You are on page 1of 13

International Journal of Rock Mechanics & Mining Sciences 128 (2020) 104258

Contents lists available at ScienceDirect

International Journal of Rock Mechanics and Mining Sciences


journal homepage: http://www.elsevier.com/locate/ijrmms

The usability of Cerchar abrasivity index for the estimation of mechanical


rock properties
Ahmet Teymen
Department of Mining Engineering, Ni�gde Omer
€ Halisdemir University, 51240, Ni�gde, Turkey

A R T I C L E I N F O A B S T R A C T

Keywords: One of the most commonly used methods of determining the rock abrasivity is Cerchar Abrasiveness Index (CAI).
Cerchar The basic parameters affecting abrasiveness are density, cementation degree, strength properties and the number
Abrasivity of abrasive minerals in the rock. In this study, 20 different tests were carried out on 80 different rocks, mainly
Multiple regression
Young’s modulus (E), uniaxial compressive strength (UCS) and CAI. The main aim of the study is to estimate the
Rock properties
Performance index
basic mechanical rock properties which are difficult to determine and time-consuming with the help of CAI
values. In addition to simple equations, multiple regression equations were developed using CAI and other test
methods (11 tests). Among the obtained equations, the most reliable ones were determined with the help of a
developed performance index (PIat1). The results of the study show that there are significant relationships be­
tween basic mechanical rock properties and CAI.

1. Introduction because of its quick and simple applicability and the use of small
amounts of rock samples. CAI testing has been recommended by the
In mining operations, either drilling-blasting or mechanized exca­ International Society of Rock Mechanics ISRM2 as a rock mechanics
vations methods (roadheaders, tunnel boring machines, dragline, and testing method. CAI testing is one of the most commonly used testing
excavator) are performed. All the tools used in the mentioned methods procedures in the UK and France coal mining industry to estimate tool
interact with the rock. This interaction does not only leads to frag­ life and determine tool cost.3–6,17 The test was originally developed by
mentation or deformation of the rock, but also abrasion of the used tools. the Laboratoire du Centre d’E0 tudes et Recherches des Charbonnages
Increased advancement rate of the tunnel or increased production rates, (Cerchar) de France for coal mining applications.7 Three standards exist
especially in mining operations, require more knowledge of rock abra­ for this test method: the AFNOR NF P 94-430-1,8 ASTM D7625-10.9 and
sivity. It is important to know or predict the interaction of rock cutting ISRM.10 The test is widely used in research and practice. There are two
tools with changing rock types and geological formations. The unpre­ basic test apparatus designs, the original design developed at the Cer­
dictability of rock abrasiveness leads to unexpected high cutting tool char Center11 and modified by West.12
wear rates, which greatly affects the project budget.1 Generally, abra­ Suana and Peters13 used the equivalent quartz content obtained from
sion can be defined as the tearing or wearing away of particles from the thin section analyzes prepared to estimate CAI. They also emphasized
material surface. Abrasion in the excavation tool can be defined as the that other parameters, such as matrix properties and grain size, played
loss of tool material when interacting with the rock. Rock abrasivity an important role in the CAI estimation. They determined that the
plays an important role in the characterizing of a rock material for abrasiveness of Quartz mineral, which has a very high abrasive char­
excavation purposes. Rock abrasiveness is a function of the amount of acter, had a 100% equivalent quartz content and was equivalent to six
quartz and similar abrasive minerals in the rock, average quartz grain (6) CAI. West14 in his study on coal-measures sedimentary rocks (31
size, cementation, and cement type. rocks), determined the directly proportional relationship between CAI
The resistance of tool wear when in contact with rock is the most rock abrasiveness and quartz content, Mohs hardness value. Johson and
common principle for measuring rock abrasivity. The Cerchar Abrasivity Fowell15 using in-situ measurements, showed that the cutter consump­
Test is a method for determining the index of abrasiveness of rock called tion used in excavation machines was directly related to the CAI values
the Cerchar Abrasivity Index (CAI) and has great popularity in academic of the rocks. They also stated that in addition to abrasive minerals and
research as well as in industry. The CAI test method is highly preferred cementing properties, strength properties had an effect on the CAI.

E-mail address: ateymen@ohu.edu.tr.

https://doi.org/10.1016/j.ijrmms.2020.104258
Received 7 October 2019; Received in revised form 11 February 2020; Accepted 11 February 2020
Available online 28 February 2020
1365-1609/© 2020 Elsevier Ltd. All rights reserved.
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Al-Ameen and Wallner16 investigated the relationship between CAI and measures rocks in the Zonguldak Hard Coal Basin were investigated by
rock strength. They determined that some high strength rocks contain­ Yaralı.26 For this purpose, CAI, (SSH, and Is50) and strength (UCS and
ing low abrasive minerals may have a high abrasive index, and some low BTS) tests were conducted on sedimentary rocks. Then, relationships
strength rocks with high abrasive minerals may have a low abrasive among strength and CAI, index test values were evaluated using
index. They reported that cementing and strength properties with cor­ regression analysis. As a result, positive linear relationships were found
rosive mineral content had an effect on CAI. Plinninger et al.,17 in their among CAI and UCS, BTS, SSH, and Is50. Moradizadeh et al.27 investi­
study on 109 rock samples, demonstrated test conditions and geo­ gated the correlations between the CAI and quartz content (EQC), Is50,
mechanical properties that affect the CAI value. They proposed a slake durability index, and percentage of water absorption of 36
correction (CAI ¼ 0.99CAIs þ 0.48) for CAI values obtained from different rock samples using simple and multivariate regression. Er and
saw-cut rock samples. Also, they conducted some standard tests, such as Tugrul28-29 developed empirical relationships between the chemical,
UCS, E, and BTS; as well as petrographic thin section analysis, grain size mineralogical, petrographical and physico-mechanical properties and
analysis, and density. Finally, they compared the results of their study CAI values of 12 different granitic rock samples using the simple
with West’s12 results. They have indicated that the equivalent quartz regression analysis. Majeed and Abu Bakar30 tested the characteristics of
content (Qeq) alone is not suitable for interpreting CAI values. Plin­ CAI, UCS, BTS, PG, Vp of a large number of rocks from different regions
ninger et al.18 presented the correlations between CAI and other of Pakistan and evaluated the CAI measurement methods. They also
abrasivity-related parameters (elastic modulus and equivalent quartz carried out petrographic thin section analyses to determine the abrasion
content). In their comprehensive study, they compared the CAI values factors such as Schimazek’s F value and Rock Abrasivity Index in some
obtained from rough and smooth rock surfaces. Effects of test length on of the rocks. Ko et al.31 emphasized that a single independent variable is
the CAI were shown, the non-existent relationship between CAI and not sufficient when estimating the CAI value. Also, they showed the
equivalent Quartz content was discussed and a possible dependence of positive effect of UCS and brittleness index for magmatic rocks and UCS
CAI to the modulus of elasticity was discussed. In summary, the CAI test and BTS for metamorphic rocks in the models. Yaralı32 tested some
has been successfully applied in the exploration process of rock engi­ index properties (CHI, SSH, SHH, and Is50) and strength (UCS and BTS)
neering. Yaralı and Akçın19 studied some rock properties in the Zon­ properties of 29 sedimentary coal measure rocks in Zonguldak hard coal
guldak Hard Coal Basin. They determined the hardness of rocks by drill basin and determined linear relationships between CHI and UCS, SHH,
bits with two different tip angles and investigated the effect of drill bit power relationships between CHI and SSH, Is50 using statistical
angle on CAI values. Finally, they identified correlations between the methods. The author also showed that there was an excellent linear
CAI and mechanical properties. Michalakopoulos et al.20 examined the relationship between the test results made with bits having 99 and 125�
influence of the steel hardness on the CAI value. Lassnig et al.21 studied tip angles. In the study of Ozdog �an et al.,33 the relationship between
different testing instructions on the influence of the grain size on the SSH, PG, UCS and CAI values of building stones were examined using
CAI. The main purpose of their work was to verify the influence of grain simple and multiple regression analysis. In particular, they found a
size on the number of required single examinations per sample. They strong relationship between SSH and CAI. They emphasized that if the
performed grain size analyses to obtain the total curves of each tested CAI value is greater than one, the coefficient of determination can be
rock sample and calculated the median and interquartile range of par­ higher. Kadhodaei et al.34 created a data set using experimental results
ticle sizes from the grain size data. After 5 and 10 scratch tests, they from several studies in the literature. They trained and tested this data
compared the CAI values with the median and grain size qu range. set with the Gene expression programming (GEP) approach. They used
Alber22 demonstrated that the CAI was stress-dependent. A novel test the abrasivity index (RAI) and BTS parameters in the GEP model to es­
procedure in which the CAI tests were conducted on samples in a triaxial timate CAI values. They also compared the model with multiple
cell shows the stress dependency for various rock types by higher CAI regression and other existing models in the literature.
values upon confining pressure. Kahraman et al.23 focused on the pre­ The main purpose of this study is to estimate the basic rock prop­
dictability of Misis Fault Breccia’s E and UCS values from some indirect erties which are carried out using samples prepared in specific sizes with
methods such as CAI, UPV, density using artificial neural network the CAI test which is widely and practically used in the determination of
analysis and regression analysis. The CAI was included in the best rock abrasiveness. Based on this idea, firstly the relationships between
regression model presented for UCS estimation but was not used in UCS, E, BTS, Is50, BSA, ROP, BPI, KIC tests and CAI were determined. In
estimating the E. In conclusion, the CAI was found to be more useful in addition to the correlation studies using all 80 rocks, the data set
estimating UCS of Misis Fault Breccia and the ANN model was more grouped using various classification systems were evaluated in detail.
reliable than the regression model. In Deliormanlı’s study,24 it was used Besides, multiple regression studies were performed for the eight
simple and multiple regression methods to determine the strength values different tests listed above. Multiple regression studies were performed
of rocks such as UCS, direct shear (DSS) and abrasion properties such as using two, three and four independent variables, one of which was CAI
BSA, Wide-Wheel (WW) with CAI. According to the results of the study values. Multiple regression equations were compared with the help of
using 15 commonly used marbles, two new converter charts were various statistical tools. The most reliable equations were determined by
developed. The first chart represents the relationship between using the performance index criterion proposed in this study. According
CAI–UCS–DSS, while the second chart displays the relationship between to the detailed analyses, the CAI test is very successful for estimating the
CAI–BA–WW. The researcher stated that these charts can be used basic engineering properties.
practically by engineers and academicians who study on construction
and building materials to estimate their strength and abrasion values 2. Sampling and test procedure
using the CAI of marbles. Dipova25 tested Texas tunnel rock samples and
investigated the relationship between the CAI data and the strength The different originated rocks used in this study were collected from
characteristics of weak limestones. Considering the abrasion of the rock the factories, outcrops and natural stone quarries in different locations
and pin together, it has measured and correlated the wear on the pin and of Turkey. Tests were carried out on blocks or pieces taken from fresh
indentation on the rock. The researcher stated that the CAI method used parts of 80 rocks. The rocks tested are composed of volcanic (20),
for estimating the digger life in hard rocks can be used to estimate the plutonic (13), pyroclastic (10), subvolcanic (7), sedimentary (18) and
strength properties of weak rocks. As a result of statistical studies, it was metamorphic (12) rocks. The rock samples were macroscopically
determined that there is a relationship between the CAI and UCS, BTS inspected and samples were used that did not have any alters and dis­
values. It was found that there was a relationship between the CAI and continuities. The Brillant 250 wet abrasive cut-off machine was used for
indentation depth (ID) and between ID and UCS, BTS values. The re­ sizing and smoothing the core samples. During the sample preparation
lationships between mechanical properties and the abrasiveness of coal phase, standard laboratory test procedures were used ISRM.35 A

2
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

classification of CAI proposed by ISRM10 is given in Table 1. When CAI


ranges from 0 to 2.0, rock abrasiveness changes from slight to medium.
For CAI of 4–6, rock is defined as very to extremely abrasive.
According to the classification in Table 1; seven of rocks are very low
abrasive (8.8%), twenty-three are low abrasive (28.8%), nineteen are
medium abrasive (23.8%), nineteen are high abrasive (23.8%), eleven
are very high abrasive (13.8%), and one is extremely high abrasive
(1.3%) rock class.

3. Experimental studies

By the aim of the study, a total of 20 different tests were applied to 80


rocks with the CAI test. These tests; Uniaxial compressive strength
(UCS), Young’s modulus (E), Brazilian tensile strength (BTS), Point load
index (Is50), Bohme surface abrasion (BSA), Rate of penetration (ROP),
Block punch index (BPI), Fracture toughness (KIC), Cerchar Abrasive­
ness Index (CAI), P-wave velocity (UPV), Shore Schlerescope hardness
(SSH), Schmidt hammer hardness (SHH), Brinell hardness (HB), Vickers
hardness (HV), Water absorption by the weight (WA), Effective porosity
Fig. 1. Measurement of abrasivity on the steel pin by a computer program
(PG), Impact strength index (ISI), Coefficient of rock strength (CRS),
(screen caliper).
Crushability index (CI) and Protodyakonov index (f). The CAI tests were
performed using the original Cerchar apparatus by scratching pins of
HRC 54–56. CAI apparatus features a vice holding the rock specimen, a BTS test apparatus was used to determine the tensile strength of
rack for the steel stylus, a deadweight of 70 N and a hand crank with rocks. For each rock type, at least seven core samples with a diameter of
which the specimen is moved underneath the steel pin. The tests were 54 mm and a thickness of about 27 mm were used and the test was
carried out on the disc specimens having 42 mm diameter, 30–50 mm carried out following the ISRM35 standard. The load on the rock samples
thickness. The disc samples were manually moved by the hand crank was continuously applied at a constant stress rate such that failure took
with 1 mm/s for a length of 10 mm. This was repeated for a total of five place within 5 min. The point load test (Is50) is considered an indirect
pins. An apparatus was prepared using a high-resolution camera to measure of the rock strength. The point load tool is widely used in
visualize the amount of wear occurring at the steel pins in a standard practice because no special sample is needed and can be used at the field
way. Measurements on the photographs were performed utilizing a application. The diametric method was preferred for the Is50 test. Ac­
calibrated digital caliper software (Digimizer) (Fig. 1). A wear surface of cording to this method, the ratio of the length of the sample to the
1/10 mm at the steel pin is considered a unit of Cerchar abrasivity index diameter (54 mm) should be between 1.0 and 1.4. For each rock type,
and is used without units Eq. (1). samples ranging in number from seven to ten were tested and the
average values obtained from this test were given in Table 2.
CAI ¼ 10ðd=cÞ (1) The breakage of particles by mechanical impacts such as friction and
impact from the building stone surface can be described as the abrasion
CAI: Cerchar abrasivity index resistance (BSA). The problem of abrasion in rocks develops due to the
d: Diameter of wear flatness (mm) load and movement between the surfaces. BSA is an effective feature for
c: Unit correction factor (c ¼ 1 mm). determining the resistance of stones used in engineering projects and is
an important indicator of the availability of stone in higher movement
The UCS and E tests were performed on cylindrical rock specimens areas. To determine the abrasion behavior of rocks, cubic samples with a
having a 42 mm diameter and a length-to-diameter ratio of 2–2.5. The 71 mm side (50 cm2 surface areas) were dimensioned for the abrasion
ends of the cores cut off from the main rock, were cut parallel to each test. For this purpose, the Bohme apparatus described in detail at36-37
other, and then the surfaces of the cores were smoothed to make sure so was used. In compliance with the standards, the abrasion system has a
that there were no sudden irregularities. The compression machine steel disc, which has a diameter of 750 mm and a rotating speed of 30 �
(ELE-3000 kN) was used for the determination of the UCS. Stress ratio 1 cycle/min, a counter and a lever, which could apply load on the
was applied uniformly until the rock was completely deformed and this specimen. The samples placed on the disc were exposed to 440 cycles
ratio remained between 1-1.2 MPa/s. For determining the E, strain under 300 N load using 20 g of abrasion dust (crystalline Al2O3) each
gauges were used. The displacements were recorded using strain gauges time. BSA was calculated from the differences in dimensions before and
during axial loading. The E of rock was derived from the slope of the after the testing of the specimen.
axial stress-axial strain curves at a stress level equals to 50% of the ul­ Laboratory type core drilling machine and 22 mm diameter diamond
timate UCS. The mean UCS and E values were calculated by taking the core were used to determine the rate of penetration (ROP). For the test,
average of the five test values. 54 mm core samples with a length/diameter ratio of approximately one
were prepared. The rocks were cut by the weight of the drill without
applying any compressive force. The ROP was calculated by dividing the
Table 1 progression distance via the progression time. The BPI test apparatus
Classification of CAI.10 used in this study is a test apparatus, which is described in detail by
Classification Mean CAI
Ulusay et al.38 and can be mounted on a manually loaded point load
device. The BPI apparatus consists of base support and a hardened steel
Extremely High
punching block. Smoothness disc rock specimens having 10 mm thick­
�5
Very High 4.0–4.9
High 3.0–3.9 ness were installed at the center of the BPI test apparatus. The load was
Medium 2.0–2.9 increased manually so that the failure occurred at a maximum of 60 s. In
Low 1.0–1.9 the calculations, test samples, which are regularly broken into three
Very Low 0.5–0.9
sections were used. At least seven test samples were used for each rock
Extremely Low 0.1–0.4
type.

3
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Table 2
Ranges of experimental data.
Parameter N Range Minimum Maximum Mean Std. Deviation Variance Skewness Kurtosis

CAI (*) 80 4.8 0.5 5.3 2.6 � 0.1 1 1.3 0.16 � 0.27 0.92 � 0.53
UCS (MPa) 80 297.0 6.6 303.7 102.1 � 7.5 67 4462.1 0.82 � 0.27 0.22 � 0.53
E (GPa) 80 57.8 1.0 58.8 18.9 � 1.3 12 133.2 0.67 � 0.27 0.67 � 0.53
BTS (MPa) 80 20.3 1.0 21.3 8.2 � 0.5 4 20.1 0.64 � 0.27 0.08 � 0.53
Is50 (MPa) 80 14.6 1.2 15.7 5.4 � 0.4 3 9.9 0.94 � 0.27 0.61 � 0.53
BSA (cm3/50 cm2) 80 134.1 5.0 139.1 28.7 � 3.0 27 737.6 2.31 � 0.27 5.61 � 0.53
ROP (mm/sec) 80 235.9 0.4 236.3 23.4 � 5.0 45 2026.4 3.24 � 0.27 11.0 � 0.53
BPI (MPa) 80 41.0 1.2 42.2 14.5 � 1.0 9 77.9 0.87 � 0.27 0.58 � 0.53
KIC (MN/m3/2) 80 3.5 0.1 3.6 1.4 � 0.1 1 0.8 0.80 � 0.27 0.10 � 0.53
UPV (km/sec) 80 5.8 0.9 6.7 4.6 � 0.2 1 2.0 0.82 � 0.27 0.11 � 0.53
SSH (rebound) 80 91.8 7.2 99.0 58.9 � 2.6 23 521.8 0.28 � 0.27 0.65 � 0.53
SHH (rebound) 80 48.3 16.9 65.1 46.5 � 1.3 12 135.1 1.00 � 0.27 0.31 � 0.53
HB (kg/mm2) 80 813.0 1.9 814.9 156.7 � 16.9 151 22879.5 1.90 � 0.27 4.62 � 0.53
HV (kg/mm2) 80 573.4 7.5 580.8 123.2 � 13.5 121 14684.4 1.87 � 0.27 3.53 � 0.53
WA (%) 80 39.7 0.0 39.7 3.4 � 0.8 7 47.6 3.16 � 0.27 11.2 � 0.53
PG (%) 80 49.0 0.0 49.0 5.7 � 1.1 10 94.8 2.50 � 0.27 6.15 � 0.53
ISI (%) 80 53.6 39.3 92.8 74.9 � 1.4 13 156.6 0.85 � 0.27 0.39 � 0.53
CRS (%) 80 6.7 0.3 7.0 2.1 � 0.2 2 2.4 1.29 � 0.27 1.20 � 0.53
CI (%) 80 41.7 51.6 93.2 72.4 � 1.1 10 97.3 0.05 � 0.27 0.46 � 0.53
f (*) 80 320.0 6.0 326.0 88.7 � 8.5 76 5784.7 1.55 � 0.27 2.19 � 0.53

Fracture toughness can be defined as the ability of the rock to resist simplicity of the test comes from the fact that the calculations are in­
the fracturing and propagation of pre-existing cracks. In this study, the dependent of the dimension of the indenter and the hardness of the
semi-circular bend method proposed by Chong and Kuruppu39 was used. applied material is insignificant. The HV test device, which is used for
Semi-circular core samples with a 20 mm radius (r), 10 mm thickness (t) different rocks and has wide scales. Loads ranging from 1 to 50 kg were
and 10 mm crack length (a) were used for the test. KIC values were applied to prepare rock samples for 10–15 s. The two diagonal values of
calculated using the following equation. the indenter’s mark remaining on the rock surface were measured with a
pffiffiffiffiffiffi microscope and the average of these two values was calculated. The
Pmax π:a
KIC ¼ :YI (2) sloping surface area of the indentation was calculated. The HV value was
obtained by dividing the load (kg) by the calculated area (mm2).
2:r:t

where; KIC: fracture toughness (MN/m3/2), Pmax: maximum load (MN), Drilling speed should be predicted for the drilling plan. A factor of
a: crack length (m), r: disc radius (m), t: disc thickness (m) ve YI: rock in drilling is a drillability, which is quantified by CRS. The test was
dimensionless stress intensity factor. carried out using two aggregate pieces having a grain size of between
For the SHH, an N-type Schmidt hammer was used. Before coring 25.4 mm and 19.1 mm. A hammer having 2.4 kg weight was dropped
operations, the test was conducted on a flat rock block surfaces. The from a height of 63.5 cm on the selected aggregates and this operation
Schmidt hammer was pointed perpendicularly and touched the surface was repeated 5 times with new test samples and different drop numbers.
of the rock. ISRM35 suggested the method was followed in the tests. In The crushed aggregate pieces were sieved from 0.5 mm sieve. The
order to determine the SHH value of the rocks, 20 strokes were applied smallest value obtained by calculations among the five tests was
to the smooth surfaces of the block-shaped rocks and the average of the accepted as the CRS value.
highest 10 values was calculated. A correction factor (standard anvil For the CI test, 500 g of dry aggregate was used between 10 mm and
value/average of 10 readings taken over anvil) was used to correct the 5 mm. For the test, 500 gr dried aggregate having 10 þ 5 mm grain size
measured SHH values. To measure the SSH of the rocks used in this was used. Crushed test aggregates were fed to a laboratory type jaw
study, the C-2 model non-destructive measuring instrument was used in crusher having 5 mm outlet opening, then the crushed aggregates were
accordance with the rules specified in the ISRM35 standard. The test was sieved from a 5 mm sieve. The ratio of the weight of the aggregate under
performed on placed rock samples, which have a 14 square centimeter the sieve to the 500 gr weight was defined as CI value. The ISI test used
area and 2 cm thickness inside the cradle. Upper areas of cores were for classifying coal seams was developed by Protodyakonov for the first
polished with emery. The diamond-tipped hammer of the SSH test de­ time. The test was then modified by several researchers.40–42 The ISI test
vice was fallen vertically from 30 cm height on the surface of the rock was carried out using 100 gr aggregate ( 9.525 þ 3.175 mm) placed in a
specimen. Measurements were repeated 20 times at a spacing of a 42.86 mm diameter cylinder. The specimens were crushed by dropping a
minimum of 5 mm to represent the complete surface of the sample. The 1.8 kg hammer from a height of 30.48 cm (20 times). The aggregate
average of the top ten measurement results was calculated and this value weight on the 3.175 mm sieve was used as the ISI value of the aggregates
was taken as the SSH value of related rock. (repeated 3 times).
HB is a widely used test method for metals, but it is generally not One of the non-destructive experiments used for predictability of the
preferred for rocks because of its brittle nature. It is a measure of the mechanical properties of rocks and concretes are the UPV test and
rock resistance to indentation and has a direct correlation to rock designed for measuring the dynamic properties of these materials. This
strength. The ends of the cores having a thickness of 30 and a diameter of test is a non-destructive test and conducted by measuring the pulse ve­
42 mm were cut and flattened parallel to each other. The HB tests were locity between the transducer of the receiver and emitter. For the test,
performed by a 10 mm diameter spherical steel ball that was in contact both ends were parallel and polished cylinder rock samples were used.
with the rock sample. Testing loads of 50 kg–3000 kg were applied for The samples for this test had a cylindrical shape with a 42 mm diameter
30 s standard time on the rocks surfaces. The diameter of the indentation and 85 mm length. The gel was used to ensure full contact between the
area was measured and the applied load recorded. The hardness value sample and the transducers (receiver and transmitter). After this pro­
was determined from the ratio of applied load to the surface area of the cedure, the travel time of longitudinal P-wave was recorded in micro­
indentation and was expressed as kg/mm2. HB values were calculated seconds. The UPV was calculated by dividing the sample length via the
using the average of three tests. The HV test, which is easier than other travel time of P-wave.
hardness tests, developed as an alternative to the HB method. The Five samples from each rock were prepared. Each sample weighed

4
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

between 50 to 75 gr and consisted of rock pieces greater than 15 mm in


103
size. The sample was poured into the cylinder and a drop weight of 2.4 f ¼ (3)
I
kg was allowed to fall 5 times from a fixed height of 640 mm. Then, the
crushed aggregate was removed and the next sample was put in the where f is the Protodyakonov index, I is the height of dust in the vol­
cylinder, the process was repeated. After all the 5 samples were crushed, umemeter tube.
the broken material was put on a 0.5 mm (30 mesh) sieve. Fines passed Saturation techniques were used for testing such as PG and WA tests.
through the sieve were poured into the tube of the volume meter. The The water absorptions by the weight were calculated as the ratio of the
height of the dust in the tube was noted after tapping it gently ten times weight difference of the saturated-surface-dry weight and the dry weight
on the table. The Protodyakonov index was calculated by: of the sample to the weight of the dry sample. The pore volumes were
determined from the weight difference of the saturated weight (surface
dry) and the dry weight, and the PG values were calculated as the ratio

Fig. 2. Relationship between CAI and basic mechanical tests a) UCS b) E c) BTS d) Is50 e) BSA f) ROP g) BPI and h) KIC.

5
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

of pore volumes to bulk sample volume. contain the maximum and minimum confidence interval lines as well as
the maximum and minimum prediction lines calculated for the 95%
4. Statistical analysis and results confidence interval. The scatter diagrams shown in Figs. 3–10 represents
the main rock properties, respectively. When the figures are examined in
One of the frequently used methods to obtain estimation models detail, it will be seen that multiple prediction models are more reliable
between rock properties is the regression analysis, in which both simple than simple regression equations.
and multiple regression analyses were used. Both methods have been All multiple regressions revealed in this study were successful at a
used in this study. SPSS v.17.0. computer package was used for all sta­ 95% confidence level for the F test. When the equations obtained from
tistical analyses including Anova, F-tests, and t-tests. Firstly, simple multiple regression analyses and their reliability analyses are examined
regression analyses were performed using the linear, power, logarithmic in detail, it is seen that the most reliable results and high correlation
and exponential curve fitting approximations (Fig. 2 and Table 3) then, a coefficients can be obtained with UCS, BTS, E and BPI experiments.
series of multiple regression analyses summarized in Table 4 were per­ Following these, the most reliable results were obtained with the ROP
formed. The equations obtained from multiple regressions for the esti­ test. Among the eight different tests, the least reliable results belong to
mation of basic mechanical parameters were given in Table 4 (Eqs. (1.2) KIC, BSA, and Is50 tests, respectively.
- (8.4)). Some relations between the measured and predicted values When Table 4 is examined in detail, it is clear that multiple equations
were shown in Figs. 3–10, respectively. can be used for more accurate and strong estimations. Because multiple
For the simple regression analysis, the least-squares technique was models include more than two independent variables, the correlation
used and the details of the obtained models were given in Table 3. To be coefficients, and estimation capabilities are higher than the simple
able to show the relationships between the main properties of rocks and models. In Figs. 3–10, the measured and predicted values were corre­
CAI, regression curves were drawn. In Fig. 2, correlation graphs between lated, respectively. As can be seen in these figures, good predictions
UCS, E, BTS, Is50, KIC, BSA, ROP, BPI, and CAI tests are given. It is seen were obtained using multiple regression models. Correlation coefficients
that the mentioned rock mechanic tests give a very significant correla­ range from 0.72 to 0.96.
tion with the basic strength parameters. These experimental methods In addition to the statistical studies in which all the data were used,
which are specifically developed to define a rock were used as depen­ some classification methods, which are detailed in Table 5, were used.
dent variables in simple and multiple regression studies. Correlation To distinguish variability due to properties such as data set, material
coefficients obtained from the best equations related CAI vary between inhomogeneity and measurement reproducibility, main rock properties
0.655 and 0.804. The weakest correlation between these eight tests was are grouped according to the same nominal values. Thus, the measure­
obtained in the KIC test (Eq. 8.1). In the correlation studies using the CAI ment capability of each method was tried to be understood. Existing
test, the strongest relationships were obtained by UCS, Is50 and BPI tests, classification systems have been re-adapted to form nominal groups
respectively. containing sufficient rocks from the data set.
95% confidence interval values were calculated for the validity of the Classification 1: The rocks were divided into two main groups based
equations. The t-test was used to determine the significance level of the on the CAI classification in Table 1, which is greater than medium
correlation coefficients of the equations and it was seen that the ob­ abrasive (>3.0) and lower than medium abrasive (<3.0).
tained t-values were greater than the tabulated t-values. It was also Classification 2: The porosity classification, which is divided into six
observed that the significance of the t-values (p-value) was <0.05 value. categories (very compact < 1%; low porous, 1–2.5%; medium porous,
Regression analysis of variance was used to determine the importance of 2.5–5%; highly porous, 5–10%; very porous, 10–20%; very high porous,
regressions. It has been seen that F-values of all equations have > 20%), is used as three categories in this study. The first is compact
considerably high values than the tabulated F-values. Derived equations rocks (<1 %), the second is low porosity rocks (1–2.5 %) and the third is
that were proven to be reliable with F and t-tests can be reliably used, porous rocks (>2.5 %).
especially for predictive purposes. Classification 3: The UPV classification, which is divided into five
Multiple regression analyses were performed using two, three and categories (very low < 2.5 km/s; low, 2.5–3.5 km/s; medium, 3.5–4 km/
four independent variables with the condition that one of the indepen­ s; high, 4–5 km/s; very high, > 5 km/s), is used as three categories in this
dent variables was a CAI test. For multiple regression analysis, some study. The first is low-medium velocity (<4 km/s), the second is high
physical and index test methods (UVP, HB, HV, SHH, SSH, CRS, CI, f, P, velocity (4–5 km/s) and the third is very high velocity (>5 km/s).
and WA) that can be easily applied using simple test instruments and Classification 4: Another classification applied is rock origin classifi­
very few samples were used as independent variables with the CAI test. cation. The rocks are divided into three main groups as igneous
For estimating the main properties, the most reliable and meaningful (plutonic þ subvolcanic þ volcanic þ pyroclastic), metamorphic, and
multiple regression equations that could be obtained by the statistical sedimentary.
analysis were summarized in Table 4. In addition to F and t-tests, it was The most important finding to be mentioned in the evaluation ac­
also examined whether there is a multicollinearity problem (VIF) to cording to the CAI classification is the rather low correlation coefficient
make a stronger validation in multiple regression equations (Table 4). of the equation obtained from the ROP method applied to high abrasive
Scatter plots (Figs. 3–10) of measured and estimated values were plotted rocks. This is the presence of many rocks with different origin and
to see the predictability of some multiple equations. The scatter plots strength characteristics with the same CAI value in the high abrasive

Table 3
Validation of the derived simple models of rocks (F-test and t-test).
Equations no Independent variable Simple equations Best fit model F value |t | independent p |t | constant p R2
value value

Eq. 1.1 UCS UCS ¼ 51.82CAI-31.58 Linear 319 17.87 0.00 3.86 0.00 0.80
Eq. 2.1 E E ¼ 4.84CAI1.32 Power 159 12.59 0.00 9.72 0.00 0.67
Eq. 3.1 BTS BTS ¼ 2.8CAI1.086 Power 221 14.85 0.00 13.97 0.00 0.74
Eq. 4.1 Is50 Is50 ¼ 0.475e1.33CAI Exponential 310 17.62 0.00 13.14 0.00 0.80
Eq. 5.1 BSA BSA ¼ 55.2CAI 1.16 Power 162 12.71 0.00 11.21 0.00 0.67
Eq. 6.1 ROP ROP ¼ 57.12CAI 2.49 Power 203 14.23 0.00 5.85 0.00 0.72
Eq. 7.1 BPI BPI ¼ 6.59CAI-2.54 Linear 227 15.05 0.00 2.06 0.04 0.74
Eq. 8.1 KIC KIC ¼ 0.38CAI1.24 Power 148 12.17 0.00 9.99 0.00 0.66

6
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Table 4
Validation of the derived multiple models of rocks.
Eqs. no Multiple equations Adj. R2 Fvalue Ftabulated VIFmax ttabulated |t|min. sig|t|min. Durbin-Watson

Eq. 1.2 UCS ¼ 11.31CAI þ0.18HV þ 0.97SHH þ16.51CRS - 27.72 0.96 433 2.49 5.39 1.9921 3.77 0.00 2.15
Eq. 1.3 UCS ¼ 16.55CAI þ0.99SHH þ25.55CRS - 39.1 0.95 394 2.73 4.38 1.9917 4.60 0.00 2.22
Eq. 1.4 UCS ¼ 24.04CAI þ25.96CRS - 13.17 0.93 501 3.12 2.85 1.9913 2.52 0.01 2.32

Eq. 2.2 E ¼ 1.04CAI þ0.03HV þ 3.09UPV þ2.06CRS - 5.88 0.89 163 2.49 5.39 1.9921 1.51 0.14 2.26
Eq. 2.3 E ¼ 1.59CAI þ0.05HV þ 3.34UPV - 6.34 0.88 191 2.73 3.27 1.9917 2.25 0.03 1.99
Eq. 2.4 E ¼ 5.23CAI þ3.9UPV - 12.42 0.80 157 3.12 1.60 1.9913 6.23 0.00 1.60

Eq. 3.2 BTS ¼ 0.81CAI þ0.01HB - 0.08P þ 1.1CRS þ2.87 0.90 174 2.49 5.03 1.9921 3.02 0.00 1.97
Eq. 3.3 BTS ¼ 1.19CAI þ0.01HB þ 1.05CRS þ1.48 0.88 190 2.73 5.01 1.9917 3.06 0.00 1.69
Eq. 3.4 BTS ¼ 1.58CAI þ0.02HB þ 1.53 0.85 230 3.12 2.81 1.9913 2.88 0.01 1.49

Eq. 4.2 Is50 ¼ 0.86CAI þ0.01HV - 0.04ISI þ0.35CRS - 1.43 0.86 120 2.49 5.77 1.9921 1.40 0.17 1.53
Eq. 4.3 Is50 ¼ 0.90CAI þ0.01HV þ 0.05ISI – 1.91 0.85 155 2.73 4.22 1.9917 1.94 0.06 1.48
Eq. 4.4 Is50 ¼ 1.25CAI þ0.01HV þ 0.55 0.84 212 3.12 3.00 1.9913 1.39 0.17 1.44

Eq. 5.2 BSA ¼ 1.31CAI þ2.03WA - 0.29ISI - 0.73SHH þ80.62 0.84 104 2.49 4.57 1.9921 0.68 0.50 2.13
Eq. 5.3 BSA ¼ 2.62CAI þ2.18WA - 0.81SHH þ65.58 0.84 136 2.73 4.35 1.9917 1.47 0.15 2.14
Eq. 5.4 BSA ¼ 7.20CAI þ2.77WA þ 37.94 0.81 170 3.12 1.38 1.9913 5.31 0.00 2.17

Eq. 6.2 ROP ¼ 1.22CAI þ5.68WA - 0.94ISI þ5.7UPV þ51.82 0.88 146 2.49 3.89 1.9921 0.52 0.61 1.66
Eq. 6.3 ROP ¼ 0.96CAI þ5.12WA - 0.67ISI þ59.09 0.87 182 2.73 3.24 1.9917 0.40 0.69 1.54
Eq. 6.4 ROP ¼ 5.06CAI þ5.59WA þ 17.59 0.86 251 3.12 1.38 1.9913 2.66 0.01 1.52

Eq. 7.2 BPI ¼ 1.89CAI þ0.02HB þ 0.11ISI þ1.7CRS - 5.3 0.87 138 2.49 5.47 1.9921 1.97 0.05 1.73
Eq. 7.3 BPI ¼ 0.89CAI þ0.02HB þ 2.08CRS þ0.89 0.87 170 2.73 5.01 1.9917 0.88 0.38 1.74
Eq. 7.4 BPI ¼ 3.18CAI þ3.19CRS - 0.28 0.85 223 3.12 2.85 1.9913 0.28 0.78 2.04

Eq. 8.2 KIC ¼ 0.14CAI þ0.09UPV - 0.03CI þ 0.004f þ 2.68 0.76 64 2.49 4.97 1.9921 1.61 0.11 1.90
Eq. 8.3 KIC ¼ 0.14CAI - 0.04CI þ 0.004f þ 3.56 0.76 82 2.73 4.42 1.9917 1.65 0.10 1.91
Eq. 8.4 KIC ¼ 0.21CAI - 0.06CI þ 4.88 0.72 102 3.12 3.62 1.9913 2.37 0.02 1.78

Fig. 3. Comparison of UCSP and UCSM values for Eq. 1.2. Fig. 4. Comparison of EP and EM values for Eq. 2.2.

rocks category according to the CAI classification (Fig. 11b). Further­ combining two clusters of points; for two quartzites with a strength
more, the ROP value of the rocks in this group varies within a very value higher than 200 MPa and ten rocks with a strength value of less
narrow range due to the presence of quartz minerals in the rocks. In than 100 MPa (Fig. 11e). Correlation coefficients of sedimentary rocks
rocks with low abrasive, the effect of the origin difference on the drilling are relatively low. The main reason for this is the very low presence of
progression is not observed much (Fig. 11a). The rocks in this group corrosive minerals in sedimentary rocks. Two rocks with very different
have a wide range of ROP and UCS values. In the group with a narrow strengths CAI values can be the same.
range according to the UPV classification (high velocity, 4–5 km/s), the
low correlation coefficient for Bohme surface abrasion (BSA) is notice­ 5. Comparisons of the performance of derived models
able (Fig. 11d). While 20 rocks in this group show very wide scale
strength values, they have BSA and UPV values in a narrower range. The estimation values obtained from multiple regression analyses
When the correlation coefficients obtained from the correlation and whether the measured values have normal distribution were
studies based on rock origin are examined, it is seen that the best results examined. Five parameters are taken into consideration when deciding
belong to the group of igneous rocks consisting of 50 rocks (Fig. 11f). All the normal distribution. These include the coefficients of variation,
main rock properties show a strong relationship with the CAI. The cor­ histogram graph, detrended Q-Q plots, Kurtosis-Skewness, and
relation coefficients obtained for metamorphic rocks are also rather high Kolmogorov-Smirnov normality test. As a result of evaluating a total of
and there is a linear relationship. However, the fact that there is almost five parameters, one of them is the measured value, it is observed that
no metamorphic rock between 100 MPa and 200 MPa reduces the reli­ most of the equations have a normal distribution and very few have a-
ability of the derived equations. However, the graph is essentially normal distribution. This procedure was repeated for eight main rock

7
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Fig. 5. Comparison of BTSP and BTSM values for Eq. 3.2. Fig. 8. Comparison of ROPP and ROPM values for Eq. 6.2.

Fig. 6. Comparison of Is50P and Is50M values for Eq. 4.4. Fig. 9. Comparison of BPIP and BPIM values for Eq. 7.2.

Fig. 7. Comparison of BSAP and BSAM values for Eq. 5.2. Fig. 10. Comparison of KICP and KICM values for Eq. 8.2.

properties. After the normal distribution examination, the data set was is close to perfect.
analyzed using parametric and nonparametric methods. Assuming the a-normal distribution of data; the Welch Anova
Assuming the normal distribution of data; the homogeneity of the method was used for the nonparametric test. Tamhane’s T2 test (post-
variations of the measured and estimated values was tested by the Anova hoc) was used for comparison of multiple tests to investigate the re­
method (Table 6). There was no difference between the mean values (F lationships between the measured and estimated values. The variations
statistical significance level of 0.97–1.00) in all groups (Table 7). The of these values were also tested using one-way analysis of variance
significance level approaching 1.0 indicates that variance homogeneity (Welch Anova) and were seen to be homogeneous (Welch statistic

8
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Table 5
Classification systems and correlation coefficients according to categories.
Classification All rocks CAI Rock porosity (%) UPV (km/sec) Rock origin

Parameters <3.0 >3.0 <1.0 1–2.5 >2.5 <4 4–5 >5 Metamorphic Sedimentary Igneous

low high low-med. medium high low-med. high very high

Rock Number 80 49 31 28 23 29 22 20 29 12 18 50
% 100 61.25 38.75 35 28.75 36.25 27.5 25 47.5 15 22.5 62.5
Nominal value * (CAI-I) (CAI-II) (P–I) (P-II) (P-III) (UPV–I) (UPV-II) (UPV-III) (O–I) (O-II) (O-III)

UCS 0.80LN 0.66PW 0.64LN 0.83LN 0.76PW 0.72LN 0.69LN 0.70EX 0.81EX 0.91LN 0.60PW 0.86PW
E 0.67PW 0.53PW 0.50LN 0.73LN 0.68PW 0.64PW 0.53EX 0.27EX 0.65EX 0.75LN 0.40PW 0.86PW
BTS 0.74PW 0.61PW 0.62LN 0.81EX 0.79PW 0.62PW 0.60PW 0.50LN 0.68LN 0.86LN 0.40PW 0.85PW
Is50 0.80EX 0.62EX 0.40PW 0.80EX 0.75LN 0.81EX 0.61LN 0.68EX 0.77EX 0.84LN 0.62LN 0.86EX
BSA 0.67PW 0.53LG 0.26EX 0.67EX 0.45LN 0.64PW 0.64PW 0.19EX 0.56EX 0.92EX 0.43PW 0.84EX
ROP 0.72PW 0.65PW 0.16EX 0.74LG 0.71EX 0.70EX 0.57EX 0.62LG 0.77LG 0.81EX 0.48LG 0.76PW
BPI 0.74LN 0.61PW 0.61LN 0.84LN 0.70PW 0.63PW 0.50PW 0.43EX 0.77EX 0.84LN 0.34PW 0.82PW
KIC 0.65PW 0.47PW 0.36PW 0.66LN 0.58PW 0.62PW 0.40PW 0.32EX 0.63EX 0.81LN 0.28PW 0.79PW

LN: linear, PW: power, LG: logarithmic, EX: exponential.

Fig. 11. Comparison of some simple equations according to classification systems a) CAI-I, b) CAI-II, c) UPV-I, d) UPV-II, e) O–I and f) O-III

significance levels change between 0.91-1.00). According to Tamhane’s 1 for all equations) > 0.005, ie the groups were homogeneously
T2 test (post-hoc) test results, it can be said that there is no significant distributed (Table 6).
difference between the groups because the significance values (equal to Dunnett’s two-sided T-test (post-hoc) was used to determine the

9
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Table 6
Homogeneity of variances and Welch Anova test.
Parameters Test of homogeneity of variances Robust tests of equality of means

LEVENE STATISTICS WELCH ANOVA

Levene Statistic df1 df2 Sig. Statistica df1 df2 Sig.

UCS 0.024 4 395 0.999 0.000 4 197.4 1.000


E 0.346 4 395 0.847 0.000 4 197.2 1.000
BTS 0.147 4 395 0.964 0.000 4 197.4 1.000
IS50 0.722 4 395 0.577 0.003 4 197.4 1.000
BSA 0.603 4 395 0.660 0.023 4 196.9 0.999
ROP 1.028 4 395 0.392 0.243 4 195.6 0.914
BPI 0.224 4 395 0.925 0.000 4 197.4 1.000
KIC 0.692 4 395 0.598 0.001 4 197.3 1.000
a
Asymptotically F distributed.

Table 7
Anova test.
Parameters Sum of Squares df Mean Square F sig.

UCS Between Groups 0 4 0 0.000 1.000


Within Groups 1632188 395 4132
Total 1632188 399

E Between Groups 0 4 0 0.000 1.000


Within Groups 44686 395 113
Total 44686 399

BTS Between Groups 0 4 0 0.000 1.000


Within Groups 6979 395 18
Total 6979 399

Is50 Between Groups 0 4 0 0.003 1.000


Within Groups 3371 395 9
Total 3371 399

BSA Between Groups 42 4 11 0.018 0.999


Within Groups 236063 395 598
Total 236106 399

ROP Between Groups 1003 4 251 0.153 0.961


Within Groups 646035 395 1636
Total 647038 399

BPI Between Groups 0 4 0 0.000 1.000


Within Groups 26765 395 68
Total 26765 399

KIC Between Groups 0 4 0 0.001 1.000


Within Groups 252 395 1
Total 252 399

relationships between estimation and measurement values. The percentages. The relative error is calculated as the absolute error divided
measured values (J) were taken as the control group for each main rock by the magnitude of the measured value. The absolute error is the
properties used in the study (Table 8; Fig. 12). The significance value to magnitude of the difference between the measurement and the estimate.
be used in the post hoc tests should be revised according to the total According to the percentage error analysis, the best equations for each
number of compared groups. The new significance value can be calcu­ rock feature are; Eq. (1.3), Eq. (2.2), Eq. (3.2), Eq. (4.4) Eq. (5.3), Eq.
lated using the following equation: (6.4), Eq. (7.4) and Eq. (8.3). Among the eight equations given above,
the equation with the lowest error rate is the one developed for the UCS
0:05
NSV ¼ � � (4) test, while the equation with the highest error rate is related to ROP. As
n:ðn 1Þ
2
can be seen from Table 8, Eq. (1.3) has an error value of less than 10%
for 38 samples (47.5% of 80 samples), while Eq. (6.4) has an error value
of less than ten percent for 14 samples (17.5% of 80 samples).
where NSV is the new sig. value, 0.05 is the 95% confidence interval
To check the capacity performance of all predictive models, some
value, n is the number of compared groups. This value is calculated as
statistical performance indices (the root mean square error to observa­
0.005 for all properties.
tion’s standard deviation ratio-RSR; weighted mean absolute percentage
According to the Dunnett test, each pair of data had very close
error-WMAPE; the variance accounted for-VAF; adjusted determination
average differences, equal standard error, and significance levels. One
coefficient-Adj.R2) were calculated for each model according to the
can infer from the findings that the main properties of rock materials can
formulas given below.
be reliably estimated from any equations considering the correlation and
� � � �
Anova analysis results. VAF
Percentage error analysis was performed after correlation and Anova PIat1 ¼ Adj:R2 þ RSR WMAPE (5)
100
analyses (Table 9). These are relative errors and expressed as

10
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Table 8
Multiple comparisons of the measured and predicted test values.
Multiple Comparisons Mean Difference (I-J) Std. Error Dunnett t (2-sided)a

Eqs. No (I) Groups (J) Groups sig. 99.5% Confidence Interval

Lower Bound Upper Bound

Eq. 1.1 UCSP UCSM 0.0008 10.16 1.000 32.84 32.84


Eq. 1.2 UCSP UCSM 0.0018 10.16 1.000 32.84 32.84
Eq. 1.3 UCSP UCSM 0.0058 10.16 1.000 32.84 32.83
Eq. 1.4 UCSP UCSM 0.0020 10.16 1.000 32.84 32.84

Eq. 2.1 EP EM 0.0011 1.68 1.000 5.44 5.43


Eq. 2.2 EP EM 0.0014 1.68 1.000 5.43 5.44
Eq. 2.3 EP EM 0.0001 1.68 1.000 5.43 5.43
Eq. 2.4 EP EM 0.0026 1.68 1.000 5.43 5.44

Eq. 3.1 BTSP BTSM 0.0000 0.66 1.000 2.15 2.15


Eq. 3.2 BTSP BTSM 0.0037 0.66 1.000 2.15 2.14
Eq. 3.3 BTSP BTSM 0.0012 0.66 1.000 2.15 2.15
Eq. 3.4 BTSP BTSM 0.0025 0.66 1.000 2.15 2.15

Eq. 4.1 Is50P Is50M 0.0393 0.46 1.000 1.45 1.53


Eq. 4.2 Is50P Is50M 0.0018 0.46 1.000 1.49 1.49
Eq. 4.3 Is50P Is50M 0.0033 0.46 1.000 1.50 1.49
Eq. 4.4 Is50P Is50M 0.0008 0.46 1.000 1.49 1.49

Eq. 5.1 BSAP BSAM 0.8186 3.87 0.999 11.67 13.31


Eq. 5.2 BSAP BSAM 0.0049 3.87 1.000 12.48 12.49
Eq. 5.3 BSAP BSAM 0.0224 3.87 1.000 12.47 12.51
Eq. 5.4 BSAP BSAM 0.0039 3.87 1.000 12.49 12.48

Eq. 6.1 ROPP ROPM 3.9579 6.39 0.927 16.70 24.62


Eq. 6.2 ROPP ROPM 0.0034 6.39 1.000 20.66 20.66
Eq. 6.3 ROPP ROPM 0.0016 6.39 1.000 20.66 20.66
Eq. 6.4 ROPP ROPM 0.0004 6.39 1.000 20.66 20.66

Eq. 7.1 BPIP BPIM 0.0046 1.30 1.000 4.20 4.21


Eq. 7.2 BPIP BPIM 0.0004 1.30 1.000 4.21 4.21
Eq. 7.3 BPIP BPIM 0.0034 1.30 1.000 4.20 4.21
Eq. 7.4 BPIP BPIM 0.0041 1.30 1.000 4.21 4.20

Eq. 8.1 KICP KICM 0.0014 0.13 1.000 0.41 0.41


Eq. 8.2 KICP KICM 0.0001 0.13 1.000 0.41 0.41
Eq. 8.3 KICP KICM 0.0051 0.13 1.000 0.40 0.41
Eq. 8.4 KICP KICM 0.0026 0.13 1.000 0.41 0.41
a
Dunnett t-tests treat one group as a control, and compare all other groups against it.

Fig. 12. Comparison of the mean values of the main mechanical tests obtained from different methods.

ðn 1Þ � � �
Adj:R2 ¼ 1 1 R2 (6) Pn ��ti oi ��
ðn p 1Þ i¼1 � ti �
WMAPE ¼ Pn (8)
� � i¼1 ti
varðti oi Þ
VAF ¼ 1 :100 (7)
varðti Þ

11
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

Table 9
Percentage error and PIat1 values for predicted methods.
Eqs. number Percentage error �10 % Differences between Differences PIat1

Number of samples Percent of samples (%) Best Eq. and other Eqs. (%)

Eq. 1.1 20 25.0 (Eq. 1.3-Eq. 1.1) 22.5 1.16


Eq. 1.2 36 45.0 (Eq. 1.3-Eq. 1.2) 2.5 1.71
Eq. 1.3 38 47.5 (Eq. 1.3-Eq. 1.3) 0.0 1.63
Eq. 1.4 33 41.3 (Eq. 1.3-Eq. 1.4) 6.3 1.59

Eq. 2.1 10 12.5 (Eq. 2.2-Eq. 2.1) 27.5 0.71


Eq. 2.2 32 40.0 (Eq. 2.2-Eq. 2.2) 0.0 1.45
Eq. 2.3 29 36.3 (Eq. 2.2-Eq. 2.3) 3.8 1.40
Eq. 2.4 21 26.3 (Eq. 2.2-Eq. 2.4) 13.8 1.14

Eq. 3.1 24 30.0 (Eq. 3.2-Eq. 3.1) 8.8 0.94


Eq. 3.2 31 38.8 (Eq. 3.2-Eq. 3.2) 0.0 1.47
Eq. 3.3 29 36.3 (Eq. 3.2-Eq. 3.3) 2.5 1.39
Eq. 3.4 30 37.5 (Eq. 3.2-Eq. 3.4) 1.3 1.30

Eq. 4.1 20 25.0 (Eq. 4.4-Eq. 4.1) 11.3 0.98


Eq. 4.2 25 31.3 (Eq. 4.4-Eq. 4.2) 5.0 1.26
Eq. 4.3 26 32.5 (Eq. 4.4-Eq. 4.3) 3.8 1.30
Eq. 4.4 29 36.3 (Eq. 4.4-Eq. 4.4) 0.0 1.32

Eq. 5.1 14 17.5 (Eq. 5.3-Eq. 5.1) 16.3 0.58


Eq. 5.2 22 27.5 (Eq. 5.3-Eq. 5.2) 6.3 1.29
Eq. 5.3 27 33.8 (Eq. 5.3-Eq. 5.3) 0.0 1.27
Eq. 5.4 17 21.3 (Eq. 5.3-Eq. 5.4) 12.5 1.18

Eq. 6.1 1 1.3 (Eq. 6.4-Eq. 6.1) 16.3 0.17


Eq. 6.2 12 15.0 (Eq. 6.4-Eq. 6.2) 2.5 1.35
Eq. 6.3 8 10.0 (Eq. 6.4-Eq. 6.3) 7.5 1.31
Eq. 6.4 14 17.5 (Eq. 6.4-Eq. 6.4) 0.0 1.31

Eq. 7.1 22 27.5 (Eq. 7.4-Eq. 6.1) 5.0 0.96


Eq. 7.2 25 31.3 (Eq. 7.4-Eq. 6.2) 1.3 1.39
Eq. 7.3 20 25.0 (Eq. 7.4-Eq. 6.3) 7.5 1.36
Eq. 7.4 26 32.5 (Eq. 7.4-Eq. 6.4) 0.0 1.30

Eq. 8.1 12 15.0 (Eq. 8.3-Eq. 8.1) 10.0 0.30


Eq. 8.2 19 23.8 (Eq. 8.3-Eq. 8.2) 1.3 0.82
Eq. 8.3 20 25.0 (Eq. 8.3-Eq. 8.3) 0.0 0.81
Eq. 8.4 18 22.5 (Eq. 8.3-Eq. 8.4) 2.5 0.64

RMSE follows; Eq. (1.2) (1.71-UCS); Eq. (2.2) (1.45-E); Eq. (3.2) (1.47-BTS);
RSR ¼ qffiffiP
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (9)
1 n 2 Eq. (4.4) (1.32-Is50); Eq. (5.2) (1.29-BSA); Eq. (6.2) (1.35-ROP); Eq.
i¼1 ðti ti Þ
n
(7.2) (1.39-BPI); Eq. (8.2) (0.82-KIC).
Table 9 shows the calculated PIat1 values for the equations developed
where; t is the target value obtained from the experimental studies. o is
using one, two, three, and four independent variables for each main rock
the output value of models. ti is the arithmetic mean of target variables
property. In principle, the comparison of PIat1 values of these four
obtained from the formulas. oi is the arithmetic mean of model vari­
equations was made according to the arithmetic mean. According to this
ables. var is variance. and n is the number of experimental study data. p
criterion, the most reliable correlations with the CAI were obtained from
denotes the model input quantity. PIat1 is the performance index value
UCS (1.52), BTS (1.27), and BPI (1.25) tests, while the lowest-
developed by author in this study based on the PI value recommended
performing tests were KIC (0.5), ROP (1.04), and BSA (1.08),
by Yagiz et all.43 VAF is the variance account factor. WMAPE measures
respectively.
the weighted mean absolute percentage error of the prediction. Adj.R2 is
the adjusted determination coefficient. RSR incorporates the benefits of
6. Conclusions
error index statistics and includes a scaling/normalization factor so that
the resulting statistic and reported values can apply to various constit­
In this study, a series of statistical studies have been carried out to
uents. The optimal value of RSR is zero.
estimate the basic rock mechanics test results carried out using samples
It is rather difficult to select a more accurate model for predicting the
prepared in specific sizes with the CAI test which is practiced in rock
independent variables. Essentially, the accuracy of models can be
wear.
examined separately using performance indices such as WMAPE, VAF,
In addition to presenting the relationship between some commonly
RMSE, RSR, and R2, but one of these indices is not superior to the others.
known parameters (UCS, E, BTS, and Is50) and CAI, it is aimed to
Four of these performance indices were combined in the same formula
contribute to the literature the practical equations concerning some
and the performance index given in Eq. (5) was calculated to compare
parameters (BSA, ROP, BPI, and KIC) which the researchers do not work
the performance of the developed models. Theoretically, for a model to
a lot.
be considered excellent, the Adj. R2 value must be “1”, the VAF value
In addition to the correlation studies using 80 rocks, the data set
“100”, the WMAPE value “0” and the RSR value must be “0”. Theoret­
grouped using various classification systems were evaluated in detail.
ically, the PIat1 value of excellent prediction models is equal to 2 as
The results from multiple and simple regression analysis performed with
expected. Therefore, the model with the highest average PIat1 value
test results were summarized in the following items.
should be the most reliable and accurate one. The most reliable and
In the simple regression analysis, the weakest correlation between
accurate equations according to the PIat1 values given in Table 9 are as
these eight tests was obtained from the KIC test (Eq. 8.1). In the

12
A. Teymen International Journal of Rock Mechanics and Mining Sciences 128 (2020) 104258

correlation studies using the CAI test, the strongest relationships were 18 Plinninger R, Kasling H, Thuro K. Wear prediction in hardrock excavation using the
Cerchar abrasivity index (CAI). In: Schubert, ed. Proc. EUROCK 2004 and 5rd
obtained by UCS, Is50 and BPI tests, respectively.
Geomechanics Colloquium VGE. Germany: Essen; 2004:599–604.
Equations obtained from multiple regression analyzes have higher 19 Yaralı O, Akçın NA. Investigation of relationships between Cerchar hardness index
predictive power and determination coefficient than simple equations. values and strength properties. In: Proceedings of the 19th International Mining
Simple equations are reliable and sufficiently practical when used Congress and Fair of Turkey (IMCET2005). Izmir,
_ Turkey. 2005:271–281, 09-12 June.
20 Michalakopoulos T, Anagnostou VG, Bassanou ME, Panagiotou GN. The influence of
especially estimation purposes. However, if more than one independent steel styli hardness on the Cerchar abrasiveness index value. Int J Rock Mech Min Sci.
variable is obtained, multiple equations can be used for more precise 2006;43:321–327. https://doi.org/10.1016/j.ijrmms.2005.06.009.
estimation. Several multiple regression equations have been developed 21 Lassnig K, Latal C, Klima K. Impact of grain size on the cerchar abrasiveness. Test.
Geomech. Tunn.. 2008:71–76 https://doi.org/10.1002/geot.200800008.
for the estimation of main rock properties. According to the determi­ 22 Alber M. Stress dependency of the Cerchar abrasivity index (CAI) and its effects on
nation coefficients, all equations have high prediction performance. wear of selected rock cutting tools. Tunn Undergr Space Technol. 2008;23(4):351–359.
Only some aggregate pieces are used in a part of the tests while very https://doi.org/10.1016/j.tust.2007.05.008.
23 Kahraman S, Alber M, Fener M, Gunaydin O. The usability of Cerchar abrasivity
small samples are needed in some of the tests used in the estimation of index for the prediction of UCS and E of Misis Fault Breccia: regression and artificial
main properties. Equations with high correlation coefficients presented neural networks analysis. Expert Syst Appl. 2010;37:8750–8756. https://doi.org/
in this study can be reliably used for rocks in situations where it is not 10.1016/j.eswa.2010.06.039.
24 Deliormanli AH. Cerchar abrasivity index (CAI) and its relation to strength and
possible to determine an important mechanical parameter such as UCS. abrasion test methods for marble stones. Construct Build Mater. 2012;30. https://doi.
It is clear that the estimation equations with two independent vari­ org/10.1016/j.conbuildmat.2011.11.023.
ables are more reliable than the others in multiple regression studies. 25 Dipova N. Investigation of the relationships between abrasiveness and strength
properties of weak limestones along a tunnel route. Geol. Eng. 2012;36(1):23–34.
26 Yaralı O. Investigation of the effect into mechanical properties of rocks on Cerchar
Abrasivity Index. Karaelmas Fen ve Müh. Derg. 2016;6(1):218–229.
Declaration of competing interest 27 Moradizadeh M, Cheshomi A, Ghafoori M, TrighAzali S. Correlation of equivalent
quartz content, Slake durability index and Is50 with Cerchar abrasiveness index for
The author declare that they have no conflict of interest. different types of rock. Int J Rock Mech Min Sci. 2016;86:42–47. https://doi.org/
10.1016/j.ijrmms.2016.04.003.
28 Er S, Tu� grul A. Estimation of Cerchar abrasivity index of granitic rocks in Turkey by
References geological properties using regression analysis. Bull Eng Geol Environ. 2016;75(3):
1325–1339. https://doi.org/10.1007/s10064-016-0853-y.
1 Deketh HJR. Wear of Rock Cutting Tools Laboratory Experiments on the Abrasivity of 29 Er S, Tu� grul A. Correlation of physico-mechanical properties of granitic rocks with
Rock. Rotterdam, Netherlands: AA Balkema; 1995:143. Cerchar Abrasivity Index in Turkey. Measurement. 2016;91:114–123. https://doi.
2 ISRM. Suggested methods for determining hardness and abrasiveness of rocks. Int J org/10.1016/j.measurement.2016.05.034.
Rock Mech Min Sci Geomech Abstr. 1978;15:89–97. 30 Majeed Y, Abu Bakar MZ. Statistical evaluation of CERCHAR Abrasivity Index (CAI)
3 McFeat-Smith I, Fowell RJ, Babtie S, Babtie M. Correlation of rock properties and the measurement methods and dependence on petrographic and mechanical properties
cutting performance of tunnelling machines. In: Proceedings Rock Eng. University of of selected rocks of Pakistan. Bull Eng Geol Environ. 2016;75(3):1341–1360. https://
Newcastle Upon Tyne; 1977:581–602. doi.org/10.1007/s10064-015-0799-5.
4 West G. A review of rock abrasiveness testing for tunnelling. In: Proceedings of the 31 Ko TY, Kim TK, Son Y, Jeon S. Effect of geomechanical properties on Cerchar
International Symposium on Weak Rock. 1. 1981:585–594. Tokyo. Abrasivity Index (CAI) and its application to TBM tunneling. Tunn Undergr Space
5 Evenden M. Methods of Determining Rock Hardness and Abrasivity. PhD Thesis. Technol. 2016;57:99–111. https://doi.org/10.1016/j.tust.2016.02.006.
Institute of Technology and Science, Univ. of Notthingham; 1983:134. 32 Yaralı O. Investigation into relationships between Cerchar hardness index and some
6 Atkinson T, Cassapy VB, Singh I. Assessment of abrasive wear resistance potential in mechanical properties of coal measure rocks. Geotech Geol Eng. 2017;35:1605–1614.
rock excavation machinery. Int J Min Geol Eng. 1986;3:151–163. https://doi.org/ https://doi.org/10.1007/s10706-017-0195-y.
10.1007/BF01560672. 33 Ozdogan MV, Deliormanli AH, Yenice H. The correlations between the Cerchar
7 The Cerchar Abrasivity Index. Verneuil: Centre d’E’ tudes et des Recherches des abrasivity index and the geomechanical properties of building stones. Arab. J. Geosci.
Charbonnages de France; 1986. 2018;11(20):604. https://doi.org/10.1007/s12517-018-3958-8.
8 AFNOR. NF P 94-430-1. Determination du pouvoir abrasif d’une roche-Partie 1: Essai de 34 Kadkhodaei MH, Ghasemi E. Development of a GEP model to assess CERCHAR
rayure avec une pointe. 2000. Paris. abrasivity index of rocks based on geomechanical properties. J. Min. Environ. 2019;
9 ASTM D7625-10. Standard Test Method for Laboratory Determination of Abrasiveness of 10(4):917–928. https://doi.org/10.22044/jme.2019.8141.1684.
Rock Using the CERCHAR Method. West Conshohocken. PA: ASTM International; 35 ISRM. Rock Characterization, Testing and Monitoring – Commission on Standardization
2010. https://doi.org/10.1520/D7625-10. Laboratory and Field Results. Suggested Methods for Determining Hardness and
10 ISRM. Suggested Method for Determining the Abrasivity of Rock by the Cerchar Abrasiveness of Rocks. Oxford: Pergamon; 1981:102–103. Part 4.
Abrasivity Test. The ISRM Suggested Method for Rock Characterization, Testing and 36 TSE 699. Methods of Testing for Natural Building Stones. Ankara, Turkey: Institute of
Monitoring: 2007-2014. In: Ulusay R, ed. USA: Springer; 2015:101–106. Turkish Standards (TSE); 1987.
11 Valantin A. Examen des diffe’rents proce’de’s classiques de de’termination de la nocivite’ 37 ASTM C 241-90. Standard Test Method for Abrasion Resistance of Stone Subjected to
des roches vis-a-vis de l’abattage me’canique. Expose Presente’ Aux Journe’es Foot Traffic. Annual Book of ASTM Standards, Ame. Soc. Testing Mater. (ASTM);
d’information Techniques de Creusement, les 28–29 Novembre. Luxemburg. 1974: 1990.
133–140. 38 Ulusay R, Gokceoglu C, Sulukcu S. Draft ISRM suggested method for determining
12 West G. Rock abrasiveness testing for tunnelling. Int J Rock Mech Min Sci. 1989;26: block punch strength index (BPI). Int J Rock Mech Min Sci. 2001;38(8):1113–1199.
151–160. https://doi.org/10.1016/0148-9062(89)90003-X. https://doi.org/10.1016/S1365-1609(01)00078-8.
13 Suana M, Peters T. The Cerchar Abrasivity Index and its relation to rock mineralogy 39 Chong KP, Kuruppu MD. New specimen for fracture toughness determination for
and petrography. Rock Mech. Felsmech. Mecanique Des. Roches. 1982;15(1):1–8. rock and other materials. Int J Fract. 1984;26:59–62. https://doi.org/10.1007/
https://doi.org/10.1007/BF01239473. BF01157555.
14 West G. A relation between abrasiveness and quartz content for some coal measures 40 Paone J, Madson D, Bruce WE. Drillability studies: laboratory percussive drilling.
sediments. Int J Min Geol Eng. 1986;4:73–78. https://doi.org/10.1021/jp3049982. USBM RI. 1969;7300.
15 Johnson ST, Fowell RJ. Compressive strength is not enough (assessing pick wear for 41 Tandanand S, Unger HF. Drillability determination: a drillability index of percussive
drag tool-equipped machines). In: Proceedings of the 27th US Symp Rock Mechanics. drills. 1975. USBM RI 8073.
Alabama, USA: Tuscaloosa; 1986:840–845. 42 Rabia H, Brook W. An empirical equation for drill performance prediction. In:
16 Al-Ameen SI, Waller MD. The influence of rock strength and abrasive mineral content Proceedings of the 21st US Symposium on Rock Mechanics. Univ Missouri-Rolla; 1980:
on the Cerchar Abrasive Index. Eng Geol. 1994;36:293–301. https://doi.org/ 103.
10.1016/0013-7952(94)90010-8. 43 Yagiz S, Sezer EA, Gokceoglu C. Artificial neural networks and nonlinear regression
17 Plinninger R, Kasling H, Thuro K, Spaun G. Testing conditions and geomechanical techniques to assess the influence of slake durability cycles on the prediction of
properties influencing the Cerchar abrasiveness index (CAI) value. Technical note. Int uniaxial compressive strength and modulus of elasticity for carbonate rocks. Int J
J Rock Mech Min Sci. 2003;40(2):259–263. https://doi.org/10.1016/S1365-1609(02) Numer Anal Methods GeoMech. 2012;36:1636–1650. https://doi.org/10.1002/
00140-5. nag.1066.

13

You might also like