Studia Geotechnica et Mechanica, 2023; 45(s1); 333–349

Special Issue 19th KKMGiIG Open Access

Artur Góral*, Marek Lefik , Marek Wojciechowski

Reduction of Numerical Model in Some

Geotechnical Problems
[Link]
received March 1, 2023; accepted August 18, 2023.
1 Introduction
Abstract: The concept of equivalence of the realistic, Despite the enormous progress in modeling of soil–
initial reference model and the simplified, reduced model structure interaction by discretization of the structure
is proposed. In reduced models, the action of the soil and the real domain of soil using finite element method
on the structure is replaced by the action of a layer with (FEM), discrete element method (DEM), or finite difference
prescribed properties, defined by a set of parameters. method (FDM), the reduced models are still very popular in
The main difficulty here is to find the parameter values engineering practice and scientific literature (e.g. [1, 2, 6,
required by the simplified theory. The subject of this work 11, 12, 25, 27, 35] and many other articles). There are various
is to find the dependence of the parameters of the reduced sources of complexity in numerical models. For example,
model on the parameters of the full model, including for composites, the model that takes into account the
realistic soil behavior, in order to ensure the equivalence microstructure at the whole macroscale by a simple fitting
of both models. We show the potential of the method by the FE mesh would lead to huge numerical task, impossible
presenting two examples: Winkler and Pasternak’s model in practical applications. In this case the homogenous-
of a plate on the ground. We assume that both models are like, theoretically homogenized medium plays a role
equivalent if they give identical results (displacements) at of the reduced model. In the field of civil engineering,
a finite number of observation points. An artificial neural there is an obvious need of analysis of the performance
network (ANN) is built in order to approximate and record of the entire structure before it is built, with FE mesh
the dependence of the parameters of the reduced model (at defined for all elements of the structure. In geotechnics,
the network output) from the parameters of the full model the entire structure includes the structure itself and a
(given at the network input). The complex network acts part of soil’s domain supposed to be in interaction with
as a formula that assigns the parameters of the reduced the structure. Traditional reduced models in geotechnics
model to a realistic description of the soil structure that operate as the action of some boundary constraints on
is used for finite element method (FEM) modeling. The the structure’s boundary, replacing the soil’s domain.
formalism we propose is quite general and can be applied These constraints must properly reflect the interaction
to many engineering problems. The presented procedure between the structure and the soil that depends on both
is entirely numerical; it allows to calculate the parameters constitutive properties of the structure and those of the
of the reduced model without resorting to symbolic soil. For example, the action of the soil on the structure
calculations or additional theoretical considerations. is replaced by the action of a linear or nonlinear elastic
layer (e.g. Winkler’s model, [34, 35]; Pasternak’s two-
Keywords: ANN in geotechnics; Winkler model; parameters model [22]) or a layer with other properties,
Pasternak model; reduced models. described by a number of parameters [11, 12]. In the oldest
model, that of Winkler, the unique parameter, stiffness
of the Winkler’s spring, was several times generalized [1,
18, 20, 23, 24, 32, 33]. In fact, this one-parameter model
may require a few additional characteristics if the spring
*Corresponding author: Artur Góral, Division of Geotechnics and rate is to be different for compression and extension, or
Engineering Structures, Department of Concrete Structures Lodz for varying spring’s stiffnesses for different stress levels.
University of Technology, Al. Politechniki 6, 93-590 Łódź, E-mail:
In the dynamic version of this model, it is necessary to
[Link]@[Link]
Marek Lefik, Marek Wojciechowski, Division of Geotechnics and enter the mass of the spring, parameters of viscous and
Engineering Structures, Department of Concrete Structures Lodz dry friction damping. Finally, as many as six parameters
University of Technology, Al. Politechniki 6, 93-590 Łódź can be needed. The formulae for the simplest set of

Open Access. © 2023 Artur Góral, Marek Lefik, Marek Wojciechowski, published by Sciendo. This work is licensed under the
Creative Commons Attribution alone 4.0 License.
334 Artur Góral, Marek Lefik, Marek Wojciechowski

parameters are basing on arbitrary assumptions, like for indicated as ANN-1. The inverse network ANN_2-1 is trained
example the depth of the soil’s active layer, and thus are by presenting the sets of values of the displacement in the
always disputable or uncertain. The same can be stated observation points obtained as a solution of the reduced
for the two-parameters model of Pasternak [22, 29, 30]. In theory at the input of the ANN_2-1 and the trial parameters
this model also the second parameter can be interpreted of the reduced model are elements of its output. This will
in two different manners: as a stiffness of a membrane- be explained in detail in Section 3.3.
like interface or Kirchhoff stiffness of the subgrade. The ANNs have been used in geotechnics for a long time.
situation becomes more difficult if the strength of the Most often, they are a tool for modeling constitutive
springs must be accounted for, like for piles or retaining relationships. The pioneering article here is the work
walls. In this last case, we need the surrogate of two [19], in which a classical neural network simulates the
Rankine’s coefficients and the Jaki’s type-parameter [35]. constitutive properties of a stratified soil medium based
Each of these parameters results from individual analysis on tests with the falling weight deflectometer (FWD)
requiring various “a priori” assumptions. In the complex dynamic test. A review of constitutive relationship
soil condition this reasoning requires a lot of engineering modeling strategies is a very interesting issue, but it is
intuition. A significant number of works are devoted to not related to the topic of this article. It seems that the
this issue. Recommendations regarding the values of formulation of the reduced model was not the subject of
these parameters can be found in each of the articles cited ANN application. If it is assumed that the homogenized
so far, some of which are devoted exclusively to this topic homogeneous medium is a model reduced in relation to
[14, 20, 21, 22, 23, 24, 25]. Winkler stiffness calculators the nonhomogeneous model, then the work [37] should
have been developed in programs used for engineering be mentioned, but there the classical neural network
calculations. The various hypotheses adopted in the serves only as a surrogate for the calculation of the
cited analyses regarding the physical nature of the soil effective modules of the homogenized medium performed
response, the depth to which the calculations are carried classically with the use of FEM. This network is trained
out, qualitative assumptions about the interaction of the in such a way that the periodicity cell described with the
soil with the structure lead to different estimates of the necessary number of parameters is assigned effective
physical parameters of traditional models (which can be modulus of elasticity. It seems that the concept of using
easily seen by analyzing any engineering problem). a complex neural network to evaluate the parameters of a
The subject of this article is to find the dependence of reduced model is a novelty.
the parameters of the reduced model on the parameters Even though the work deals with issues related to
of the full model in an automatic manner, via a numerical soil–structure interaction, we are convinced that the
procedure. First, we formalize the equivalence problem of general scheme of conduct may be useful in constructing
two different models, the initial (model of reference) and simplified models in a more general context. This
the reduced one. The algorithm of the procedure leading procedure seems to be novel and applicable in many
to the definition of the equivalent model will be described domains of engineering.
in turn. To obtain the equivalent properties of the reduced
model, the numerical tool involving a superposition of
two artificial neural networks (ANN) is developed. The
first neural network, called ANN_1, approximates a direct
2 Equivalence of the complex
solution of the model of reference. Here, the input will model (model of reference) and the
be soil and structure characteristics and mechanistic
properties of the materials. The output will contain some
reduced one
selected elements of solution of the reference problems
The main difficulty in the definition of the reduced
(e.g. beam deflection or pile settlement). The second
model is finding values of the parameters required by
neural network in this superposition, namely ANN_2-1,
the reduced theory. The values of these parameters are
approximates a solution of inverse problem related to the
obviously functions of the engineering properties of the
reduced model. This network is inverse with respect to the
soil and structure, and are given by algebraic formulas
ANN_2 - the network approximating the direct solution
only for elementary cases.
of the reduced problem. Please note, that the symbol
Under the term “initial model” or “reference model”
denoting the inverse function has been adopted here: if
we will understand a complex, realistic description of soil
the network ANN approximates the function f, then the
and the structure interaction, formulated originally in the
network that approximates the function inverse to f is
 Reduction of Numerical Model in Some Geotechnical Problems 335

form of a boundary value problem (BVP). Engineering this deflections fully characterize the pavement and the
approximations of the solutions of this BVP (with rare (possibly layered) subgrade, so the mechanical properties
exceptions) can only be efficiently obtained by discretizing of them can be computed by back analysis. More about
it, using the finite elements, the finite difference, or the this method can be found in [7, 21, 24, 28]. All sets of
discrete element method. We limit ourselves to the case, observation points in this article coincide with the ones
when the reference model is always FEM discretization from FWD test.
built for the analyzed engineering problem. Therefore, The equivalence criterion assures that a complex,
the reduced model parameters will be selected to imitate realistic behavior of the soil is taken into account in the
the FEM results. In this article, a Coulomb–Mohr model frame of the reduced model.
implemented by FEM is assumed to serve as a constitutive
model of soils. However, other constitutive models are also
possible to be applicable within the proposed framework.
The typical set of parameters (pCM1,…, pCMn) appearing
3 Scheme of the procedure
in the Coulomb–Mohr reference model of a layered soil
would consist of 5 elements by layer: 3.1 Approximator of direct problems

�𝐸𝐸𝐸𝐸𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 �, thickness of the layer (ℎ ))1, … ,( (𝑐𝑐𝑐𝑐), (𝜑𝜑𝜑𝜑The

), (𝜓𝜓𝜓𝜓crucial
), �𝐸𝐸𝐸𝐸𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒point
�, (ℎ )of
)n},the proposed algorithm is the use of
ANN, which is trained with results of FE computations
thickness of the layer (ℎ ))1, … ,( (𝑐𝑐𝑐𝑐), (𝜑𝜑𝜑𝜑), (𝜓𝜓𝜓𝜓), �𝐸𝐸𝐸𝐸𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 �, ({(ℎℎ)))n)},n}, of reference model for many sets of model’s parameters.
ANNs are operators that process an input dataset into
where n is the number of layers. This set of known an output set, taking expected values. ANN acts here
parameters can be extended as far as the FE code permits as a shortcut that, in the simple form and in a very low
and the BVP requires. computational cost, replaces the solution of the complex
The “reduced model” is formalized to be a pair of model. If there exist a function of many variables, the
two elements: differential equation that it defines and arguments of which are the parameters of the reference
an ordered set of parameters (p1,…, pn ) appearing in this model and the values are the displacements in the
equation. The physical range of the reduced model’s observation points, the ANN is able to approximate this
parameter values should always be known. From this function. It is guaranteed by the theorem that proves that
range, the test values of the parameters used to train the the ANN is a universal approximator of any function,
ANN are selected. Only within this range the equivalence functional, or operator [4, 9].
of the models will be valid. We assume that the physical Figure 1 depicts the schematic for the feed-forward
sense of the equation and the parameters is clear. The ANN, which is a general approximator of multivalued
assumption also is that this equation is solvable, with the functions of multiple variables [8, 10]. If the input and
solution being an ordered set of fields of displacement. output of the network are interpreted as an argument
This solution will be called “direct solution of the reduced and image of a certain operator or function that the ANN
problem”. approximates, the coincidence of the output values with
It is important to formulate the criterion of the expected values should be satisfied for an infinite
equivalence of the two models. We assume that the two number of all possible sets of arguments at input, with the
models are equivalent if they give identical (up to given error (in the sense of minimal squares) remaining below
tolerance) results in finite number of observation points. an acceptable threshold. To achieve this, the network
We assume that the reference model and the reduced one parameters (weights) are selected through successive
are formulated in displacements; thus, these results are corrections in an iterative process called “learning or
simply the displacements in the observation points. network training”. The training of the network minimizes
The choice of number and the placement of the the difference between the ANN output signal and the
observation points is not trivial in general. In all examples target signal expected at the ANN output. The most
presented in this article (there are always beams or plates common procedure of weight’s tuning is a minimization
on the elastic support) this choice is inspired by the well- of the sum of squared errors calculated for all output
known FWD test. In FWD test, the given mass is dropped neurons and for all input patterns. The back propagation
down on the surface of the pavement and the deflections error (BPE) procedure for such a minimization was used
due to this impact are measured in maximum nine points in this article. The complexity of the relation between
in some prescribed distances. It is assumed that the set of parameters and the resulting set of displacements is
336 Artur Góral, Marek Lefik, Marek Wojciechowski

Minimization
of the
difference

Figure 1: Scheme of the approximation of known learning output data (targets) by the transformation of the input signal on input layer
by trained ANN. Segments between nodes are symbolic representations of weights (multipliers) that modify the nodal activities before
attributing it to the next node.

adjusted by the number of layers and number of nodes Both solutions are classical, direct solution of the BVP.
in the layers that results with a set of approximation’s In what follows, we will need a so-called inverse solution
parameters (weights and biases of the ANN), sufficiently of the reduced model.
rich. Detailed description of construction and operation
of ANN have already been published in many textbooks or
monographs [8, 9, 10, 13] and articles [14, 15, 16]. We refer 3.2 Approximator of inverse problem
to any of the above-listed manuals for details concerning
the ANN. The process of determining the parameters appearing
In Figure 2, the approximation by ANN_1 of the FE in the “direct” problem through the interpretation of
solution for the initial model and the approximation by measurement data obtained as a result of the conducted
ANN_2 of the direct solution for the reduced model are physical experiment in the real world, consists of
shown in parallel. the formulation and solution of the inverse problem.
The training scheme of these networks is shown Unknowns of the inverse problem are the parameters
in the example: the reference model is a two-layer soil occurring in the direct problem and should be selected so
characterized by Young’s modulus and Poisson’s ratio that the solution obtained then is the closest one to the
with indices denoting the layer’s number, and EJ – the given, observed values. In our case we need the values
stiffness of the beam resting on this substrate (Figure of the parameters of the reduced model, such that for
2.a.). The trial values of these parameters are directed to the given displacements in the observation points, the
the network input. The activation values of the network’s direct solution is the closest to them. The minimum of
output neurons are interpreted as the deflection values the distance is understood in terms of the sum of squared
at the observation points. The network is trained using deviations from the given data in observation points.
the deflection values at these points obtained by FEM For any given direct problem (any typical BVP), there
calculations. The scheme of training a network ANN_2 are various inverse problems, the mathematical nature
that approximates the direct solution of the reduced of which is different than that of the direct one. In our
model is presented in Figure 2.b. In this case, the three previous articles [15, 16], we described an application of
parameters of the model are the Winkler’s ground ANN for the solution of inverse problem.
stiffness, the beam stiffness, and the ground parameter It is to note, that in the algorithm we propose to
related to the Pasternak model. Output neurons have the construct the final, superposed neural network ANN_3
same interpretation as in Figure 2.a. (defined later in Section 3.3) we execute at once two
 Reduction of Numerical Model in Some Geotechnical Problems 337

a.

Figure 3: Scheme of numerical solution of inverse problem defined

for the reduced model. In training, input to the “inverse ANN” are
displacements in the observation points—the direct solutions of the
reduced model while targets are the values of the parameters for
which these displacements have been computed. In recall mode,
the well-trained ANN responds with correct model parameters for
the set of displacements obtained at the input.

3.3 Procedure description

Assuming, that for the reduced problem the inverse

solution is unique, the parameters of the reduced model
can be easily computed via composed ANN_3:
b.
nthetic representation of the results of the model solution in observation points in the form ANN_3
of = ANN_1@(ANN_2-1) (1)
Figure 2: Synthetic representation of the results of the model
solution in observation points in the form of the ANN. Figure 2.a.
represents the ANN approximation of the FE solution for the model Symbol @ is used here to denote an action of an operator
of reference while in Figure 2.b. the same scheme applies for the (ANN_1) on an object (ANN_2-1)).
reduced model. The structure of the ANN_3 is illustrated in Figure 4.
It is seen that the composed, complex ANN_3 that assigns
the parameters of the reduced model to the parameters of
trainings: the one illustrated in Figure 2.a. and the other the initial model is composed as follows:
described in Figure 3. – Input layer contains a number of neurons equal the
The most important advantage of this method is that it number of parameters of the initial model.
is not necessary even to formulate the inverse problem. It – Block of hidden layers consists of:
is enough to generate, using the well-known engineering o all hidden layers from the direct ANN_1 (coding
tools, a set of direct solutions. These direct solutions are the direct solution of the initial problem);
then used as an input of the ANN, the output of which are o additional layer of nine neurons (in general—as
corresponding parameters known in training. The well- many as the number of control points is) with
trained ANN in the “recall mode” attributes at output weights: at the layer input—taken from network
the correct parameters of the BVP for any direct solution ANN_1, at the layer output—taken from ANN_2-1;
presented at the input layer. In the case of inverse solution o all hidden layers from the inverse network ANN_2-1
for reduced model, this procedure is illustrated in Figure (coding the inverse solution of the reduced
3. Unfortunately, the solution of inverse problem by well- problem).
trained ANN has also some disadvantages. The inverse – Output layer contains a number of neurons equal the
solution is often nonunique and in this case the ANN number of parameters of the reduced model.
requires many additional efforts to obtain all solutions.
338 Artur Góral, Marek Lefik, Marek Wojciechowski

h1 ,
E1, y1
y1
ν1, Hidden y2
layers of y2 kW
h2 , y3 Hidden
ANN_1 y3 layers of EJ
E2, …
…
ANN_2(-1) G
ν2, y9
y9
EJ

Figure 4: The complex ANN_1@(ANN_2-1) acts as a formula that assigns the parameters of the reduced model to the realistic parametric
descriptions of the problem that is used for its FE solution.

The following steps should be accomplished to define 4 Illustrative examples

the complex network ANN_3:
– Execution of direct solutions of the initial problem, In this section, two examples of application of the proposed
using FEM for representative set of input structural procedure will be presented. The first one illustrates the
and soils parameters. In the results of this operation, application of the proposed procedure in the case of the
an its element of training patterns set for the network Winkler model of the plate, in the second one we consider
ANN_1 has the form: the beam on the two-parameter Pasternak soil model.

{input, target}i = { (combination of structural and

soils parameters of BVP)i, (y1, y2, .. yi, .. y9)i } (2) 4.1 Stiffness coefficient for Winkler model of
soil–structure interaction
– Execution of direct solutions of the reduced problem,
for representative set of parameters of the reduced The oldest reduced model is the model of a plate or beam
model. In the results of this operation an its element of resting on a Winkler elastic foundation. This model
training patterns set for the network ANN_2-1 has the was formulated in 1876 by Winkler [34]. According to
form: Winkler’s formulation, the ground reaction vector at a
certain point on the soil-building boundary is determined
{ (y1, y2, .. yi, .. y9)i, (corresponding combination by the structure displacement vector at that point and is
of parameters of reduced model)i } (3) proportional to the beam deflection. The coefficient of
proportionality—Winkler stiffness is denoted by kw in the
– Training of the network ANN_1 with the set of training following. The assumptions of this model are well known
data (2); and do not require discussion in this article.
– Training of the network ANN_2-1 with the set of
training data (3);
– Creation of the resulting ANN_3 according to (1). 4.1.1 Reference model

The application of complex network ANN_3 consists in Figure 5 shows a diagram of the considered models.
introducing parameters of the initial, exact model of the The concept of model corresponds to the finite concrete
problem at the network input. Then the parameters of pavement on subgrade. Model A subgrade is represented by
the reduced problem will be calculated at the networks one layer of natural soil. Model B subgrade is represented
output. by two layers of natural soil; each differ where thickness
 Reduction of Numerical Model in Some Geotechnical Problems 339

ls
load
8 x dL reading points
x
(W0) Slab

(W1) Soil layer

y
l

ls
load
8 x dL reading points 3 x dL
x
(W0) Slab

(W1) Soil layer (model B) / Soil-cement (model C)

(W2) Soil layer

y l

Figure 5: Diagram of the considered layered pavement models. From top: model A; model B, C.

of upper layer of natural soil change in a given range. depth where the effective stresses due to external loading
Model C subgrade is represented from top: by a layer of are less than 20% of the effective primary stresses due to
stabilized soil (soil–cement mixture) and natural soil. the soil’s own weight σy. The horizontal dimension of top
Model A will be associated with the Winkler model of the slab layer is ls=4 m. This condition ensures proper decay
plate. Model B and C will be associated with the Pasternak of deflection along the horizontal direction of subgrade.
model of soil, respectively. The extract of data for the As boundary condition horizontal displacements are
models are summarized in Table 1. Owing to the symmetry blocked in the axis of symmetry and on the right edge of
of the pavement model, only half of the geometry model the model. At the bottom of the model, a full displacement
was considered. Dimensions of the analyzed area of lock is used. A static constant load is on the top surface
subgrade are l=8 m in horizontal direction and h depends with width of 15 cm and value of 700 kPa. The value and
on the model. The last layer of natural soil across models range of the load corresponds to the static vertical contact
represents the last layer in elastic half-space theory and stresses apparent under the load plate of FWD vehicle.
meets it requirements. This dimension ensures decay of In engineering practice, the dynamic load from a falling
general stresses according to Eurocode 7 guidelines, for mass is commonly replaced with an evenly distributed,
example, recommends that integration be carried out to a equivalent static load [25]. Its realistic nonlinear course
340 Artur Góral, Marek Lefik, Marek Wojciechowski

Table 1: Summary of data for the applied model of the layered pavement.

Model A Model B Model C

Layer hj vj Ej hj vj Ej hj vj Ej
[cm] [-] [MPa] [cm] [-] [MPa] [cm] [-] [MPa]

W0 15÷25 0,2 32000 15÷25 0,2 32000 15÷25 0,2 32000

W1 300 0,25 50÷200 100÷200 0,25 50÷200 20 0,17 9100÷13850

W2 - - - 300 0,25 50÷200 380 0,25 50÷200

Figure 6: Examples of finite element meshes of the models.

leads to dynamic analysis, which is very interesting, to 15 cm at the bottom surface. The FEM model and the
but is beyond the scope of this article. Such an analysis calculations described below were performed using the
was presented in a conference keynote presentation [17]. fempy software [36].
Standard surface quality testing using FWD test is only For the models defined above, total 150 sets of data
inspiration. FWD device is used to apply a dynamic load were randomly generated. Total thickness hj for concrete
to the road, which simulates the pressure from the vehicle slab or subgrade layer and deformation modules EJ
tire. Pavement deflection is measured using geophones. for subgrade layers each time were random and it was
However, in this example, only the number and type of assumed all other parameters of the task as invariant.
measurement data and their distance from the point The sampling was carried out using the Latin hypercube
where the load is applied will be taken from the real FWD sampling (LHS) method. It is a statistical method of
test. generating samples with a multivariate distribution [13]. It
Figure 6 shows the finite element mesh adopted for ensures that the random set of samples is representative for
the analyzed boundary problem. Six-node triangular the given ranges of parameter variability (while the usual
elements were used. Edge size of equilateral triangle is random sampling is just a collection of random numbers
5 cm at the top surface of the model, evenly increased with no guarantees whatsoever). Then, calculations
 Reduction of Numerical Model in Some Geotechnical Problems 341

Figure 7: Deflections of the pavement for the selected sets of stiffnesses. From left: model A, model B. Deflections between (from 2.4 to 4.0) [m]
are not recorded.

were performed for all generated samples, registering uk

deflections on the top surface of subgrade at 9 points 0,
30, 60, 90, 120, 150, 180, 210, and 240 cm from the plane
of symmetry under concrete pavement. Next additional 4
deflections at 400, 415, 430, and 445 cm from the axis of
symmetry on the top surface of subgrade. In these points,
the influence of deflected plate on surrounding subgrade
is registered for the Pasternak model. In this way, data was
obtained in sequence (Ej(p), uk(p) where Ej(p)—thickness of
slab layer or subgrade and elasticity modules of subgrade
layers, uk(p)—deflections of top surface, p = 1,…, 150, j = 1, 2
in model A, j = 1, 2, 3 in model B and C, k = 1,…, 13. sample

4.1.2 Artificial Neural Networks that surrogate the

results of the FE computations

Having the data from FE computations, the ANN that

approximates the relation among the soil’s parameters
and the deflections of the plate in the nine observation
points will be constructed. According to (1), this is the
first component of the complex network, namely ANN_1.
The structure of that network depends on the case of
the reference model, A, B, or C, having from two to four,
input nodes for Young’s moduli of each of the layer and
sample
its widths (see Fig. 5). At the output we deal with nine or
thirteen observable deflections in each case. In the case
of 10 nodes in the hidden layer (model A) and 14 nodes Figure 8: Direct Nk network learning results for two of all nine
reading point of pavement deflection: 0 in title—1st reading point, 7
(model B and C), the structure of the ANN_1 is similar
in title—8th reading point, target—reference deflection values, opt—
for each of the analyzed examples. Network training was deflection values identified by the trained ANN. On the horizontal
carried out on 125 sets of training data, the remaining 25 axis—number of patterns. The target and output coincide (the blue
were used for network testing. The criterion for stopping line is not visible!)
the training is always the minimum of the root mean
square error (RMSE) error for the testing set.
342 Artur Góral, Marek Lefik, Marek Wojciechowski

u_0 uniform load

u_1
u_2
reading points
8 x dL

E, ν, h x

y(x)

kw

Winkler elastic soil

y

Figure 9: Normalized response of the network versus the pattern

Figure 10: Scheme of the task of identifying the mechanical
deflection values: u_0—results for 1st reading point, u_1—results for
parameters of the surface and soil, description of symbols in the
2nd reading point, u_2—results for 3rd reading point. The points on
text.
the graph also contain test data.

The solution to the above equation is of the form:

The learning results are very good. A perfect match of
the deflection values was achieved even though the ANNs 𝑤𝑤𝑤𝑤(𝑥𝑥𝑥𝑥) = 𝑤𝑤𝑤𝑤𝑠𝑠𝑠𝑠 (𝑥𝑥𝑥𝑥) + 𝑒𝑒𝑒𝑒 δ 𝑥𝑥𝑥𝑥 (𝐴𝐴𝐴𝐴 sin δ 𝑥𝑥𝑥𝑥 + 𝐵𝐵𝐵𝐵 cos δ 𝑥𝑥𝑥𝑥 ) + 𝑒𝑒𝑒𝑒 −δ 𝑥𝑥𝑥𝑥 (𝐶𝐶𝐶𝐶 sin δ 𝑥𝑥𝑥𝑥 + 𝐷𝐷𝐷𝐷
have unfavorable architecture (more exits than inputs).
(4.2)
It is worth noting that the quality 𝑤𝑤𝑤𝑤(𝑥𝑥𝑥𝑥) =of𝑤𝑤𝑤𝑤the
𝑠𝑠𝑠𝑠 (𝑥𝑥𝑥𝑥)fit
+ on
𝑒𝑒𝑒𝑒 δ 𝑥𝑥𝑥𝑥the testδ 𝑥𝑥𝑥𝑥 + 𝐵𝐵𝐵𝐵 cos δ 𝑥𝑥𝑥𝑥 ) + 𝑒𝑒𝑒𝑒 −δ 𝑥𝑥𝑥𝑥 (𝐶𝐶𝐶𝐶 sin δ 𝑥𝑥𝑥𝑥 + 𝐷𝐷𝐷𝐷 cos δ 𝑥𝑥𝑥𝑥 )
(𝐴𝐴𝐴𝐴 sin (4.2)
results does not differ from that on the training set, what
indicates is that the ANN has been trained correctly.
where: ws(x) – is a particular integral of equation (4.1),
the form of which depends on the form of the function
p(x), while the constants A, B, C, and D are obtained from
4.1.3 Reduced model
boundary conditions.
We assume that p(x) is zero. In the reduced model, the
The reduced model is here the classic Winkler model
load is concentrated at the symmetry plane.
with bilateral soil reactions and the Euler–Bernoulli
Using suitable loops over the parameters, the Maple
hypothesis for beam deflection was adopted. We consider
code computes these deflections for given x-coordinates.
a model of an elastic, homogeneous, finite beam band on
In order to solve the inverse problem, it is now necessary
a homogeneous Winkler elastic soil, loaded with a linear
to construct a set of learning patterns (input patterns),
load on the plane of symmetry. In the inverse problem
allowing for the approximation of inverse relationship by
being solved, the thickness of the beam strip layer and the
ANN. At the input of this network, solutions calculated
Winkler stiffness of the subsoil are identified.
from the formula (4.2) should appear. The selected
In order to construct ANN_2-1, it is necessary to solve
measurement points are, of course, geophones. The
the inverse problem with respect to following boundary
standard distance between geophones ΔL was assumed
value problem (reduced model):
equal to 0.3 m. The output of the network should contain
Find the deflection line y(x) for the finite plate strand:
the parameters of the reduced problem for which the
solution (4.2) was obtained. Natural questions arise here
0≤x≤L,
about how to choose the test parameters of the problem
explicitly. There should be as few of them as possible
where: L – half of the slab span [m].
so that the task of training the network is not time-
The deflection line is a function of one variable that
consuming, and at the same time enough to make the
satisfies the equation:
approximation of the inverse relation accurate enough. In
this example, there are two variables. The training set will
𝑑𝑑𝑑𝑑44𝑦𝑦𝑦𝑦(𝑥𝑥𝑥𝑥)
𝑑𝑑𝑑𝑑 𝑦𝑦𝑦𝑦(𝑥𝑥𝑥𝑥) 4
4 𝑝𝑝𝑝𝑝(𝑥𝑥𝑥𝑥)
𝑝𝑝𝑝𝑝(𝑥𝑥𝑥𝑥) 44 𝑘𝑘𝑘𝑘 𝑘𝑘𝑘𝑘 contain pairs that match the notation:
4ο
+4 l4 𝑦𝑦𝑦𝑦(𝑥𝑥𝑥𝑥) = 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 , οl = �4𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 (4.1) (4.1)
𝑑𝑑𝑑𝑑𝑥𝑥𝑥𝑥44
𝑑𝑑𝑑𝑑𝑥𝑥𝑥𝑥 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 4𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
 Reduction of Numerical Model in Some Geotechnical Problems 343

Reduced problem, displacements uy

0,002

0,000
0 0,3 0,6 0,9 1,2 1,5 1,8 2,1 2,4
-0,002
uy [m]

-0,004

-0,006

-0,008

-0,010
pavement alog x-axis [m]

Figure 11: Trial deflections computed in the frame of the reduced model.

{ (y1, y2, .. yi, .. y9)i, (corresponding combination of

parameters of reduced model: k, h)i } (5)

It should be noted that in any real technical problem

related to the selected theoretical model, the range of
identified quantities is usually well defined. For example,
we can estimate with a high degree of certainty the range
in which the Winkler stiffness or the value of the Pasternak
constant, naturally related to the Kirchhoff constant of the
soil medium, should lie. These assumptions are important
because network training will only be performed for
values in this range. It is generally believed that ANN
sample extrapolates the results of the approximation very poorly
beyond the interval for which it was trained. There are
theoretical rules for sampling the parameter space, in
the former example the LHS procedure has been used
to this purpose. In this example, however, a simplified
sampling method was used. All inputs were prepared
using the Excell RAND() function, which returns evenly
distributed random real numbers. In this simple example,
we did not observe an unfavorable effect on the ANN’s
ability to generalize. This means that the observed RMSE
error on the test set decreased as fast as when using the
LHS procedure in the previous example. In the solved
example, the following limitations of the task parameters
were adopted: the thickness of the beam is in the range [15
sample cm, 25 cm], and the stiffness of the Winkler subsoil is in
the range [9 000 kPa, 60 000 kPa]. A simplified random
Figure 12: The results of learning the “inverse” network ANN_2-1 . method of sampling the space of these parameters, basing
Targets—are values of mechanical properties to learn, opt—values of
on Excel RAND() function was adopted. In Figure 11, a set
mechanical properties identified by the trained net.
of trial deflection of the beam-plate is shown.
344 Artur Góral, Marek Lefik, Marek Wojciechowski

4.1.4 Artificial Neural Networks that approximate the model is also excellent. Thus, the necessary condition for
inverse relation for reduced model constructing the complex neural network ANN_3 (1) are
fulfilled.
According to (3), the network ANN_2-1 has 9 input and 2
output neurons. The output parameters of pavement
mechanical properties are W0 for slab layer and k (Winkler 4.1.5 Quality of the solution
modulus) for subgrade layer. The network architecture
consists of five neurons in one hidden layer. Network In this section, the ANN1 net was verified with three
training was carried out on 125 sets of training data, the samples of input parameters. The output results will be
remaining 25 were used for network testing. used for ANN_2-1 net verification. Figure 14 shows testing
Figures 12 and 13 show the results of the pattern fits of the complex network.
and deflections determined by the trained ANN_2-1 for all It is seen from the Figure 14 that the deflections of the
150 samples. plate for reduced model and the reference model coincide.
As in the reference model (for direct dependence of
the displacements on the soil’s parameters), the quality of
the approximation of the inverse problem for the reduced 4.2 Identification of two parameters of
Pasternak model of soil–structure interaction

Pasternak’s model (also called the two-parameter model),

published in 1954 [16], should be mentioned as another,
qualitatively different classical reduced model. Despite
various physical interpretations, this model is based on
introducing a second derivative of deflection into the beam
equation. According to the well-known interpretation,
the parameter G of the Pasternak’s model is obtained by
introducing a layer exhibiting an elastic shear response
into the subsoil’s idealization scheme. In a situation
where an elastic layer of the Winkler type also appears
in the soils scheme—such a model is associated with the
name of Kerr. Another interpretation suggests that the
coefficient accompanying the second derivative of the
beam’s deflection results from an action of the membrane
Figure 13: Normalized response of the “inverse” network versus
expected output. The points on the graph also contain test data. underlying the beam or plate. While the first interpretation
permits to guess the value of G as determined by Kirchhoff

Quality of solution
0
-0,0005 0 0,3 0,6 0,9 1,2 1,5 1,8 2,1 2,4
-0,001
-0,0015
uy [m]

-0,002 ANN1_s1

ANN1_s2
-0,0025
ANN1_s3
-0,003
ANN2^(-1)_net_verif_s1
-0,0035
ANN2^(-1)_net_verif_s2
-0,004
pavement alog x-axis [m] ANN2^(-1)_net_verif_s3

Figure 14: Reference (target) and ANN_3 identified (net_approx.) deflections of the pavement for three random test cases.
 Reduction of Numerical Model in Some Geotechnical Problems 345

uniform load reading points

8 x dL 3 x dL

E, ν, h x
G
elastic shear layer
w(x)
kw

y Winkler elastic soil

Figure 15: Idealization of the interaction of the pavement and soil, the positive direction of the axis and load as well as the assumed material
parameters are marked. A detailed description of this scheme—in the text of the article.

modulus of the soil, the second one the value of G in a very −L ≤ x ≤ L

speculative manner. This problem can be important in the
modeling of soils reinforced with geogrids.
The deflection line w(x) is a function of one variable that
satisfies the equation:

4.2.1 Reference model

d 4 w ( x ) G d 2 w ( x ) kW p( x)
− + w( x) = (6.1)
The reference model is the same as for the Winkler model, dx 4
EJ dx 2
EJ EJ
described in Section 4.1.1. The only exception is that for
the Pasternak model the last four points, external with
The soil deflection line u(x) is a function of one variable that
respect to the plate strip, are taken into account. To keep
satisfies the equation:
the same number of displacements as in the previous
example, we skip the four last observation points under
the plate strip. d 2u( x )
−G + kW u( x ) = q( x ) (6.2)
dx 2

4.2.2 Reduced model

These equations should be solved for a given load p(x) with
boundary conditions assuming zeroing of u(x) at an infinite
The scheme of the Pasternak’s model is shown in Figure
distance from the origin of the coordinate system.
15. We assume that the plate-strip deflection line w(x) is
The solution to the above equations is of the form:
identical to the soil deflection line u(x) at all points under
the beam. Apart from the beam, the soil deflection line
satisfies other differential (6.2) and coincides with the 𝑤𝑤𝑤𝑤(𝑥𝑥𝑥𝑥) = 𝑤𝑤𝑤𝑤𝑠𝑠𝑠𝑠 (𝑥𝑥𝑥𝑥) + 𝑒𝑒𝑒𝑒 δ 𝑥𝑥𝑥𝑥 (𝐴𝐴𝐴𝐴 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥 + 𝐵𝐵𝐵𝐵 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥) + 𝑒𝑒𝑒𝑒 −δ 𝑥𝑥𝑥𝑥 (𝐶𝐶𝐶𝐶 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥 + 𝐷𝐷𝐷𝐷 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑠𝑠𝑠𝑠
(6.3)
plate-strip deflection. 𝑤𝑤𝑤𝑤(𝑥𝑥𝑥𝑥) = 𝑤𝑤𝑤𝑤𝑠𝑠𝑠𝑠 (𝑥𝑥𝑥𝑥) + 𝑒𝑒𝑒𝑒 δ 𝑥𝑥𝑥𝑥 (𝐴𝐴𝐴𝐴 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥 + 𝐵𝐵𝐵𝐵 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥) + 𝑒𝑒𝑒𝑒 −δ 𝑥𝑥𝑥𝑥 (𝐶𝐶𝐶𝐶 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥 + 𝐷𝐷𝐷𝐷 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑠𝑠𝑠𝑠 δ 𝑥𝑥𝑥𝑥) (6.3)
Boundary problem written by equation (6), is
associated with the scheme presented in Fig. 15 understood
where: ws(x)—is a particular integral of equation (6.1), the
as the reduced model:
form of which depends on the form of the function p(x), while
Find the deflection lines of the plate strip w(x) and the
the constants A, B, C, and D are obtained from appropriate
soil u(x) for x from the interval,
boundary conditions.
346 Artur Góral, Marek Lefik, Marek Wojciechowski

prediction of Young's modulus E

45000

40000

35000
E [MPa]

30000

25000

20000
0 20 40 60 80 100
sample number

expected values computed values

a.

prediction of Winklers's modulus kW

120000

100000

80000

60000
kw [kN/m]

40000

20000

0
0 20 40 60 80 100
sample number

expected (target) values computed values

b.

prediction of Pasternak's coefficient G

160000
140000
120000
100000
80000
G [kPa]

60000
40000
20000
0
0 20 40 60 80 100

sample number

expected (target) value computed value

c.

Figure 16: Comparison of the results of the network calculations with the known values of the material parameters of the Pasternak model:
in Fig. 16.a, it is the Young’s modulus of the stiffness of the E beam; in Figure 16.b it is the Winkler stiffness of the ground kW; in Figure 16.c
it is Pasternak’s constant G. These are the result of calculations in the reminder mode for a trained network with the structure of ANN_953,
for 100 datasets, nine deflections each.
 Reduction of Numerical Model in Some Geotechnical Problems 347

The adopted designations are defined as follows: 4.3 Possible applications and limitations
of the proposed method of formulation of
1 1 1 1 1 1 k k4W kW reduced models
α = α =α4λ=2 4+λG24+
/λEJ
2
β = β =β4=λ 2 4−λG42 λ−/ 2G
G+/GEJ/ EJ / EJλ = λ4 =λW4=
− /GEJ
EJ (7) (7) (7) (7)
2 2 2 2 2 2 4 EJ4 EJ4 EJ The aim of this work is to present the method of determining
the parameters of the reduced model rather than to
Two sets of 150 pairs (3) were generated for random present spectacular engineering applications. Therefore,
values of all three parameters of the Pasternak model. The the presented examples have the value of illustrating
sampling ranges of the parameter values are as follows: the proposed procedure. However, they illustrate three
E in the range from 20 000 to 45000 MPa, kW in the range important elements of this procedure. Its success depends
from 20 000 to 110 000 kPa. on the ability of the ANN_1 network to synthetically
replace the FEM solution at several representative points,
on the possibility of well-trained ANN_2-1 inverse network.
4.2.3 Artificial Neural Networks that approximate the The third, most important advantage of this procedure is
inverse relation for reduced model (the Pasternak one) the fact that it takes into account a very wide context of
real ground conditions to determine the parameters of
In this section, we discuss the training results of the the reduced model. The examples show that the ANN_1
network ANN_2-1—the second part of the complex network is a very good surrogate for the FEN solution in
network ANN_3 (3). For a three-layer network with nine the scope we need. It was also shown that the ANN_2-1
input neurons (for nine values of measured deflections), inverse network can be trained very well. In order to create
five neurons in the hidden layer and three output the ANN_3 network, which generates the constitutive
neurons, very good training results were obtained for all parameters of the reduced model as an output, it should
three parameters of the reduced model. The value of the be trained for a situation in which the FEM model is
correlation coefficient of the training set with the network described by a large number of parameters describing, for
response was at the level of 0.997, while the mean square example, a system of layers not parallel to the surface. This
error of the calculated parameter values was of the order was not done in this article, but we are sure that ANN_1,
of 0.02. The network was not optimized; however, it seems even for an input composed of a large number of neurons,
that the adopted ANN variant is very simple. You can is a good approximation of the “straight” FEM solution at
certainly limit the number of neurons in the input layer. selected points [37]. The biggest limitation of the method
Nothing is lost in the precision of the approximation with is the ability to precisely train the inverse network. The
six input neurons with the values of the first six measured inverse relationship is often not a one-to-one function and
deflections. Discussion regarding the necessary minimum problems may arise with effective training of the network.
number of experimental deflection data is omitted here. We are convinced, however, that in a situation where a
Figures 16.a, 16.b, and 16.c show a comparison of the properly constructed reduced model is described with few
network calculation results with the known values of parameters and each of these parameters characterizes
material parameters E, G, and kW. These comparisons are a different aspect of the soil–structure interaction—the
done in the recall mode for new, 200 verification datasets. inverse network will always be a good approximator of the
None of these sequences were used during training. Only inverse relationship.
100 such comparisons are presented, for the sake of In the light of what was written above, the excellent
clarity. agreement of the FEM solution with the approximate
solution obtained using the reduced model was obtained
only in a simple case built to illustrate the presented
4.2.4 Quality of the reduced model method. We are well aware of the fact that in the case
of the subgrade under the slab, a model with variable
It should be stated that the obtained ANNs: ANN_1 and Winkler stiffness should be adopted [20, 35]. We are
ANN_2-1 are very accurate approximation of the direct convinced that also much more difficult and important
and inverse relation, i.e. the dependence of the model problems can be solved using the developed method. For
parameters on the parameters of the reference model. The example, pile-bearing capacity calculation methods that
parameters of Pasternak’s model can therefore be easily require the use of transfer functions describing pile–soil
calculated via ANN_3, without resorting to speculative interactions are, in our understanding, reduced models.
theoretical formulas. The parameters of the transfer functions depend on the
348 Artur Góral, Marek Lefik, Marek Wojciechowski

properties and state of the soil in the vicinity of the pile

lateral surface. These parameters can be obtained as the
References
output values of the ANN_3 network, the input of which [1] Ataman, M., Zbiciak A., Szcześniak, W.: Drgania własne
should be given the characteristics of the soils in the płyty cienkiej na podłożu Kerra – Jemielity, [Link]
vicinity of the node where the transfer function is to be [Link]/publication/263414439, 2009
defined. Unfortunately, in this case, even the reference [2] Ataman, M.: Inercyjne, Sprężyste modele podłoża
model is not easy to construct. odkształcalnego w zadaniach z mechaniki konstrukcji,
Logistyka-nauka, 6, s. 1515, 2014.
Such research, as well as the examples in this article,
[3] Boso, D.P., Lefik, M. and Schrefler B. A.: Generalised self
is the content of the doctoral dissertation being prepared consistent homogenisation as an inverse problem. Z. Angew.
by the first author of this article. The concept of the method Math. Mech., 90(10-11), pp. 847 – 860, 2010.
was proposed by the second author, the third author is [4] Chen T., Chen H.: Universal Approximation to Non-linear
the creator of the original FEM code, which, thanks to the Operators by Neural Networks with Arbitrary Activation
Functions and Its Application to Dynamical Systems, IEEE Trans.
attached scripts, can be used as a training data generator
on Neural Networks, Vol.6, No 4, 911-917, 1995.
[36]. [5] Favre, H.: Sur une loi régissant la déformation d’un sol
horizontal sous l’action d’une charge répartie à la surface.
Comptes Rendus, Acad. Sci., 252, 1961, 2988–2990.

5 Conclusions
[6] Ghavanloo, E., Daneshmand, F. and Rafiei, M.: Vibration and
instability analysis of carbon nanotubes conveying fluid and
resting on a linear viscoelastic Winkler foundation, Physica E,
In this article, the new method of identification of the vol. 42, no. 9, pp. 2218–2224, 2010.
parameters of reduced model has been established and [7] Gradkowski K.: Nośność podłoży nawierzchni dróg
samochodowych wzmacnianych geosyntetykami.
exemplified for two classical reduced models of soil–
Badania doświadczalne, Oficyna Wydawnicza Politechniki
structure interactions in soil mechanics: model of Winkler Warszawskiej, Warszawa, 2008
and model of Pasternak. [8] Hertz, J., A. Krogh., A., Palmer G.R.: Introduction to The
The method assures that the parameters of the reduced Theory of Neural Computation, Lecture Notes Volume I, Santa
model permit to generate the displacement (solutions of Fe Institute Studies in the sciences of Complexity, Addison-
Wesley, 1991.
the reduced model) are very close to the solution of the
[9] Hornik K., Stinchcombe M., White H.: Universal approximation
reference model in selected points.
of an unknown mapping and its derivatives using multilayered
The method is fully automatic, the correspondence feedforward networks, Neural Networks, vol. 3, 551-560, 1990.
between two models (initial and reduced) is assured [10] Hou Z.Q. and Lucifredi, A.: On the Use of Neural Networks for
without resorting to speculative theoretical consideration Identification of Linear and Nonlinear Systems, Meccanica 30,
related to the formulation of the reduced model. 377-388, 1995.
[11] Kerr, A.D.: Elastic and viscoelastic foundation models, J. Appl.
The resulting artificial neural network ANN_3 (see
Mech., 31, 1964, pp.491-498.
(3)) can be treated as a symbolic formula. Since its explicit [12] Kerr, A.D.: On the formal development of elastic foundation
form is too complex to be printed, the Excel formula that models, Ing. Arch, 54, 1984, pp. 455–464. [Link]
simulates the action of ANN_3 in recall mode will be soon org/10.1007/BF00537376
available (please mail to the first author). [13] Lefik M.: Przykłady rozwiązań zagadnienia odwrotnego metodą
aproksymacji relacji odwrotnej, Wydawnictwo Politechniki
The authors continue their work in order to apply the
Łódzkiej, 2016.
procedure presented in this article to simplified nonlinear
[14] Lefik M.: Rozwiązanie zagadnienia odwrotnego przy pomocy
models that allow modeling materials behavior in limit sztucznych sieci neuronowych. Zastosowanie metod
states. numerycznych w analizie wybranych zagadnień geotechniki
i geotechniki środowiska. Wydawnictwo Politechniki Łódzkiej,
2021.
[15] Lefik M. and Boso D. P.: Identification of parameters of
Ancknowlegment adsorption by approximation of inverse relation and using
artificial neural networks, Proceedings Solmech2016
Conference [Link] /Abstracts/0177.
This paper has been completed while the first author was
pdf
the Doctoral Candidate in the Interdisciplinary Doctoral [16] Lefik M. and Boso D. P.: Inverse problem - soft solution, in
School at the Lodz University of Technology Bytes and Science, edited by G. Zavarise and D. P. Boso (©
CIMNE, Barcelona, Spain 2012), pp. 10-34.
[17] M. Lefik, M. Wojciechowski, M.: Examples of inverse problem
in geotechnics: comparative analysis of various strategies of
 Reduction of Numerical Model in Some Geotechnical Problems 349

solutions involving artificial neural networks, International Approach, Journal of Applied Mathematics, Article ID 626287,
Symposium on Geomechanics and Applications for Sustainable p. 13, 2013, [Link]
Development, SIPS2019, Cyprus. October 2019 [27] Szeptyński, P.: Praktyczne zagadnienia interakcji budowli z
[18] Levinson, M.: On the linear, isotropic, elastic foundation of podłożem, Inżynieria i Budownictwo 72, nr 7 ss. 384-389, 2016.
grade 4. Mech. Res. Com. 6, 1979, pp. 369–378. [28] Szydło A.: Nawierzchnie drogowe z betonu cementowego.
[19] Meier R. W.: Backcalculation of Flexible Pavement Moduli Wydawnictwo Polski Cement, Kraków 2004.
from Falling Weight Deflectometer Data Using Artificial Neural [29] Świniarski J.: Ocena podatności podłoża przy wymiarowaniu
Networks, Report prepared for U.S. Army Corps of Engineers, płyt fundamentowych. Inżynieria Morska i Geotechnika, nr
technical report GL-95-3, Washington DC, 1995. 5/2003.
[20] Meyer, Z.: Propozycja określania współczynnika podatności [30] Tanahashi, H.: Two-dimensional analysis of ground surface
podłoża przy projektowaniu płyt fundamentowych, displacements using Pasternak Model, J. Struct. Constr. Engrg.,
[Link] AIJ, 530, 2000, pp.85-91
wspolczynnika-podatnosci-podloza-przy-projektowaniu-plyt- [31] Terzaghi K.: Evaluation of coefficients of subrade reaction, Inst.
[Link] of Civil. Eng., London 1956;
[21] Nita P.: Budowa i utrzymanie nawierzchni lotniskowych. WKŁ. [32] Vallabhan, C.V.G., and Das, Y.C.: Parametric study of beams on
Warszawa 2008 elastic foundations, J. Engrg. Mech., ASCE, 114(12), pp.2072-
[22] Pasternak, P.L.: On a new method of analysis of an elastic 2082, 1988
foundation by means of two foundation constants, [33] Vlasov, V.Z. and Leont’ev, N.N.: Beams, plates and shells on
Gosudarstvennoe Izdatelstvo Literaturi po Stroitelstuve i elastic foundations, Israel Program for Scientific Translations,
Arkhitekture, Moscow, 1954, (in Russian). Jerusalem (Translated from Russian; original Russian version
[23] Reissner, E.: A note on deflections of plates on a viscoelastic published in 1960). [Link]
foundation, J. Appl. Mech., ASME, 25, 1958, pp.144-145. techdoc_19670004909/page/n43/mode/2up.
[24] Ruta P., Krawczyk B., Szydło A.: Identification of pavement [34] Winkler, E.: Die Lehre von der Elastizität und Festigkeit, H.
elastic moduli by means of impact test, Engineering Structures, Dominicus, Prague, 1867 (in German).
100, 2015, ss. 201-211. [35] Wojciechowski, M.: Posadowienie na płycie fundamentowej –
[25] Sadrekarimi J., Akbarzad M.: Comparative study of methods ustalanie współczynniów podatności podłoża, Wydawnictwo
of determination of coefficient of subgrade reaction. The Politechniki Łódzkiej, 2016.
Electronic Journal of Geotechnical Engineering, tom 14/2009, [36] Wojciechowski M.: Fempy - finite element method in python,
zeszyt E, 2009. [Link] [Link] (access:
[26] Suchart Limkatanyu, Woraphot Prachasaree, Nattapong 15, 03, 2021).
Damrongwiriyan, Minho Kwon, andWooyoung Jung.: Exact [37] Wojciechowski, M., Lefik, M., D. Boso.: Inverse problems
Stiffness for Beams on Kerr-Type Foundation: The Virtual Force in the light of homogenisation methods: identification