minerals
Article
Prediction of Uniaxial Compression Strength of Limestone
Based on the Point Load Strength and SVM Model
Shaoqian Li 1 , Yu Wang 2, * and Xuebin Xie 1, *






Citation: Li, S.; Wang, Y.; Xie, X.
Prediction of Uniaxial Compression
Strength of Limestone Based on the
Point Load Strength and SVM Model.
Minerals 2021, 11, 1387. https://
doi.org/10.3390/min11121387
Academic Editor: Gianvito Scaringi
Received: 15 November 2021
Accepted: 6 December 2021
Published: 8 December 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Minerals 2021, 11, 1387. https://doi.org/10.3390/min11121387
1 School of Resources and Safety Engineering, Central South University, Changsha 410083, China;
tushan@csu.edu.cn
2 School of Civil Engineering, Beijing Jiaotong University, Beijing 100044, China
* Correspondence: yu_wang@bjtu.edu.cn (Y.W.); xbxie@csu.edu.cn (X.X.)
Abstract: Uniaxial compression strength (UCS) is a fundamental parameter to carry out geotechnical
engineering design and construction. It is simple and efficient to predict UCS using point load
strength (PLS) at engineering sites. However, the high dispersion of rock strength limits the accuracy
of traditional fitting prediction methods. In order to improve the UCS prediction accuracy, 30 sets
of regular cylindrical specimen tests between PLS and UCS are conducted on limestone mines.
The correlation relationship between PLS and UCS is found by using four basic fitting functions.
Then, a prediction model is established by using SVM algorithm. Multiple training test data are
used to achieve high-precision prediction of UCS and the results show it is less different from the
actual values. Especially, the R2 coefficient reached 0.98. The SVM model prediction performance is
significantly better than the traditional fitting function. The constructed SVM model in this study can
accurately predict the UCS using the PLS obtained in the field, which has a great significance to the
rock stability judgment in the actual construction environment.
Keywords: point load strength; uniaxial compression strength; correlation; support vector machine
1. Introduction
Construction projects such as water and hydropower, transportation, mining, indus-
trial and civil construction engineering all involve rock engineering. In order to accurately
and promptly assess the stability of the mountains involved in the construction project and
ensure the safety of the structure, rock classification methods are often used in the rock
engineering project and design. To evaluate the stability of engineering rock mass, the rock
classification divides engineering rock mass into several levels with different basic quality
and strength. In this way, it provides an important basis and safety protection selection for
the technical planning and various construction engineering [1].
Quantitative indicators of strength in rock classification are generally expressed in
terms of uniaxial compression strength of rocks. However, the determination of the
uniaxial compression strength usually needs to be measured in the laboratory, which
cannot meet the requirements of obtaining data quickly at the engineering site [2]. In
1972, Broch and Franklin [3] gave the first point load strength formula by studying the
damage model analysis of cylindrical specimens analyzed along the diameter direction and
found that the uniaxial compression strength values of certain rocks were approximately
24 times higher than those of rock specimens tested with a 50 mm diameter point load. This
discovery provided an alternative method of simple operation and high accuracy for testing
the uniaxial compression strength on the engineering site. Compared with the uniaxial
compression strength, testing the point load strength of the rock has the advantages of
convenience and efficiency, and the strength index of the rock can be obtained in a shorter
time period. This makes the point load strength very important in engineering practice. It
is widely used in practical scenarios such as tunnel engineering, slope stability analysis,
https://www.mdpi.com/journal/minerals
Minerals 2021, 11, 1387 2 of 14

mining roadway support, etc [4,5]. Subsequently, many scholars and scientific research
institutions have launched a large number of related research.
Bieniawski [6] stated that the uniaxial compression strength of rocks is about 23 times
the point load strength. The International Society of Rock Mechanics [7] and the American
Society for Testing Materials [8] proposed a general test method and procedure for point
load strength and gave a general formula R = (20~25) Is(50) between them in 1985. Tsi-
ambaos and Sabatakakis [9] studied the ratio of uniaxial compression strength to point load
strength for intact sedimentary rocks. The test results showed that this ratio varied with
different point load strength distributions. For example, different ratios existed from soft
sedimentary rocks with point load strength less than 2 MPa to hard rocks with point load
strength more than 5 MPa. A. Basu and M. Kamran [10] have explored the feasibility of
point load tests to evaluate anisotropic rocks uniaxial compression strength such as schists
based on successful estimation of uniaxial compression strength by point load tests on
isotropic rocks. Through evaluating uniaxial compression strength of these rocks, relevant
fitting formulas have been obtained with good prediction results. Robina H. C. Wong [11]
investigated the correlation between point load strength and uniaxial compression strength
of irregular volcanic rock masses from 16 different locations and obtained a good linear fit
based on the analysis of all test data. Accordingly, they found a suitable size correction
factor by least squares method.
Based on the summary of a large number of literatures, these research results are
arranged in the table by year, as shown in Table 1. It can be identified that many
scholars [12–27] have found the correlation between point load strength and uniaxial
compression strength by experimental methods. However, the data processing of these
research results is carried out by function fitting, and there are large deviations in predicting
the test points that deviate from the data center. Moreover, it can be found from Table 1
that most of the derived correlation formulas are linear, but some of them are nonlinear
functions such as exponential, logarithmic, and power functions. These results also imply
that there exists a nonlinear relationship of limestone between the point load strength and
the uniaxial compression strength [14,26]. This shows that large deviation will be produced
if the general linear formulas are simply used to perform the uniaxial compression strength
estimation. So linear relationship cannot be blindly extended to all engineering scenarios
for undeveloped projects [28,29].

Table 1. Statistical table of correlation between PLS and UCS (arranged by year).

NO. Year Authors Rock Type Equations R2

1 1985 ISRM [7] granite, marble et al. R = (20~25) Is(50) -
2 1989 Vallejo et al. [12] granite R = 12.6 Is(50) R = 17.4 Is(50) -
3 2003 Quane, S.L., Russell, J.K. [13] welded ignimbrite R = 3.86 (Is(50) )2 + 5.65 Is(50) -
4 2004 Tsiambaos et al. [9] limestone R = 7.3 (Is(50) )1.71 0.90
5 2008 KiliÇ, A., Teymen, A. [14] limestone, tuff et al. R = 100 ln(Is(50) ) + 13.9 0.99
6 2008 Sabatakakis, N. et al. [15] limestone, sandstone R = 28 Is(50) 0.73
7 2009 Yilmaz, Yuksek [16] sandstone, conglomerate R = 12.4 Is(50) − 9.08 0.85
8 2009 Diamantis, K, et al. [17] serpentinites R = 16.45 e0.39Is(50) 0.89
9 2010 Basu, A. et al. [10] apatite R = 11.103 Is(50) + 37.66 0.86
10 2012 Kohno, M., Maeda, H. [4] volcaniclastic, tuff R = 16.4 Is(50) 0.90
11 2013 LI.D, Wong, L. [18] metasandstone R = 21.27 Is(50) -
12 2013 Mishra, D.A., Basu, A. [19] granite, schist, sandstone R = 14.63 Is(50) 0.94
R = 7.201 Is(50) + 14.074 (Axial) 0.99
13 2013 FUZhiliang, et al. [20] sandstone, mudstone
R = 13.938 Is(50) + 17.529 (Diametral) 0.97
14 2014 Kahraman, S. [21] ignimbrite R = 7.73 (Is(50) )1.25 0.91
15 2014 LI anping et al. [22] sandstone, mudstone R = 14.933 ln(Is(50) ) + 89.012 0.79
16 2015 ZHANG Jianming et al. [23] granite, andesite R = −0.66 (Is(50) )2 + 21.15 Is(50) 0.94
17 2017 Jian-Hua Yin et al. [24] granite R = 21.605 Is(50) 0.74
Minerals 2021, 11, 1387 3 of 14

Table 1. Cont.

NO. Year Authors Rock Type Equations R2

18 2017 Robina H.C. Wong et al. [11] granite, tuff R = 18.897 Is (50) 0.87
19 2018 LIN jun et al. [25] cement block R = 29.87 Is(50) − 2.242 0.99
R = 22.72 (Is(50) )0.82 (Irregular)
20 2018 CHEN Jiaqi, WEI Zuoan [26] limestone, sandstone 0.86
R = 26.24 (Is(50) )0.72 (Regular)
21 2019 LI Hongpeng et al. [27] marble R = 21.28 Is(50) -
22 2020 SHA Peng et al. [2] igneous R = 0.14 (Is(50) )2 + 13.25 Is(50) 0.93

The shortcomings shown above indicate that the currently derived conversion formu-
las still have large errors for engineering applications, and some of them even fail to meet
the engineering accuracy requirements for direct application in actual engineering. In view
of this, it is necessary to further study and establish the correlation model between rock
uniaxial compression strength and point load strength to improve the prediction accuracy
and calculation.
With the continuous progress in computational science, techniques such as machine
learning and artificial intelligence, many scholars have applied these to the field of civil
and mining engineering to solve many engineering challenges encountered. For example,
Yilmaz and Yuksek [16] developed a high-performance adaptive neural network model
for predicting gypsum strength and elastic modulus to reduce rock engineering project
uncertainty. Ruhul Amin Mozumder [30] used an artificial neural network model to analyze
the effect of different geopolymer ratios on uniaxial compression strength and predicted
uniaxial compression strength by this model. The usage of these methods improved the
prediction model reliability and accuracy to a greater extent and were more accurate in
predicting discrete points that deviate from the data center region compared to traditional
fitting methods.
Based on these research experiences, a prediction model based on machine learning
algorithm can be developed in conjunction with the correlation relationship existing be-
tween point load strength and uniaxial compression strength, which can further enhance
the prediction accuracy. In this study, 30 groups of corresponding point load tests and
uniaxial compression strength tests were conducted, and four basic function types were
used to fit and regress the data in order to investigate the correlation relationship between
uniaxial compression strength and point load strength. Then a computational model
with high accuracy was established for predicting uniaxial compression strength based on
support vector machine algorithm. The model can be used to quickly determine the rock
strength level at engineering sites, which provides a reference for construction design and
safe production.

2. Methods and Materials

2.1. Rock Sampling and Fabrication
Point load strength tests can be performed on all types of rocks in nature, and test
specimens are mainly divided into regular and irregular blocks according to the specimen’s
integrity. In general, the use of irregular specimens makes the field operation procedure
more convenient, but the accuracy of this method is relatively low, so regular blocks for
point load strength parameter determination is more in line with the practical needs [31–33].
In order to improve accuracy of the point load strength index, this study uses regular
cylindrical specimens for testing and analysis.
As we all know, rock is a kind of non-uniform material, and there are pores and
micro cracks in it. These factors will affect the macro mechanical strength of rock such
as UCS and PLS [34], so complete specifications and no obvious fissures rocks were
selected when collecting. The rock samples in this study were obtained from the limestone
mine of Sichuan Red Lion Cement in China. The main component of the rock is calcium
carbonate, which locally contains 10%~25% argillaceous limestone sand debris. The main
Minerals 2021, 11, x FOR PEER REVIEW 4 of 15

Minerals 2021, 11, 1387 4 of 14

which locally contains 10~25% argillaceous limestone sand debris. The main density dis-
tribution of limestone in the mine is 2.65~2.83 g/cm3. The nature of the rock is limestone
density distribution 3 . The nature of the rock is
with uniform block of limestone
size, in the mine
good integrity and isregular
2.65~2.83 g/cm
shape, which can meet the test re-
limestone with
quirements. Theuniform block size,
rock specimens weregood integrity
prepared onand
thisregular
basis. shape, which can meet the
test requirements. The rock
In order to prepare test specimens
specimens,were holesprepared on this
were drilled basis.
in the collected rocks to obtain
In order to prepare test specimens, holes were
initial specimens with a 50 mm diameter, which were marked. Aboutdrilled in the collected
4 torocks to obtain
6 initial speci-
initial specimens with a 50 mm diameter, which were marked. About 4
mens were obtained on a single rock. Subsequently, these initial specimens were cut and to 6 initial specimens
were obtained
polished on a single
to produce UCSrock.and PLSSubsequently, these initial
rock specimens specimens
with smooth andwere
flat cut and polished
surfaces and no
to produce
obvious UCS and
fissures. PLS rock
Ultimately, specimens
the with smoothstrength
uniaxial compression and flat test
surfaces and nowere
specimens obvious
pre-
fissures.
pared as Ultimately, the uniaxial
standard cylindrical compression
specimens strength
with 100 test specimens
mm height and 50 mm were preparedThe
diameter. as
standard cylindrical specimens with 100 mm height and 50 mm diameter.
point load test specimens were cylindrical in overall shape with 50 mm diameter, the di- The point
load testloading
ametral specimens were cylindrical
specimens’ height wasin overall
70 mm,shape withaxial
and the 50 mm diameter,
loading the diametral
specimen’s height
loading
was 30 mm.specimens’ height was 70 mm, and the axial loading specimen’s height was 30 mm.
The rock
The rock specimens
specimens and and equipment
equipment usedused in in the
the test
test are
are shown
shown in in Figure
Figure 1. In this
1. In this
study, 30
study, specimens were
30 specimens separately prepared
were separately prepared forfor axially
axially loaded
loaded point load test,
point load test, diametral
diametral
loaded point
loaded point load
load test
test and
and uniaxial
uniaxialcompression
compressiontest,test,totaling
totaling9090rock
rockspecimens.
specimens.The Thewater
wa-
content of the specimens was natural and the tests were run at room
ter content of the specimens was natural and the tests were run at room temperature. temperature.

Figure 1. The
Figure 1. The main
main test process, in
test process, in which
which (a)
(a) is
is the
the diametral
diametral point
point load
load test
test specimen,
specimen, (b)
(b)is
is the
the uniaxial
uniaxial compression
compression test
test
specimen, and (c) is the axial point load test specimen. （The test equipment mainly adopts HCT-206A rock true triaxial
specimen, and (c) is the axial point load test specimen. (The test equipment mainly adopts HCT-206A rock true triaxial
compression tester produced by Shenzhen Wance Test Equipment Co., Ltd., Shenzhen,, China and STDZ-3 digital display
compression tester produced by Shenzhen Wance Test Equipment Co., Ltd., Shenzhen, China and STDZ-3 digital display
point load tester produced by Zhejiang Geotechnical Instrument Manufacturing Co., Ltd., Zhejiang China.)
point load tester produced by Zhejiang Geotechnical Instrument Manufacturing Co., Ltd., Zhejiang, China).
2.2. Point Load Strength Test
2.2. Point Load Strength Test
Point load test applies a concentrated load at the tip of the rock specimen until the
Pointisload
specimen test applies
destroyed, and athe
concentrated load at the
point load strength of tip
theof the rock can
specimen specimen until the
be obtained by
specimen is destroyed, and the point load strength of the specimen can be obtained
calculation. The larger the concentrated load on the specimen, the greater the point load by
calculation.
strength The larger
obtained. the concentrated
The concentrated loadmethods
loading on the specimen, the greater
mainly include the point
diametral load
and axial,
strength obtained. The concentrated loading methods mainly include diametral
from which the diametral point load strength (DPLS) and axial point load strengths and axial,
from which
(APLS) the diametral
are obtained, bothpoint loadhave
of which strength (DPLS) and
comparable axial point
reliability and load
samestrengths (APLS)
applied [35,36].
are obtained, both of which have comparable reliability and same applied [35,36].
The test method procedure in common use today is mainly the recommended
The test method procedure in common use today is mainly the recommended method
method for point load strength testing proposed by the International Society of Rock Me-
for point load strength testing proposed by the International Society of Rock Mechanics in
chanics in 1985. The ISRM method uses the concept of equivalent diameter De to calculate
1985. The ISRM method uses the concept of equivalent diameter De to calculate the point
the point load strength. Moreover, to make the resultant data more comparable, the cal-
load strength. Moreover, to make the resultant data more comparable, the calculated raw
culated raw Is is generally converted to the point load strength Is(50) at De = 50 mm as
Is is generally converted to the point load strength Is (50) at De = 50 mm as follows:
follows:
P
Is = P (1)
Is De2 2 (1)
De
Minerals 2021, 11, 1387 5 of 14

In Formula (1), P represents the point load, De is the equivalent core diameter (mm).

Is(50) = FIs (2)

In Formula (2), F represents the correction factor.

 m
De
F= (3)
50

In Formula (3), m is the correction index, generally desirable 0.40~0.45, or according

to the value of similar rocks to determine the experience.
In this study, point load strength tests were respectively conducted using both di-
ametral loading and axial loading. When the diametral test was used, the equivalent core
diameter was calculated according to the following equation:

De2 = D2 (4)

In Formula (4), D is the loading point spacing (mm). When the axial test is used, the
equivalent core diameter is calculated according to Formula (5):

4WD
De2 = (5)
π
In Formula (5), W is the width or the smallest section average width through the two
loading points (mm).
The test procedure was carried out according to ISRM. During the test, specimens
were placed between the upper and lower loading cone heads of the point load strength
tester, and the lower cone head was raised by cranking the hand crank oil pump to maintain
the loading speed between 0.05~0.1 MPa. The damage occurred within 10~60 s. The rock
specimen is damaged between the two loading heads, forming a damage surface through
the upper and lower loading points. A typical damage diagram is shown in Figure 1.

2.3. Uniaxial Compression Strength Test

Uniaxial compression strength is the main index to evaluate rock stability and rock
classification. Put the rock sample between the upper and lower pressing plates of the
press. When the sample is crushed, its pressure strength value can be measured, which is
the uniaxial compression strength. The greater pressure the rock can bear, the higher the
uniaxial compression strength is [37].
Uniaxial compression strength test also refers to the ISRM standard. A total of 30 Rock
standard specimens are used in the test. The specimens are tested on the HCT-206A rock
true triaxial compression testing machine (Figure 1) and loaded to failure at the speed of
0.6 MPa/s. The ultimate stress value on the compression surface at the moment of failure is
the uniaxial compression strength of the rock, and the specimens will undergo monotonic
failure under axial load. The uniaxial compression strength is calculated according to
Formula (6):
P
R= (6)
A
In Formula (6), R is the uniaxial compression strength of the rock, MPa; P is the peak
breaking load, N; A is the initial cross-sectional area of the specimen, mm2 .

2.4. Data Analysis and Inspection

In the point load strength test and uniaxial compression strength test, data distortions
caused by improper test operations often occur. In this study, such apparently abnormal
data were eliminated, and Grubbs’ criterion was used to test whether there were suspicious
Minerals 2021, 11, x FOR PEER REVIEW 6 of 15

Minerals 2021, 11, 1387 6 of 14

data were eliminated, and Grubbs’ criterion was used to test whether there were suspi-
cious data, and when the suspicious data satisfied Formula (7), they should be eliminated
from the group.
data, and when the suspicious data satisfied Xp  X  G p ( n(7),
Formula )S
they should be eliminated from(7)
the group.  
In Formula (7), Xp is the data being X p −tested,
X  > GXp(nis
) Sthe average value, S is the standard(7)
deviation, Gp(n) is the test threshold, where p is the confidence probability, taken as 0.95,
and nInisFormula
the number Xpdata,
(7), of is thetaken
data as 30. tested, X is the average value, S is the standard
being
deviation, Gp(n) is the test threshold,
After rigorous calculations, the Grubbs where ptest
is the confidence
value probability,
of diametral takenstrength
point load as 0.95,
and nisis1.97,
data the the
number
Grubbsof data, takenofasaxial
test value 30. loading point load strength data is 2.77, and the
After rigorous calculations, the Grubbs test value of diametral point load strength
Grubbs test value of uniaxial compression strength test data is 2.15, which has no abnor-
data is 1.97, the Grubbs test value of axial loading point load strength data is 2.77, and
mal value at 5% significance level and satisfies the condition of no abnormal value of test
the Grubbs test value of uniaxial compression strength test data is 2.15, which has no
data. The above analysis indicates that test data obtained from the experiment has high
abnormal value at 5% significance level and satisfies the condition of no abnormal value
reliability.
of test data. The above analysis indicates that test data obtained from the experiment has
high reliability.
3. Result and Analysis
3.1. Experimental
3. Result Data Results
and Analysis
3.1. Experimental
After the test Data Results to the scheme, the test data were measured in this study.
according
Then,After
we drew a Q-Q
the test chart to
according to examine
the scheme,the all
normality
the test of thewere
data data,measured
as shownininthis
Figure 2.
study,
The data
which candistribution,
be obtained mean, and standard deviation
from supplementary materials of the three
(Tables experiments
S1–S3). alsoaccord-
After the test can be
obtained
ing to theinscheme,
Figure 2. It test
the can be found
data werethat the UCSinaverage
measured value
this study. is 76.07
Then, weMPadrewand standard
a Q-Q chart
deviation
to examineisthe 42.46 MPa. The
normality DPLS
of the average
data, value
as shown in is 4.03 MPa
Figure 2. Theanddatastandard deviation
distribution, mean,is
1.22 MPa. Thedeviation
and standard APLS average
of the value is 4.13 MPa and
three experiments alsostandard deviation
can be obtained is 0.99 MPa.
in Figure 2. It can be
found It that
can betheseen
UCSfrom the figure
average that
value is the MPa
76.07 threeand
experiments data are inisaccordance
standard deviation 42.46 MPa.with
The
DPLS
the average
normal value is 4.03
distribution law MPa
within and
thestandard
confidencedeviation
intervalisof1.22
95%MPa.
level.The
TheAPLS average
sample aver-
value
age canis well
4.13 MPa and standard
represent the overalldeviation is 0.99
data level and MPa.
data concentration degree [38].
mu=76.07 sigma=42.46 mu=4.03 sigma=1.22 mu=4.13 sigma=0.99
200 8
Test data Test data Tese data
Normality reference line Normality reference line Normality reference line
6
150
Expected normal value (MPa)
Expected normal value (MPa)

Expected normal value (MPa)

100

4 4

50

2
0
2

-50 0
-50 0 50 100 150 200 2 4 6 2 4 6
UCS (MPa) DPLS (MPa) APLS (MPa)

(a) (b) (c)

Figure 2.
Figure 2. Q-Q
Q-Q diagram
diagram for
for normality test of
normality test of test
test data,
data,(a) UCS, (b)
(a) UCS, (b) DPLS,
DPLS, (c)
(c) APLS.
APLS.

It can be seen
3.2. Discussion from the figure that the three experiments data are in accordance with
and Analysis
the normal distribution law within
Firstly, the relationship the APLS
between confidence interval
and UCS of 95%As
is found. level.
shownThein
sample
Figureaverage
3a, the
can well represent the overall data level and data concentration degree [38].
correlation coefficient by linear fitting is 0.8883, logarithmic fitting is 0.7768, exponential
fitting is 0.9485, and power function fitting is 0.9485. Then four relations between the
3.2. Discussion and Analysis
APLS and UCS are obtained, as shown in Figure 3b. The correlation coefficient of linear
Firstly,
fitting the relationship
is 0.6873, between
logarithmic fitting APLS and
is 0.5313, UCS is found.
exponential fittingAs shown and
is 0.8846, in Figure
power3a, the
func-
correlation coefficient
tion fitting is 0.7680. by linear fitting is 0.8883, logarithmic fitting is 0.7768, exponential
fitting is 0.9485, and power function fitting is 0.9485. Then four relations between the APLS
and UCS are obtained, as shown in Figure 3b. The correlation coefficient of linear fitting is
0.6873, logarithmic fitting is 0.5313, exponential fitting is 0.8846, and power function fitting
is 0.7680.
 180 180
Test data Test data
160 Linear fitting 160 Linear fitting
Power fitting Power fitting
Minerals 2021,140
11, 1387 Logarithmic fitting 140 7 of 14
Minerals 2021, 11, x FOR PEER REVIEW Logarithmic fitting 7 of 15
Exponential fitting Exponential fitting
120 120
y=111.08ln(x)-73.121 R2=0.7768
UCS (MPa)

UCS (MPa)
y=99.338ln(x)-60.918 R2=0.5313
100 100
180 180
y=32.88x-56.404
Test data R2=0.8883
80 80 Test data
160 Linear fitting 160 Linear fitting
Power fitting y=35.704x-71.307 R2=0.6873
60 60 Power fitting
140 Logarithmic fitting 140 Logarithmic fitting
Exponential fitting y=6.8668x1.6741 R2=0.9289 Exponential fitting
40 40
120 120
0.4697x y=6.7667x1.6412 R2=0.7680
y=111.08ln(x)-73.121 R2=0.7768y=9.8043e R2=0.9485
UCS (MPa)

UCS (MPa)
20 20 y=99.338ln(x)-60.918 R2=0.5313
100 100
y=6.4614e0.5595x R2=0.8846
0 y=32.88x-56.404 R2=0.8883 0
80 1 2 3 4 5 6 7 80 1 2 3 4 5 6
y=35.704x-71.307 R2=0.6873
60 DPLS (MPa) 60 APLS (MPa)
(a) y=6.8668x1.6741 R2=0.9289 (b)
40 40
1.6412
Figure 3. Fitting R2=0.7680
y=9.8043ecurve
0.4697x between
R2=0.9485 UCS and PLS for (a) is DPLS, (b) is APLS.
y=6.7667x
20 20
y=6.4614e0.5595x R2=0.8846
0 From these fitting results, the 0correlation coefficients using exponential function fit-
1 2 3 4 5 6 7
ting are greater than other function1 fitting, whether 2 3
axial 4
loading 5
or diametral 6
loading,
DPLS (MPa) APLS (MPa)
which has better fitting effect, indicating that the relationship between PLS and UCS in
(a) mine is more aptly described in the form of exponential
this (b) function.
Figure
Figure 3. According
3. Fitting
Fitting curveto
curve the exponential
between
between UCS
UCS and function
PLS
and PLS (a) relationship,
for (a)
for is DPLS,
is DPLS, (b) is the
(b) is APLS.
APLS. predicted value by using this
formula is compared with the experimental value, and the error value is presented in Fig-
ure 4.FromIt canthese
be seen
these fitting
fitting fromresults, thecorrelation
correlation
the prediction
results, the effect coefficients
of the final
coefficients using
fitting
using exponential
function function
exponential function
that if fit-
evenfitting
the
ting
are are greater
exponential
greater than than
function
otherwithother
function function
the fitting,fitting,
largest whether
correlation
whether axial loading
coefficient
axial loading isorused or strength
for
diametral diametral loading,
prediction,
loading, which
which
the better
has hasfitting
calculationbetteraccuracy
fittingindicating
effect, effect, indicating
still deviates
that thetoothat
much thefrom
relationshiprelationship
the actualbetween
between PLS PLS and
requirements,
and UCS UCS
in especially
this minein
this
is mine
inmore
the is more
aptly aptly
described
high-strength described
in the
rocks. Theform inof
maximumthe form
exponentialof exponential
deviation reachedfunction.
function. 47.53 MPa, and concentrated
in theAccording
axial point to the
load exponential
strength area. function
This relationship,
shows that the thetraditional
predicted value fittingbymethod
using thishas
formula
formula isiscompared
poor accuracy compared
and cannot withthe
with the experimental
experimental
adapt to the high value,
value, and
dispersionand
thetheerrorerror
value
disadvantages value is rock
presented
is presented
of in
in Fig-
strength
Figure
[39].4. It4.can
ure It can
be be
seen seenfromfrom thethe prediction
prediction effectofofthe
effect thefinal
finalfitting
fittingfunction
functionthat that even
even if the
Through
exponential the above
function withtests largest
and data
the largest correlation coefficient
analysis,coefficient
correlation it can be known usedthat
is used for strength
for strength prediction,
there is prediction,
an obvious
the calculation accuracy
non-linear relationship still
accuracy between deviates
still deviates too much
too much
the point from
load from the
strength actual
the ofactual requirements,
requirements,
the mine’s especially
limestone especially
and thein
the
uniaxial
in high-strength
compression
the high-strength rocks. The
strength.
rocks. maximum deviation
On this basis,
The maximum reached
the artificial
deviation 47.53 MPa,
reached intelligence
47.53 MPa, andand concentrated
algorithm in
can be
concentrated
the
in
used axial
the to point
axial
constructload
point strength
load
the strengtharea.area.
relationship This shows
This
model that that
shows
between the traditional
the Data
both. traditionalfitting
training method
fitting
can methodhas poor
improve has
the
accuracy
poor accuracy
accuracy and
of thecannot
and
model adapt
cannotto meetto the
adapt high
to thedispersion
engineering high disadvantages
dispersion
needs. of rock strength
disadvantages [39].
of rock strength
[39].
50
Through the above tests and data analysis, it can be known that there is an obvious
DPLS-UCS
non-linear APLS-UCS
relationship between the point load strength of the mine’s limestone and the
uniaxial
40 compression strength. On this basis, the artificial intelligence algorithm can be
deviation value (MPa)

used to construct the relationship model between both. Data training can improve the
accuracy
30 of the model to meet engineering needs.

20
50 DPLS-UCS
APLS-UCS
10
40
deviation value (MPa)

0
30
0 5 10 15 20 25 30
20 specimen number

Figure 4.4. The

Figure The error
error comparison
comparison chart
chart between
between the
the UCS
UCS predicted
predicted value
value and
andexperimental
experimentalvalue
value
10
generated by the exponential function formula.
generated by the exponential function formula.
0
Through the above tests and data analysis, it can be known that there is an obvious
non-linear
0 relationship
5 10 between
15 20the point
25 load
30 strength of the mine’s limestone and the
uniaxial compression specimen numberOn this basis, the artificial intelligence algorithm can be
strength.
used to4. construct
Figure The error the relationship
comparison chart model
betweenbetween
the UCS both. Data
predicted training
value can improvevalue
and experimental the
accuracy of
generated by the model to meet
the exponential engineering
function formula. needs.
Minerals 2021, 11, 1387 8 of 14

4. Support Vector Machine Model

4.1. Theoretical Basis
Support vector machine (SVM) is a classical artificial intelligence algorithm. SVM
model can theoretically approximate any nonlinear function in a global sense. It improves
the generalization ability of the model according to the principle of structural risk mini-
mization and can obtain better statistical laws with a small statistical sample size [40]. SVM
has the significant advantage of transforming the problem to a high-dimensional feature
through nonlinear transformation space, thus avoiding catastrophic data phenomena, local
miniaturization, and other problems that often occur in the prediction process [41–43].
SVM is a machine learning mechanism. The essence of machine learning is to obtain
the relationship estimation between input and output of the system from a given finite
sample [44]. Thus, it can make a more accurate prediction of the unknown output. In terms
of development, support vector machine theory is developed based on statistical learning
theory, which overcomes many shortcomings of classical statistical methods and empirical
nonlinear methods [45]. The rock test has the nature of small samples. Each strength index
has an obvious non-linear relationship, so it is very suitable to use the SVM model to solve
the data prediction problem related to rock specimens. The principle of SVM nonlinear
regression is briefly described below [46,47].
For a set of real numbers H = {(xi , yi )}, where i = 1, 2,. . . , n, it is assumed that the
functional relation y = g(x) is fitted between xi and yi , where xi is the input quantity in
this relation, yi is the output quantity, and n is the sample capacity. For this functional
relationship, the fitted relationship equation used in SVM is:

f ( x ) = ωϕ( x ) + b (8)

In Formula (8), f (x) is the fit function, ϕ(x) is the nonlinear mapping function, ω is the
weight vector, and b is the partiality constant.
Support vector machines use a structural risk minimization strategy by which ω and
b can be obtained:
n
1
Sm = C ∑ Lε (yi , f ( xi )) + kω k2 (9)
i =1
2

In Formula (9), Lε (yi , f ( xi )) is the empirical risk function, 1/2 kω k2 is the regulariza-
tion term, C is the penalty coefficient, and Lε is the insensitive loss function of ε.
Sm can prevent the decision function from over-fitting, and then continue to introduce
slack variables ξ i and ξ i∗ , so that the minimization of Sm is abbreviated as:
n
1
minSm = kω k2 + C ∑ (ξ i + ξ i∗ ) (10)
2 i =1

 yi − b[ωφ( xi )] ≤ ε + ξ i
Subject to [ωφ( xi )] + b − yi ≤ ε+ ξ i∗ (11)
ξ i , ξ i∗ ≥ 0, i = 1, · · · , n


To determine the range of ω, the solution is performed by Lagrange optimization

method, which leads to:
n
ω= ∑ (ai − ai∗ )φ(xi ) (12)
i =1

In Formula (12), ai and ai∗ are Lagrange multipliers corresponding to sample of i.

The final decision function is obtained by substituting Formula (12) into Formula (8):
n
f (x) = ∑ (ai − ai∗ )K(x, xi ) + b (13)
i =1

K ( x, xi ) = φ( xi )φ( xk ) (14)
 i1
i i i

K  x, xi     xi    xk  (14)

Minerals 2021, 11, 1387 After establishing the model, R2 and mean square error (MSE) are used to evaluate 9 of 14
the performance of established model. R2 can be used to measure how well the predicted
values fit the true values. In the ideal case, the value of R2 is 1. The more R2 converges to
1, the better the prediction performance of the model, which is calculated as in Formula
2
(15): After establishing the model, R and mean square error (MSE) are used to evaluate
the performance of established model. R2 can be used to measure how well the predicted

n
p  Ui )
values fit the true values. In the ideal case, the value (Uof R2 is 1. The more R2 converges to 1,
the better the prediction performance
2
R of 1the
 model,
i 1
(15)
which is calculated as in Formula (15):

n
i 1
(U i  U )
n
∑i=1 (U p − Ui )
R2 = 1 −value
In Formula (15), Up is the predicted compression strength, Ui is(15)
∑in=1 of
(Uuniaxial
i − U)
the
true value of uniaxial compression strength, U represents the average value of the true
In Formula
uniaxial (15), Ustrength
compression p is the predicted value
of the group of uniaxial
of tests, and ncompression capacity.Ui is the
is the samplestrength,
true value
The MSEof uniaxial
refers tocompression
the average strength, U represents
value of the differencethe average
between thevalue of thevalue
predicted true
uniaxial compression
and the actual strength
value after of the
the sum group
of the of tests,
squares. It isand n isused
often the sample capacity.
to measure the prediction
modelThe MSE refers
accuracy andtoevaluate
the average valuechange
the data of the difference
degree. The between
smallerthe predicted
the value ofvalue and
the MSE,
the actual value after the sum of the squares. It is often used to measure the
the better the model’s ability to fit experimental data. The calculation formula is as fol- prediction
model
lows: accuracy and evaluate the data change degree. The smaller the value of the MSE,
the better the model’s ability to fit experimental data. The calculation formula is as follows:
2
1 n
MSE 1 n 
 U p  U2 i  (16)
MSE = ∑i U
n 1 p − Ui (16)
n i =1

4.2. Model
4.2. Model and
and Results
Results
Based on
Based onthe
theabove
abovetheoretical
theoreticalbasis,
basis, the
the SVM
SVM method
method is used
is used to establish
to establish the the rela-
relation-
tionship model between PLS and UCS. As shown in Figure 5, the model
ship model between PLS and UCS. As shown in Figure 5, the model uses APLS and DPLS uses APLS and
DPLS
as asvectors.
input input vectors. Kernel function
Kernel function and partiality
and partiality constant bconstant b isthe
is used as used as the
hidden hidden
layer, and
ylayer, and y the
represents represents theUCS
measured measured
value.UCS value.

Figure 5.
Figure 5. Schematic
Schematic diagram
diagram of
of SVM
SVM model
model for
for predicting
predicting UCS
UCS using
using DPLS
DPLS and
andAPLS.
APLS.

The introduction
The introduction of of the
the kernel
kernel function
function makes
makes the
the SVM
SVM perfectly
perfectly solve
solve the
the problem
problem
of sample nonlinearity, so
of so that
that thetheinner
innerproduct
productoperation
operationofofthe vector
the vectorin in
thethe
high-di-
high-
mensional space
dimensional spaceisistransformed
transformedinto intothethefunction
functionof
ofthe
the vector
vector in the original space,
space, which
which
effectively
effectively handles
handles the the input
input of of the
the high-dimensional
high-dimensional vector
vector and
and can
can directly
directly affect
affect the
the
SVM model performance. At present, the kernel functions commonly used in SVM models
include linear kernel function, polynomial kernel function, Gaussian kernel function,
Laplacian kernel function, and Sigmoid kernel function. Among them, the Gaussian kernel
function has strong learning ability and can be used in the absence of research samples. A
good predictive effect is obtained when the prior knowledge is obtained [48]. The SVM
constructed in this study uses a Gaussian kernel function, and the specific equation is
shown in Formula (17), where γ is the kernel parameter of the kernel function [49]:
 
K ( xi , xk ) = exp −γk xi − xk k2 (17)

The SVM is used to establish the uniaxial compression strength prediction model
M1, as shown in Figure 6. The input vector currently is the APLS and the corresponding
 Minerals 2021, 11, 1387 10 of 14

DPLS, and the output vector is the UCS. After sorting out the experimental data, randomly
nerals 2021, 11, x FOR PEER REVIEW 11 of 15
divide 25 groups of data into the training set, and the remaining 5 groups of data into the
prediction set. Then run the model.

MSE=62.66 R2=0.97 MSE=50.05 R2=0.98

180 True value 180 True value
Predictive value Predictive value
160 160

140 140

120 120

UCS (MPa)
UCS (MPa)

100 100

80 80

60 60

40 40

20 20

0 0
0 5 10 15 20 25 1 2 3 4 5
Test piece serial number Test piece serial number
(a) (b)
Figure 6. Comparison between between
Figure 6. Comparison the real the
andreal
predicted valuesvalues
and predicted of model M1, M1,
of model (a) M1 training
(a) M1 setset
training andand(b)
(b)M1
M1prediction set.
prediction set.

It can be After
seen a large
fromamount
the twoofmodel
training with data in
evaluation the training
indexes set, the
that the model of
accuracy accuracy
modeland M2
generalization ability continue to improve. According to the model M1 prediction results,
is higher than that of model M3, and the2 evaluation indexes of the M2 are better. There-
the MSE in the training set is 62.66, R reaches 0.96937, the MSE of the prediction set is 50.05,
fore, forand
thisR2mining project, it is more suitable to use diametral loading for point load
is 0.98, which shows the SVM model has a high fit degree and a small deviation
strengthfrom
testthe
in the
average. Itwhich
field, is foundcan quickly
that predict
the sample areathe
withuniaxial compression
UCS lower than 100 MPastrength and
has higher
enhanceprediction
the efficiency
accuracyandthan
geotechnical
the sample design
area withcapability.
UCS higher than 100 MPa. Compared with
Combining the performance results 2 of the
some existing studies [9,15,26], the R coefficient threeof models,
model M1the overall
is higher performance
and the predictionof
model M1 is better and
performance the accuracy is the highest. This shows that the prediction method
is better.
According
of using APLS and DPLS to thetopractical
predictapplication experience
UCS is better than of SVMonly
using model, the main
a single typeparameters
of point
affecting the prediction results are penalty factor C and sensitive
load strength. When the engineering economic conditions permit, the axial point parameter γ. Byload
finding
test
the optimal C and γ, the applicability of the model can be optimized. The parameters used
and the diametral point load test should be carried out at the same time to obtain the best
in this study are shown in Table 2.
prediction effect.
Table 2. Two main parameters used in the SVM model.
MSE=48.24 R2=0.97 MSE=76.44 R2=0.99
180 True value
180 True value Model Name Penalty Factor C
Predictive value
Sensitive Parameter γ
Predictive value
160 M1 160 256.75 1.13
M2 226.27 1.41
140 M3 140 362.04 2.38
120 120
UCS (MPa)
UCS (MPa)

100
Subsequently, we establish100the relationship model M2 for predicting UCS by using
DPLS. As shown in Figure 7, the results show that the MSE in the training set is 48.24, R2
80 is 0.97, the MSE in the prediction80set is 76.44, and R2 is 0.99. There is not much difference
60 between the predicted value and60the true value in the prediction set.

40 40

20 20

0 0
0 5 10 15 20 25 1 2 3 4 5

Test piece serial number Test piece serial number

(a) (b)
 enhance the efficiency and geotechnical design capability.
Combining the performance results of the three models, the overall performance of
model M1 is better and the accuracy is the highest. This shows that the prediction method
of using APLS and DPLS to predict UCS is better than using only a single type of point
load strength. When the engineering economic conditions permit, the axial point load test
Minerals 2021, 11, 1387 11 of 14
and the diametral point load test should be carried out at the same time to obtain the best
prediction effect.

MSE=48.24 R2=0.97 MSE=76.44 R2=0.99

180 True value
180 True value
Predictive value
Predictive value
160 160

140 140

120 120

UCS (MPa)
UCS (MPa)

100 100

80 80

60 60

40 40

20 20

0 0
Minerals 2021, 11,
0 x FOR PEER
5 REVIEW
10 15 20 25 1 2 3 4 5 12 of 15
Test piece serial number Test piece serial number
(a) (b)
2
MSE=19.81 R =0.98 MSE=176.48 R2=0.96
180 180 True value
True value
Predictive value
Predictive value
160 160

140 140

120 120
UCS (MPa)
UCS (MPa)

100 100

80 80

60 60

40 40

20 20

0 0
0 5 10 15 20 25 1 2 3 4 5
Test piece serial number Test piece serial number
(c) (d)
Figure 7. Comparison
Figure 7. Comparison of the
of real value
the real andand
value predicted value
predicted of model
value M2M2
of model andand
model M3,M3,
model (a) (a)
M2M2training set,set,
training (b) (b)
M2M2
pre-
diction set, (c) set,
prediction M3 (c)
training set and
M3 training set(d)
andM3(d)prediction set. set.
M3 prediction

Under the same grouping conditions, we establish the relationship model M3 to

4.3. Limitations
predict UCS by APLS. As shown in Figure 7, the MSE also obtained in training set is 19.81,
The main content of this research is the correlation between the PLS and the UCS,
R2 is 0.98, and the MSE in the prediction set is 176.48, R2 reached 0.96. It can be found that
andthe predictedthe
exploring feasibility
value of sample andNo.predictive performance
4 and sample No. 5 has of the SVM
a large error model
with theintrue
predicting
value.
theThe
uniaxial compressive strength by using the point load strength. Although
main reason is that the training data amount is small, and the strength discreteness this model
hasinterferes
significant
withperformance
the predictionand advantages in predicting the uniaxial compressive
accuracy.
strength of rocks, it also has some limitations
It can be seen from the two model evaluation that require
indexes deeper
that theresearch.
accuracy of model M2
1. is higher than that ofstrength
The mechanical model M3, and the
of rock hasevaluation indexes
a high degree of the M2 areand
of dispersion, better.
theTherefore,
amount of
fordata
this used
mining project, it is more suitable to use diametral loading for point
in this model is small, which may easily lead to errors in the prediction load strengthof
test
the strength of individual specimens [50]. In order to indicate the accuracyenhance
in the field, which can quickly predict the uniaxial compression strength and and sta-
thebility
efficiency
of theand geotechnical
model, furtherdesign capability.
research is required of more relevant test data sets for
Combining the performance results of the three models, the overall performance of
verification.
model M1 is better and the accuracy is the highest. This shows that the prediction method
2. This study only focuses on the model relationship between the point load strength
of using APLS and DPLS to predict UCS is better than using only a single type of point
and the uniaxial compressive strength. Other indicators involved in the test, such as
load strength. When the engineering economic conditions permit, the axial point load test
specimen loading point spacing D and the minimum cross-sectional width W, have
not been involved. The specific relationship between other variables and the uniaxial
compressive strength of the rock is not clear.
3. In order to better improve the prediction performance and accuracy of the model, it
is necessary to better optimize and search the built-in parameters of the established
SVM model, such as the penalty factor C and kernel parameter γ. The calculation
Minerals 2021, 11, 1387 12 of 14

and the diametral point load test should be carried out at the same time to obtain the best
prediction effect.

4.3. Limitations
The main content of this research is the correlation between the PLS and the UCS, and
exploring the feasibility and predictive performance of the SVM model in predicting the
uniaxial compressive strength by using the point load strength. Although this model has
significant performance and advantages in predicting the uniaxial compressive strength of
rocks, it also has some limitations that require deeper research.
1. The mechanical strength of rock has a high degree of dispersion, and the amount of
data used in this model is small, which may easily lead to errors in the prediction
of the strength of individual specimens [50]. In order to indicate the accuracy and
stability of the model, further research is required of more relevant test data sets
for verification.
2. This study only focuses on the model relationship between the point load strength
and the uniaxial compressive strength. Other indicators involved in the test, such as
specimen loading point spacing D and the minimum cross-sectional width W, have
not been involved. The specific relationship between other variables and the uniaxial
compressive strength of the rock is not clear.
3. In order to better improve the prediction performance and accuracy of the model, it
is necessary to better optimize and search the built-in parameters of the established
SVM model, such as the penalty factor C and kernel parameter γ. The calculation
work in this area can be enhanced in the following research.

5. Conclusions
Through the test results analysis, it was found that the PLS of the limestone in the mine
area had an exponential relation with the UCS. Then based on the exponential relationship,
the SVM models for predicting UCS by DPLS and APLS are established respectively, and
the following conclusions are drawn:
1. Four fitting methods are both used to obtain the formula with the highest correlation
coefficient. The most accurate fitting formula and function for APLS and UCS is
R = 6.46e0.56Is(50) , where R2 is 0.8846, and the most accurate fitting formula for DPLS
and UCS is R = 9.80e0.47Is(50) , where R2 is 0.9485. The above formulas are all expressed
as exponential functions.
2. The proposed SVR model can accurately predict the UCS through the PLS. The
predicted value is basically the same as the true value, and the maximum error
rate is below 15%. The SVM model prediction result is more accurate compared
with the traditional fitting methods and it also proves that the SVM algorithm can
be better applied to the rock strength prediction scenarios under the condition of
small samples.
3. For the first time, the APLS and DPLS are both used as an entirety to predict the UCS,
and the prediction model M1 was established. By comparing the three models, it
is found that in the R2 performance: the correlation coefficient of model M2 is the
largest and the correlation coefficient of M3 is the smallest. In the MSE performance,
the MSE of model M1 is the smallest and the M3 model is the largest. Combining the
R2 and MSE judgment indexes, model M1 has the best overall performance, which
indicates that the accuracy of predicting UCS by using APLS together with DPLS is
the highest.
4. Through comparison, it is found that model M2 has better prediction performance
than model M3, which indicates from the side that the prediction of UCS by using
DPLS is more accurate than using APLS. If DPLS and APLS cannot be obtained
simultaneously at the engineering site, the point load test method of diametral loading
can be preferred.
Minerals 2021, 11, 1387 13 of 14

Supplementary Materials: The following are available online at https://www.mdpi.com/article/

10.3390/min11121387/s1, Table S1: Record of data indexes for uniaxial compression strength test,
Table S2: Record of data indexes for diametral point load strength, Table S3: Record of data indexes
for axial point load strength.
Author Contributions: S.L. and Y.W. provided experimental ideas and carried out the experiment;
X.X. directed and assisted the experiment; S.L. wrote the paper. All authors have read and agreed to
the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: The source data can be obtained in the article.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wu, A.Q.; Wang, B. Engineering rock mass classification method based on rock mass quality index BQ for rock slope. Chin. J.
Rock Mech. Eng. 2014, 33, 699–706. [CrossRef]
2. Sha, P.; Zhang, Q.T.; Lin, J.; Wu, F.Y. In-situ estimation of uniaxial compression strength of igneous rock based on point load
strength. Rock Soil Mech. 2020, 1–10. [CrossRef]
3. Broch, E.; Franklin, J.A. The point-load strength test. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 1972, 9, 669–697. [CrossRef]
4. Kohno, M.; Maeda, H. Relationship between point load strength index and uniaxial compressive strength of hydrothermally
altered soft rocks. Int. J. Rock Mech. Min. Sci. 2012, 50, 147–157. [CrossRef]
5. Kong, F.; Shang, J. A Validation Study for the Estimation of Uniaxial Compressive Strength Based on Index Tests. Rock Mech.
Rock Eng. 2018, 51, 2289–2297. [CrossRef]
6. Bieniawski, Z.T. Estimating the strength of rock materials. J. South. Afr. Inst. Min. Metall. 1974, 74, 312–320. [CrossRef]
7. ISRM. Suggested method for determining point load strength. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 1985, 22, 51–60.
[CrossRef]
8. ASTM. Standard Test Method for Determination of the Point Load Strength Index of Rock and Application to Rock Strength Classification.
Test Designation D5731−95 Annual Book of ASTM Standards; American Society for Testing and Materials: West Conshohocken, PA,
USA, 1995; p. 2.
9. Tsiambaos, G.; Sabatakakis, N. Considerations on strength of intact sedimentary rocks. Eng. Geol. 2004, 72, 261–273. [CrossRef]
10. Basu, A.; Kamran, M. Point load test on schistose rocks and its applicability in predicting uniaxial compressive strength. Int. J.
Rock Mech. Min. Sci. 2010, 47, 823–828. [CrossRef]
11. Wong, R.H.; Chau, K.; Yin, J.-H.; Lai, D.T.; Zhao, G.-S. Uniaxial compressive strength and point load index of volcanic irregular
lumps. Int. J. Rock Mech. Min. Sci. 2017, 93, 307–315. [CrossRef]
12. Vallejo, L.E.; Welsh, R.A.; Robinson, M.K. Correlation between unconfined compression and point load strengths for Appalachian
rocks. In The 30th US Symposium on Rock Mechanics (USRMS); Rock Mechanics Association: Alexandria, VA, USA, 1989;
pp. 461–468.
13. Quane, S.L.; Russell, K. Rock strength as a metric of welding intensity in pyroclastic deposits. Eur. J. Miner. 2003, 15, 855–864.
[CrossRef]
14. Kılıç, A.; Teymen, A. Determination of mechanical properties of rocks using simple methods. Bull. Int. Assoc. Eng. Geol. 2008,
67, 237–244. [CrossRef]
15. Sabatakakis, N.; Koukis, G.; Tsiambaos, G.; Papanakli, S. Index properties and strength variation controlled by microstructure for
sedimentary rocks. Eng. Geol. 2008, 97, 80–90. [CrossRef]
16. Yilmaz, I.; Yuksek, G. Prediction of the strength and elasticity modulus of gypsum using multiple regression, ANN, and ANFIS
models. Int. J. Rock Mech. Min. Sci. 2009, 46, 803–810. [CrossRef]
17. Diamantis, K.; Gartzos, E.; Migiros, G. Study on uniaxial compressive strength, point load strength index, dynamic and physical
properties of serpentinites from Central Greece: Test results and empirical relations. Eng. Geol. 2009, 108, 199–207. [CrossRef]
18. Li, D.; Wong, L.N.Y. Point Load Test on Meta-Sedimentary Rocks and Correlation to UCS and BTS. Rock Mech. Rock Eng. 2012,
46, 889–896. [CrossRef]
19. Mishra, D.; Basu, A. Estimation of uniaxial compressive strength of rock materials by index tests using regression analysis and
fuzzy inference system. Eng. Geol. 2013, 160, 54–68. [CrossRef]
20. Fu, Z.L.; Wang, L. Comparative experimental research on point load strength, uniaxial compression strength and tensile strength
for rocks in roof and floor of coal seam. Chin. J. Rock Mechan. Eng. 2013, 32, 88–96.
21. Kahraman, S. The determination of uniaxial compressive strength from point load strength for pyroclastic rocks. Eng. Geol. 2014,
170, 33–42. [CrossRef]
22. Anping, L.; Yangshu, L.; Ming, Z. Study on relationship between poit load strength and uniaxial compressive strength of rock.
Non-Ferrous Metals 2014, 66, 53–58.
23. Zhang, J.M.; Tang, Z.C.; Liu, Q.S. Relation between point load index and uniaxial compression strength for igneous rock. Rock Soil
Mech. 2015, 36, 595–602. [CrossRef]
Minerals 2021, 11, 1387 14 of 14

24. Yin, J.-H.; Wong, R.H.; Chau, K.; Lai, D.T.; Zhao, G.-S. Point load strength index of granitic irregular lumps: Size correction and
correlation with uniaxial compressive strength. Tunn. Undergr. Space Technol. 2017, 70, 388–399. [CrossRef]
25. Lin, J.; Sha, P.; Wu, F. Correlation Between Point Loading Index and Uniaxial Compression Strength of Rock-like Material Based
on Size Effect. J. Yangtze River Sci. Res. Inst. 2018, 35, 34–44.
26. Chen, J.Q.; Wei, Z.A. Comparison of rock strength from different point load tests and the uniaxial compression strength. Chin. J.
Geol. Hazard Control 2018, 29, 72–77. [CrossRef]
27. Li, H.P.; Wu, Y.Y.; Ge, C. The relation between point load strength and compression and tensile strength of marble. Sci. Technol.
Eng. 2019, 19, 294–299.
28. Zhang, C.L.; Zhang, C.P.; Xu, J. Comparison Test of Rock Point Load Strength and Uniaxial Compres-sion Strength. Chin. J.
Undergr. Space Eng. 2015, 11, 447–451.
29. Fan, J.; Guo, Z.; Tao, Z.; Wang, F. Method of equivalent core diameter of actual fracture section for the determination of point load
strength index of rocks. Bull. Int. Assoc. Eng. Geol. 2021, 80, 4575–4585. [CrossRef]
30. Mozumder, R.A.; Laskar, A.I. Prediction of unconfined compressive strength of geopolymer stabilized clayey soil using Artificial
Neural Network. Comput. Geotech. 2015, 69, 291–300. [CrossRef]
31. Heidari, M.; Khanlari, G.R.; Kaveh, M.T.; Kargarian, S. Predicting the Uniaxial Compressive and Tensile Strengths of Gypsum
Rock by Point Load Testing. Rock Mech. Rock Eng. 2012, 45, 265–273. [CrossRef]
32. Shao, Y.; Yan, C.H.; Ma, Q.H. Correlation Analysis of Dayangshan Rock Strength in Suzhou. J. Yangtze River Sci. Res. Inst. 2015,
32, 107–110.
33. Zhang, Y.Y.; Li, K.G. The Correlation Between the Point Load Strength and Uniaxial Compression Strength of Several Kinds of
Rocks. Metal Mine 2017, 19–23.
34. Zhang, Y.; Zhao, G.F.; Wei, X.; Li, H. A multifrequency ultrasonic approach to extracting static modulus and damage characteristics
of rock. Int. J. Rock Mech. Min. Sci. 2021, 148, 104925. [CrossRef]
35. Lei, S.; Kang, H.P.; Gao, F.Q.; Zhang, X. Point load strength test of fragile coal samples and predictive analysis of uniaxial
compression strength. Coal Sci. Technol. 2019, 47, 107–113. [CrossRef]
36. Liu, Q.S.; Zhao, Y.F.; Zhang, X.P.; Kong, X.X. Study and discussion on point load test for evaluating rock strength of TBM tunnel
constructed in Limestone. Rock Soil Mech. 2018, 39, 977–984. [CrossRef]
37. Ministry of Housing and Urban Rural Development of the People’s Republic of China. GB/T 50218-2014 Standard for Engineering
Classification of Rock Mass; China Planning Press: Beijing, China, 2014.
38. Ministry of Housing and Urban Rural Development of the People’s Republic of China. GB/T 50266-2013 Standard for Test Methods
of Engineering Rock Mass; China Planning Press: Beijing, China, 2013.
39. Gupta, D.; Natarajan, N. Prediction of uniaxial compressive strength of rock samples using density weighted least squares twin
support vector regression. Neural Comput. Appl. 2021, 33, 15843–15850. [CrossRef]
40. Sain, S.R.; Vapnik, V.N. The Nature of Statistical Learning Theory. Technometrics 1996, 38, 409. [CrossRef]
41. Kang, F.; Xu, Q.; Li, J. Slope reliability analysis using surrogate models via new support vector machines with swarm intelligence.
Appl. Math. Model. 2016, 40, 6105–6120. [CrossRef]
42. Zhou, J.; Huang, S.; Wang, M.; Qiu, Y. Performance evaluation of hybrid GA–SVM and GWO–SVM models to predict earthquake-
induced liquefaction potential of soil: A multi-dataset investigation. Eng. Comput. 2021, 1–19. [CrossRef]
43. Li, E.; Zhou, J.; Shi, X.; Armaghani, D.J.; Yu, Z.; Chen, X.; Huang, P. Developing a hybrid model of salp swarm algorithm-based
support vector machine to predict the strength of fiber-reinforced cemented paste backfill. Eng. Comput. 2020, 37, 3519–3540.
[CrossRef]
44. Syah, R.; Ahmadian, N.; Elveny, M.; Alizadeh, S.M.; Hosseini, M.; Khan, A. Implementation of artificial intelligence and support
vector machine learning to estimate the drilling fluid density in high-pressure high-temperature wells. Energy Rep. 2021,
7, 4106–4113. [CrossRef]
45. Joana, C.; Petra, T.; Maisa, N.; Timo, J.; Vahid, F. Machine-learning models for activity class prediction: A comparative study of
feature selection and classification algorithms. Gait Posture 2021, 89, 45–53.
46. Wang, Y.; Qiao, C.S.; Sun, C.H.; Liu, K.Y. Forecasting model of safe thickness for roof of karst cave tunnel based on support vector
machines. Rock Soil Mech. 2006, 1000–1004. [CrossRef]
47. Huang, W.Y.; Tong, M.M.; Ren, Z.H. Nonlinear combination forecast of gas emission amount based on SVM. J. China Univ.
Min. Technol. 2009, 38, 234–239.
48. Hsu, C.W.; Lin, C.J. A comparison of methods for multiclass support vector machines. IEEE Trans. Neural Netw. 2002, 13, 415–425.
49. Wang, Q.; Sun, H.B.; Jiang, B. A method for predicting uniaxial compressive strength of rock mass based on digital drilling test
technology and support vector machine. Rock Soil Mech. 2019, 40, 1221–1228. [CrossRef]
50. Teymen, A.; Mengüç, E.C. Comparative evaluation of different statistical tools for the prediction of uniaxial compressive strength
of rocks. Int. J. Min. Sci. Technol. 2020, 30, 785–797. [CrossRef]