Page 1 of 8
DOI: 10.14738/aivp.92.10131
Publication Date: 25
th April, 2021
URL: http://dx.doi.org/10.14738/aivp.92.10131
Machine Learning Solutions of Stress Concentration
Factors in Perforated Plate With Single Circular Hole
1Yafeng Chang, 2Hui Wang
1College of Civil Engineering, Henan University of Technology, Zhengzhou, China
2School of Civil Engineering and Architecture, Hainan University, Haikou, China
ABSTRACT
The discontinuities such as holes, grooves, notches and fillets in the structural
geometry in structures would lead to significant increase of stress level, which can
be represented by stress concentration factor (SCF), and further influence the
strength of structures. Therefore, SCF plays a vital role in quantitatively
understanding the influence of discontinuities on the peak stress. However,
analytical or empirical solutions for predicting localized high-stress around the
discontinuities are usually difficult to be derived, owing to the complex interaction
of the specific boundary conditions and the discontinuities. In this work, machine
learning (ML) solutions of SCF of circular holes in a finite perforated plate under
tension are modeled using finite element analysis and back propagation neutral
network (BPNN) technique. The locations and sizes of circular holes are input as
input variables, and the SCFs are target variable. The feasibility and accuracy of the
model are demonstrated through numerical examples. It’s found that the developed
ML approach base on BPNN technique can provide accurate prediction of SCF value
for each set of structural parameters, and vice versa.
Keywords: Machine learning, neutral network, stress concentration factor, circular
hole.
INTRODUCTION
Considering the fact that almost all engineering structures involve interruptions in
section and/or shape, i.e. cracks on aircraft wings, grooves or shoulders on shafts,
oil holes, screw threads, shape reentrant corners of auxetics, etc. Typically, these
interruptions arise from discontinuities of geometry and may cause a local increase
in stress in the engineering structure, where a high stress gradient exists [1-3]. Such
stress rise is known as stress concentration normally. Obviously, the stress
concentration may result in the part failure in practice, and further affects the
material or structure designing. Therefore, stress concentration analysis plays an
important role, especially in designing the structures under alternating loads.
For engineers, a simple and reliable solution can bring great convenience to
engineering applications. However, so far, most of the existing estimators for stress
concentration factor (SCF) near a circular hole in a finite plate are some empirical
formulas, which are provided by curve fitting technology based on the experimental
measurement (usually using photoelasticity) involving many contributors through
several years [4]. Moreover, these empirical solutions are usually not reliable for the
extreme cases that the circular hole diameter is large compared to the dimensions
of finite plate. As an alternative to the empirical solutions, the finite element
solutions are popularly resorted, due to the advantages of easy implementation of
programing and strong applicability to complex problems [5, 6]. However, the
Page 2 of 8
344
Chang, Y., & Wang, H. (2021). Machine Learning Solutions of Stress Concentration Factors in Perforated Plate With
Single Circular Hole. European Journal of Applied Sciences, 9(2). 343-350.
URL: http://dx.doi.org/10.14738/aivp.92.10131
efficiency of numerical solutions is doubtable and you have to conduct simulation
for each structural configuration and each loading condition. Even today, it is
difficult to determine the exact value of SCF for most extreme cases. Therefore, a
practical requirement seeking reliable solutions is naturally raised.
Machine learning (ML) solutions are a promising way to achieve this purpose [7-
11]. It has been reported that the ML solutions behave not only ease of
implementation but also low computational cost for some complex engineering
problems, i.e. fracture mechanics problems of the pentagonal cross-section
microcantilever at nanoscale [9], stress hotspots in polycrystalline materials under
uniaxial tensile deformation [12], stress concentration depiction at the low points
of rough surfaces by interpretable convolutional neural network [13]. More
recently, Ozkan and Erdemir developed nonlinear ML models based on neural
network (also known as artificial neural network, ANN) for predicting SCF for
circular/elliptical holes in an infinite plate [11]. Interestingly, in these successful
applications, the ML process thoroughly avoids the complex mathematical formulas
related to the controlling parameters and the desired mechanical response.
Taking inspiration from these successful applications of ML, in this context, the
stress concentration in a finite plate perforated with a circular hole is investigated
by using BPNN based ML technology. The size and position of the circular hole can
be arbitrary. Thus, from the view of data analysis, the input in the BPNN framework
is the geometrical information (e.g., radius of hole, location coordinates) and the
output is the results of stress concentration factor. To drive the BPNN algorithm, a
dataset including the geometric parameters of the problem and the corresponding
stress response is synthetically generated from finite element simulation. After that,
we used genetic algorithm to continuously optimize and find the maximum stress
from all the predicted data.
PROBLEM DESCRIPTION
The peak stress in a finite plate perforated by a circular hole located at an arbitrary
position is investigated in this study, as shown in Fig. 1. The plate is subject to a
uniform tension with a specific tension force σ0 along the x direction. The solid
material is isotropic and elastic.
Fig 1. Schematic diagram of finite plate with circular perforation subject to
unidirectional tension
Page 3 of 8
345
European Journal of Applied Sciences, Volume 9 No. 2, April 2021
Services for Science and Education, United Kingdom
To characterize the peak stress better, the SCF concept is normally introduced to
perform both design and analysis of the loaded members with geometric
discontinuities. Here the SCF for the problem under consideration is defined as the
ratio of the peak stress to the normal stress at the circular hole,
Ks =
σmax
σnom
σmax
σ0
(1)
where Ks denotes the SCF, σmax is the unknown peak stress, and σnom is the
reference nominal stress, which usually corresponds to the case without hole.
Obviously, the value of SCF for isotropic material is independent of material
properties and is strongly relevant to the geometries of the structure, including the
size of circular hole, the location of circular hole in the plate. Thus, Ks
is a function
of L1
, L2
, W1
, W2 and R
Ks = Ks
(L1
, L2
,W1
,W2
,R) (2)
Traditional theoretical methods to compute the SCF of the structure involving
multiple parameters are usually very complex. In order to solve this problem and
further improve the predictive efficiency, the nonlinear machine learning model
with BPNN algorithm is established to predict the SCF.
BACK-PROPAGATION NEURAL NETWORK
Generally, a typical BPNN model composes of three layers [14, 15]. The first layer is
input layer containing inputs (also known as predictors), and the last layer is output
layer containing target outputs. Between them are one or more hidden layers which
are interconnected between neuron nodes which simulate the biological nervous
system of human brain to process information, as shown in Fig. 2. The input layer
takes in input information and then transfer them throughout the middle neuron
network layer in which nonlinear activation functions are employed to capture the
nonlinear relationship in the data set. Theoretically, the more the hidden layers and
hidden neurons, the more capabilities a network can model a complex and nonlinear
system.