogeneous and layered soils under
axisymmetric and plane strain conditions. Essential features of the framework are validated by comparing
the PINN results with the results obtained from the finite element analysis. Good agreement is observed
between the PINN and FEM results in all the cases considered. The application of the model for the inverse
analysis and parameter identification of pile-soil interaction is also presented. In the examples provided,
the localized data acquired along the pile length–possibly obtained via fiber optic strain sensing– is used for
the inversion of soil parameters in both homogeneous and layered formations. As expected, a substantial
reduction in the convergence rate is observed in the inverse analysis in comparison to the forward study.
However, it is seen that the proposed PINN framework is able to identify the material parameters quite
efficiently.

References
[1] O. C. Zienkiewicz, A. Chan, M. Pastor, B. Schrefler, T. Shiomi, Computational geomechanics, volume 613, Citeseer, 1999.
[2] O. Ghasemi-Fare, P. Basu, A practical heat transfer model for geothermal piles, Energy and Buildings 66 (2013) 470–479.
[3] A. Rahmani, A. Pak, Dynamic behavior of pile foundations under cyclic loading in liquefiable soils, Computers and
Geotechnics 40 (2012) 114–126.
[4] A. Jafari, M. Vahab, N. Khalili, Fully coupled xfem formulation for hydraulic fracturing simulation based on a generalized
fluid leak-off model, Computer Methods in Applied Mechanics and Engineering 373 (2021) 113447.
[5] M. Calvello, R. J. Finno, Selecting parameters to optimize in model calibration by inverse analysis, Computers and
Geotechnics 31 (2004) 410–424.
[6] A. M. Kabe, Stiffness matrix adjustment using mode data, AIAA journal 23 (1985) 1431–1436.
[7] S. Vardakos, M. Gutierrez, C. Xia, Parameter identification in numerical modeling of tunneling using the differential
evolution genetic algorithm (dega), Tunnelling and underground space technology 28 (2012) 109–123.
[8] G. Walton, S. Sinha, Challenges associated with numerical back analysis in rock mechanics, Journal of Rock Mechanics
and Geotechnical Engineering (2022).
[9] M. Lefik, B. A. Schrefler, Artificial neural network for parameter identifications for an elasto-plastic model of supercon-
ducting cable under cyclic loading, Computers & structures 80 (2002) 1699–1713.
[10] D. Gawin, M. Lefik, B. Schrefler, Ann approach to sorption hysteresis within a coupled hygro-thermo-mechanical fe
analysis, International Journal for Numerical Methods in Engineering 50 (2001) 299–323.
[11] E. J. Parish, K. Duraisamy, A paradigm for data-driven predictive modeling using field inversion and machine learning,
Journal of computational physics 305 (2016) 758–774.
[12] N. Kardani, A. Zhou, M. Nazem, S.-L. Shen, Estimation of bearing capacity of piles in cohesionless soil using optimised
machine learning approaches, Geotechnical and Geological Engineering 38 (2020) 2271–2291.
[13] E. Momeni, R. Nazir, D. J. Armaghani, H. Maizir, Application of artificial neural network for predicting shaft and tip
resistances of concrete piles, Earth Sciences Research Journal 19 (2015) 85–93.
[14] H. Moayedi, M. Mosallanezhad, Uplift resistance of belled and multi-belled piles in loose sand, Measurement 109 (2017)
346–353.
[15] S. K. Das, P. K. Basudhar, Undrained lateral load capacity of piles in clay using artificial neural network, Computers and
Geotechnics 33 (2006) 454–459.
[16] D. J. Armaghani, R. S. N. S. B. Raja, K. Faizi, A. S. A. Rashid, et al., Developing a hybrid pso–ann model for estimating
the ultimate bearing capacity of rock-socketed piles, Neural Computing and Applications 28 (2017) 391–405.
[17] G. Cybenko, Approximation by superpositions of a sigmoidal function, Mathematics of control, signals and systems 2
(1989) 303–314.
[18] C. M. Bishop, Pattern recognition and machine learning, springer, 2006.
[19] I. Goodfellow, Y. Bengio, A. Courville, Y. Bengio, Deep learning, volume 1, MIT press Cambridge, 2016.
[20] W. Zhang, H. Li, Y. Li, H. Liu, Y. Chen, X. Ding, Application of deep learning algorithms in geotechnical engineering: a
short critical review, Artificial Intelligence Review (2021) 1–41.
[21] M. Azimi, G. Pekcan, Structural health monitoring using extremely compressed data through deep learning, Computer-
Aided Civil and Infrastructure Engineering 35 (2020) 597–614.
[22] Y. Bao, Z. Tang, H. Li, Y. Zhang, Computer vision and deep learning–based data anomaly detection method for structural
health monitoring, Structural Health Monitoring 18 (2019) 401–421.
[23] X. Li, Z. Liu, S. Cui, C. Luo, C. Li, Z. Zhuang, Predicting the effective mechanical property of heterogeneous materials by
image based modeling and deep learning, Computer Methods in Applied Mechanics and Engineering 347 (2019) 735–753.

16
[24] D. Zhang, J. Lin, Q. Peng, D. Wang, T. Yang, S. Sorooshian, X. Liu, J. Zhuang, Modeling and simulating of reservoir
operation using the artificial neural network, support vector regression, deep learning algorithm, Journal of Hydrology
565 (2018) 720–736.
[25] S. M. Azimi, D. Britz, M. Engstler, M. Fritz, F. Mücklich, Advanced steel microstructural classification by deep learning
methods, Scientific reports 8 (2018) 1–14.
[26] E. De Bézenac, A. Pajot, P. Gallinari, Deep learning for physical processes: Incorporating prior scientific knowledge,
Journal of Statistical Mechanics: Theory and Experiment 2019 (2019) 124009.
[27] M. Reichstein, G. Camps-Valls, B. Stevens, M. Jung, J. Denzler, N. Carvalhais, et al., Deep learning and process
understanding for data-driven earth system science, Nature 566 (2019) 195–204.
[28] G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature
Reviews Physics 3 (2021) 422–440.
[29] R. S. Sutton, A. G. Barto, Reinforcement learning: An introduction, MIT press, 2018.
[30] J. Han, A. Jentzen, E. Weinan, Solving high-dimensional partial differential equations using deep learning, Proceedings
of the National Academy of Sciences 115 (2018) 8505–8510.
[31] E. Haghighat, M. Raissi, A. Moure, H. Gomez, R. Juanes, A physics-informed deep learning framework for inversion and
surrogate modeling in solid mechanics, Computer Methods in Applied Mechanics and Engineering 379 (2021) 113741.
[32] H. Owhadi, Bayesian numerical homogenization, Multiscale Modeling & Simulation 13 (2015) 812–828.
[33] Y. Bar-Sinai, S. Hoyer, J. Hickey, M. P. Brenner, Learning data-driven discretizations for partial differential equations,
Proceedings of the National Academy of Sciences 116 (2019) 15344–15349.
[34] S. H. Rudy, S. L. Brunton, J. L. Proctor, J. N. Kutz, Data-driven discovery of partial differential equations, Science
Advances 3 (2017) e1602614.
[35] M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving
forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics 378
(2019) 686–707.
[36] A. G. Baydin, B. A. Pearlmutter, A. A. Radul, J. M. Siskind, Automatic differentiation in machine learning: a survey,
Journal of machine learning research 18 (2018).
[37] M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat, G. Irving, M. Isard, et al.,
Tensorflow: A system for large-scale machine learning, in: 12th {USENIX} symposium on operating systems design and
implementation ({OSDI} 16), pp. 265–283.
[38] F. Chollet, et al., keras, 2015.
[39] E. Haghighat, R. Juanes, Sciann: A keras/tensorflow wrapper for scientific computations and physics-informed deep
learning using artificial neural networks, Computer Methods in Applied Mechanics and Engineering 373 (2021) 113552.
[40] E. Haghighat, A. C. Bekar, E. Madenci, R. Juanes, A nonlocal physics-informed deep learning framework using the
peridynamic differential operator, Computer Methods in Applied Mechanics and Engineering 385 (2021) 114012.
[41] M. Vahab, E. Haghighat, M. Khaleghi, N. Khalili, A physics-informed neural network approach to solution and identifi-
cation of biharmonic equations of elasticity, Journal of Engineering Mechanics 148 (2022) 04021154.
[42] M. Khaleghi, E. Haghighat, M. Vahab, B. Shahbodagh, N. Khalili, Fracture characterization from noisy displacement
data using artificial neural networks, Engineering Fracture Mechanics 271 (2022) 108649.
[43] Z. Mao, A. D. Jagtap, G. E. Karniadakis, Physics-informed neural networks for high-speed flows, Computer Methods in
Applied Mechanics and Engineering 360 (2020) 112789.
[44] F. Sahli Costabal, Y. Yang, P. Perdikaris, D. E. Hurtado, E. Kuhl, Physics-informed neural networks for cardiac activation
mapping, Frontiers in Physics 8 (2020) 42.
[45] E. Kharazmi, Z. Zhang, G. E. Karniadakis, hp-vpinns: Variational physics-informed neural networks with domain decom-
position, Computer Methods in Applied Mechanics and Engineering 374 (2021) 113547.
[46] S. A. Niaki, E. Haghighat, T. Campbell, A. Poursartip, R. Vaziri, Physics-informed neural network for modelling the
thermochemical curing process of composite-tool systems during manufacture, Computer Methods in Applied Mechanics
and Engineering 384 (2021) 113959.
[47] Y. Chen, L. Lu, G. E. Karniadakis, L. Dal Negro, Physics-informed neural networks for inverse problems in nano-optics
and metamaterials, Optics express 28 (2020) 11618–11633.
[48] A. D. Jagtap, E. Kharazmi, G. E. Karniadakis, Conservative physics-informed neural networks on discrete domains
for conservation laws: Applications to forward and inverse problems, Computer Methods in Applied Mechanics and
Engineering 365 (2020) 113028.
[49] Q. Lou, X. Meng, G. E. Karniadakis, Physics-informed neural networks for solving forward and inverse flow problems via
the boltzmann-bgk formulation, Journal of Computational Physics (2021) 110676.
[50] H. Mohamad, K. Soga, A. Pellew, P. J. Bennett, Performance monitoring of a secant-piled wall using distributed fiber
optic strain sensing, Journal of Geotechnical and Geoenvironmental Engineering 137 (2011) 1236–1243.
[51] H. Mohamad, K. Soga, P. J. Bennett, R. J. Mair, C. S. Lim, Monitoring twin tunnel interaction using distributed optical
fiber strain measurements, Journal of geotechnical and geoenvironmental engineering 138 (2012) 957–967.
[52] A. D. Jagtap, G. E. Karniadakis, Extended physics-informed neural networks (xpinns): A generalized space-time domain
decomposition based deep learning framework for nonlinear partial differential equations., in: AAAI Spring Symposium:
MLPS, pp. 2002—-2041.
[53] S. Wang, X. Yu, P. Perdikaris, When and why pinns fail to train: A neural tangent kernel perspective, Journal of
Computational Physics 449 (2022) 110768.
[54] C. Multiphysics, Introduction to comsol multiphysics®, COMSOL Multiphysics, Burlington, MA, accessed Feb 9 (1998)
2018.

17