International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Contents lists available at ScienceDirect

International Journal of Mechanical Sciences

journal homepage: www.elsevier.com/locate/ijmecsci

Detailed modeling of cutting forces in grinding process considering variable

stages of grain-workpiece micro interactions

Hao Nan Lia,b, , Tian Biao Yua, Zi Xuan Wanga, Li Da Zhua, Wan Shan Wanga
a
School of Mechanical Engineering and Automation, Northeastern University, Shenyang, PR China
b
Machining and Condition Monitoring Group, Faculty of Engineering, University of Nottingham, UK

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

Keywords: Grinding forces are a key parameter in the grinding process, most previous studies on grinding forces, however,
Grinding force (i) were regardless of grain-workpiece micro interaction statuses and (ii) could only predict average/maximal
Grinding process grinding forces based on average/maximal cutting depths or chip thicknesses. In this study, a novel detailed
Detailed modeling modeling methodology of grinding forces has been analytically established, experimentally validated and
Grain-workpiece contact
utilised to study a specic issue that previous methods can not address. Based on the proposed method, grinding
forces with detailed information (e.g. three components including rubbing, plowing and cutting forces) could be
accurately predicted. Except for grinding forces, the proposed methodology also enable the availability of other
grinding process details at the grain scale (e.g. the ratios of grains that are experiencing rubbing, plowing and
cutting stages to the total engaging grain number). Validation experiment results have proved that, the proposed
method could, to a large extent, describe the realistic grinding forces. Based on the proposed method, the eects
of grinding conditions (including depths of cut, wheel speeds, workpiece feed speeds and grinding wheel
abrasive sizes) on each component of grinding forces (rubbing, plowing, and cutting forces) have been analyzed.
Some new ndings, which could enhance the existing understandings of grinding forces and guide industrial
manufacture, have been gained. The proposed method therefore is anticipated to be not only meaningful to
provide a new way to model grinding forces in detail, but also promising to study other grinding issues (e.g.
grinding heat, machined surface topography, grinding chatter), especially under the trend of miniaturization and
microfabrication where grinding details at the grain scale are highly needed to optimise the micro grinding tool
eciency and micro-grinding accuracy.

1. Introduction and machining parameters) and the output grinding forces.

Grinding process could be considered as a kind of most widely-used
nishing operation in the manufacturing because of low cost, high
machining eciency and good nish quality [1]. Grinding forces are a
key element in grinding, inuencing material removal rates, machined
surface qualities, grinding temperature and vibrations, and further
wheel wear and service life [2]. Grinding forces are also an important
parameter that could be utilised to optimise machining parameters and
grinding machine and xture structures so that the potential of the
grinding process could be fully explored [3]. To this end, many eorts
have been made to try to understand grinding forces.
Because grinding wheel topography is of stochastic nature due to
the random grain distribution on the wheel surface, most early studies
on grinding forces focused on the establishment of empirical models, in
which grinding forces could be obtained by using mathematical
regression functions between the input data (e.g. the grain density
The pioneer empirical model presented by Werner and Koenig [4]
expressed the normal grinding force per unit wheel width (i.e. specic
grinding force) as Fn = k1(C )k2 (vs / vw )2k31(ap )k4 (ds )1 k5 . In this model,
grains were randomly distributed on the wheel surface and chips with
varied cross-section area were also considered based on grinding
kinematics. Bttler [5] suggested that Werner's model did not consid-
ered the eect of the increasing wheel wear during the grinding
process, therefore the author introduced an additional factor (specic
material removal Vm ) to Werners model to compensate this eect.
Tnsho et al. [6] comprehensively reviewed most empirical models
proposed before the 90s and provided the basic form of specic normal
grinding forces as Fn = k6k 7(vs / vw )k8(ap )k9(ds )k10 , which has been widely
used in the later industrial manufacturing. Recent study on the
empirical model was performed by Mishra and Salonitis [7], who
modied Werners model [4] by keeping the basic form whilst adding
the two-way sensitivity analysis in the regression calculation.

Corresponding author at: School of Mechanical Engineering and Automation, Northeastern University, Shenyang, PR China.
E-mail address: lhnlwfb@163.com (H.N. Li).

http://dx.doi.org/10.1016/j.ijmecsci.2016.11.016
Received 23 May 2016; Received in revised form 16 October 2016; Accepted 18 November 2016
0020-7403/ 2016 Elsevier Ltd. All rights reserved.

Please cite this article as: LI, H.N., International Journal of Mechanical Sciences (2016), http://dx.doi.org/10.1016/j.ijmecsci.2016.11.016
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Nomenclature hc chip thickness (m).

hmax maximal grain protrusion of all the grains (m).
(i )
A1,2 area used in force modeling for grain i (see Fig. 7) (m2). k1,2,3,21 empirical coecient.
A spike amplication (see Fig. 13) (N/m). kt specic grinding energy (J m3).
ap depth of cut (m). Lcube side length of cube (see Fig. 3) (m).
b grinding wheel width (m). L 2(i ) average bearing width of grain i perpendicular to grinding
bg grain width of cut (m). direction (see Fig. 8) (m).
C1,2 experimental coecient. lc wheel-workpiece contact length (m).
C grain density (m2). M grain number of the wheel (#).
d (i ) distance between grinding wheel center to the cutting N structure number of the wheel.
point of grain i (m). ng total grain number in the wheel.
dg(i ) diameter of grain i (m). Q chip cross section area (m2).
dgmean average grain diameter (m). rc ratio of chip width and thickness.
dgmax, gmin maximal and minimal grain diameters (m). S spike interval (see Fig. 15) (s).
ds wheel diameter (m). s average grain interval (see Fig. 11) (m).
Dg measured grain diameter (m). t time (s).
E elastic modulus of the workpiece (kg m2).
Vwheel grinding wheel volume (m3).
Fn, t specic normal/tangential grinding forces (N/m).
vw workpiece speed (m/s).
Fn, t normal/tangential grinding forces (N).
vs wheel speed (m/s).
Fnc, ns, tc, ts normal/tangential grinding force induced by chip forma- (i )
x, y, zlocal coordinate of the cutting point of grain i in the local
tion and grain-workpiece friction (N). coordinate system xiOz (see Fig. 5) (m).
i i
Fg resultant grinding force for a single grain (N). (i )
x, y, zglobal coordinate of the cutting point of grain i in the global
Fnr(i ), tr normal/tangential rubbing force of grain i (N).
(i )
coordinate system xOz (see Fig. 6) (m).
Fnp , tp normal/tangential plowing force of grain i (N). xg , zg coordinate of global coordinate origin (m).
(i )
Fnc , tc normal/tangential cutting force of grain i (N). x, y, zc(i ) cube center coordinate (m).
fmin minimal sampling rate in force measurements (Hz). (i )
x, y, zran 3d random vector (m).
Hs scratch hardness of the workpiece (N m2).
Poissons ratio.
HB Brinell hardness of the workpiece (N m2).
(i ) random variable (m).
h protrusion height of grain i (m).
hm(i ) (i ) wheel rotation of grain i (degree).
maximal chip thickness of grain i (m).
r(,ip), c friction coecients for the rubbing, plowing and cutting
hmmax maximal chip thickness of all the grains (m).
(i ) stages of grain i.
hplowing , cutting critical plowing and cutting chip thicknesses of grain i
(m). workpiece density (kg m3).
ht(i ) instantaneous chip thickness of grain i (m). grain volume rate of the wheel (%).

Experiments showed that the proposed model can estimate grinding
forces with acceptable accuracy where the maximal relative error of the
average normal specic forces among the six sets of validation trials
was reported to be 10.68%.
Besides, other mathematical theories were also utilised in the
modeling of grinding forces. Fuh and Wang [8] employed the back
propagation (BP) neural network to predict grinding forces because the
authors believed the multiple regressions employed in the previous
studies were not enough to describe the complicated input/output
relations in the grinding process. The results indicated that the maximal
error rate among the ten runs of validation trials was only 2.03%,
proving the good ability of the proposed model in learning and self-
organising information based on a small amount of data. Liu et al. [9]
believed the regression of experimental force data should not be the
simple multiple regression but the multivariate multiple regression and
therefore the authors proposed another empirical model, by which
grinding forces could be accurately predicted in comparison with the
experimental values. Similar multivariate analysis was also conducted
by Guo et al. [10] and the model improvement was made by
considering the dynamic forces induced by the wheel imbalance and
non-stationary wheel-workpiece interactions. Good agreement between
predicted and experimental grinding forces was observed in the
validation trials.
Although empirical models were considered to be practical and
easy-to-use for industrial applications, they intrinsically have some
crucial drawbacks due to the empirical nature, including: (i) they
required laborious eorts on the grinding operations, measurements,
data acquisition and regression calculations, (ii) a substantial amount of
experimental data was required to obtain optimum empirical coe-
cients, and (iii) most empirical coecients were determined under a
certain condition, and therefore the obtained coecients might prob-
ably not be applicable to other cases.
To overcome the above issues, semi-analytical and analytical models
were proposed. The pioneer study of these models was given by Malkin
et al. [1,11], who observed from the experiments that, both normal and
tangential grinding forces were respectively consisted of two compo-
nents: the forces induced (i) by chip formation and (ii) by grain-
workpiece friction, i.e. Fn = Fnc + Fns and Ft = Ftc + Fts . Li et al. [12]
assumed the normal cutting force of a single cutting edge in grinding was
similar to turning forces and could be expressed by a power function of
the chip cross section area Q i.e. Fn = k11Q k12 . Malkin's model [11]
therefore was transformed to be Fn = k13(vs / vw )ap + (vs / vw )k14 (ap )k15(ds )k16 .
Based on Li's model [12], Yao et al. [13] predicted grinding forces of the
Aermet 100 steel and the predicted results showed a reasonable
agreement with those experimentally obtained. The dierent derivation
of grinding forces was given by Younis et al. [14], who believed that
grinding forces were composed of three components according to the
three possible stages of grain-workpiece interactions, i.e. forces gener-
ated during the (i) rubbing, (ii) plowing and (iii) cutting stages, and the
specic normal force in this model was nally expressed as
Fn = k17(vs / vw )ap + k18[lc k19(vs / vw )ap k 20(vs / vw )] + k 21(vs / vw ). The opti-
misation of the Younis's model [14] was recently performed by
Durgumahanti et al. [15] by assuming the coecients of friction and
plowing forces were varied according to the machining parameters and
workpiece and abrasive grain materials. Although the proposed force
expression was similar to Younis's model [14], more accurate predicted
forces were obtained. Tang et al. [16] proposed another model which
focused on the calculation of chip formation forces. The authors stated
the chip formation forces can be divided into static and dynamic

2
H.N. Li et al.
components according to the level of the region shear strain and grinding
zone temperature and the experiments validated the feasibility and the
accuracy of the proposed model. Williams and Xie [17] also proposed a
unique grinding force model, in which grain-workpiece interactions in
grinding were simplied as a tribology problem, i.e. a single hard
pyramid-shaped asperity in repeated sliding contact with a softer surface.
The primary input, however, was the measured grinding wheel prole
(specically, the asperity density), which might be time-consuming to
capture if accurate density values were expected [18]. Besides, grinding
forces were modeled by Salonitis et al. [19], where the grinding-
hardening eects were taken into consideration by (i) calculating
grinding force based on the previous study [16], and (ii) introducing
coecients to these forces based on the thermal derivation of grinding
zone temperature and the dissipation of this thermal energy into work-
piece materials. The average deviations between theoretical and experi-
mental forces were found to be in the range of 48%, indicating the
accuracy of the proposed method.
Although the above models were derived from classic grinding
mechanisms and considered dierent grain-workpiece contact statues,
there are still some empirical coecients in these models which need to
be experimentally determined. Therefore the above models could be
considered as semi-analytical models. Except for semi-analytical model,
analytical models of grinding forces without any empirical coecients
were also proposed by scholars, and most of these models were
established based on the force modeling of single abrasive grit because
macroscopic grinding forces are resulted from microscopic forces at the
grain scale.
The foundational analytical work presented in Ref. [20] claimed
that the single grain grinding forces could be established based on the
workpiece Brinell hardness [21], and the resultant grinding force
induced by a single grain was expressed as
2 2
Fg = (HBDg )Dg Dg bg . This model was optimised further in
assisted grinding [44], and ultrasonic vibration assisted side grinding
Ref. [22] by taking into consideration of the elastic wheel deformation
and expressing chip thickness as a Rayleigh probability density func-
tion, and the calculated grinding forces were compared with experi-
mental data showing the agreements in both cylindrical and surface
grinding processes. Similar grinding force model based on the single-
grain analysis in Ref. [23] expressed the instantaneous single-grit
grinding force as Fg = (1/2)ktrchc2 according to the denition of the
specic grinding energy kt , and the validation experiments proved the
accuracy of the proposed model under various grinding conditions.
Park and Liang analytically established a grinding force model based on
single-grit events but considering thermal [24] and crystallographic
eects [25], i.e. workpiece material properties were set to be variable
according to the grinding zone temperature. Validation experiments
proved the accuracy of the proposed model, but the problem of this
model might be that, the grains in this model were assumed to
experience only ploughing and cutting stages, while many experimental
measurements [1,2,17,19,20] proved that the grain-workpiece rubbing
was the main reason for high grinding temperature. Besides, Cao et al.
[26] tried to create a grinding force model by the geometrical analysis
of the relative motions between the single grit and the workpiece. By
calculating the undeformed chip cross-section area, chip length, and the
number of active grains, grinding forces were obtained by aggregating
the tangential and normal forces induced by each individual grain.
Similar work was also performed in Ref. [27] but the dierence was
that the classic Johnson-cook constitutive model [28] was introduced to
calculate the cutting grinding forces, because the authors assumed the
cutting stages could be characterised by large-strain gradient, high
strain rate and high temperature. Besides, Azizi et al. [29] proposed
another grinding force model based on the calculation of the specic
grinding energy (i.e. the energy that needs to remove per unit volume of
workpiece materials). Tangential grinding forces were obtained by
3
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
geometrically establishing the volume of the theoretical removed chips.
Based on this, normal forces were also achievable according to the
coecient of friction (or grinding force ratio) experimentally deter-
mined by other research [30]. Single grain grinding forces were also
modeled by Holtermann et al. [31,32] and Denkena et al. [33,34],
where the meso-scale nite element (FE) method was employed to
calculate the instantaneous equivalent stresses on the workpiece surface
so that the grinding forces could be obtained. The experimentally
measured grinding forces in Ref. [3234] presented an encouraging
consistency with the forces predicted by the proposed model. The study
on single grain cutting events was also performed by Siebrecht et al.
[35,36], where during the force modeling the grain geometry was
divided into horizontal slices (because of the dierent interaction
mechanisms between each slice and the workpiece) and the resultant
grain force was obtained by the superposition of the forces acting on the
individual slices. The studies having profound inuences on grinding
force modeling were related with the development of the kinematic
simulation of grinding process (KSIM), although the original aim of
KSIM was, not to model forces, but to optimise grinding process
[37,38], design grinding tools [3941], and evaluate process-machine
interactions [42]. KSIM simulated grinding process as a penetration
process between the enveloping prole of the grinding wheel and the
workpiece according to grinding kinematics, therefore grinding forces
could be generated based on the microscopic process parameters
calculated by KSIM (e.g. undeformed chip thickness, chip length and
width, and chip cross-sectional area). The creep feed grinding trials of
the steel C45 workpiece [37] and the synthetic resin aluminum work-
piece [38] as well as the surface grinding trials of AISI 4140 workpiece
(602 HRC) [40] were performed, proving the feasibility of KSIM to
simulate grinding forces.
Except for the above-mentioned models, grinding forces during
dierent forms of grinding processes (e.g. vertical grinding [43], robot-
[45]), as well as for dierent kinds of workpiece materials (e.g. optical
glass BK7 [4649], the brittle materials like SiC [50] and zirconia
ceramics [45]) were also modeled according to the chip morphology
and geometry analysis based on grinding kinematics and reasonable
agreements between theoretical force values and experimental ones
were observed.
Despite the fact that many state-of-art models could be employed to
forecast grinding forces, these models were usually based on the
assumption which might contradict with the reality. For example,
according to the classic grinding theory [1], there are three possible
grain-workpiece interaction stages (i.e. rubbing, plowing, cutting)
during the grinding process due to the interaction characteristics, it
therefore might be unreasonable and inaccurate for previous studies to
model grinding forces without any determination of the grain-work-
piece interaction stages. Also, most previous studies could only be used
to predict average values of grinding forces because the most models
were built based on average grain cutting depth or average chip
thickness, and none of the grinding force details at the grain scale
could be modeled and obtained by previous models, which would
probably limit the understanding of grinding process or the develop-
ment of small-sized grinding tools [51], especially under the trend of
miniaturisation and microfabrication [3].
1.1. Novelty of the paper
To ll the above-mentioned gaps, a new prediction methodology of
grinding forces has been analytically established and experimentally
validated in this study. The novelty of this paper might include:
for the methodology:
A new strategy for the determination of grain-workpiece interaction
H.N. Li et al.
stages for each grain at each grinding moment (Section 4.4) have
been proposed, in which both random grinding wheel topography
(Section 4.2) and grinding kinematics (Section 4.3) have been taken
into consideration; This strategy is anticipated to be transferable to
study other grinding issues (e.g. grinding heat, machined surface
topography, grinding chatter) at micro (or grain) scale;
for the results:
Instead of the prediction of average [1,4,6,7,1114] and maximal
forces [1725] in previous studies, numerical grinding forces with
detailed information (e.g. rubbing, plowing and cutting force
components) in the full time-domain, to the best knowledge of
authors, have been obtained and experimentally validated for the
rst time (Section 6);
Based on the proposed methodology, the inuence of grinding
conditions (e.g. depths of cut, wheel speeds, workpiece feed rates,
and grinding wheel abrasive sizes) on grinding forces, especially on
each grinding force component (e.g. rubbing, plowing and cutting
forces) (Section 7) has been studied, which, if possible, would be
tricky for previous methods/models to obtain.
The proposed methodology therefore are believed to be able to
enhance the existing understandings of grinding forces from macro to
micro scale, and to guide industrial grinding process to a large extent
(e.g. optimisation of grinding parameters to simultaneously achieve
machining eciency and avoid high grinding temperature induced by
high portions of rubbing grains).
2. Background
2.1. Motivation
Grinding forces can be considered as a key research topic in
grinding because they connect simple geometrical/kinematic analyses
of grinding with realistic multi-physical grinding process. To be
specic:
Availability of grinding forces F is closely related with the under-
standing of grinding power P because P could be expressed as
P=vsF where wheel speeds vs could be obtained by using monitoring
methods; Therefore grinding forces have a connection with grinding
power consumption;
Because a large proportion of grinding power P is transformed into
grinding heat [1], therefore grinding forces are also related with the
grinding temperature, which further relates with the prevention of
unwanted heat damages of ground workpieces (i.e. grinding burns,
residual tensile stresses, phase transformations) and the optimum
cooling strategies (i.e. development of cutting uids);
Grinding forces could also be considered as the excitation forces for
the machine-wheel system, therefore grinding forces could be
employed to study issues with grinding chatter (i.e. machining
stability of machine-wheel system and prevention of machining
resonance);
Based on above, the accurate prediction of grinding forces could also
be a key point to foresee realistic ground workpiece surface
topography (i.e. roughness and waviness) as well as surface/subsur-
face integrity (i.e. microstructure, fatigue/fracture strength) where
grinding chatter needs to be considered;
Besides, grinding forces are closely related with grinding wheel wear
and wheel service life because they act as impacts for wheels, and
therefore are connected with grinding cost;
Based on above, accurate prediction and more in-depth under-
4
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
standings of grinding forces are needed for the development of grinding
process, which is also the motivation that drives this study.
2.2. Problem description
Unlike other machining processes conducted by tools with dened
cutting edges, grinding process could be considered as one kind of
complex machining techniques and has some unique characteristics
because of the stochastic grinding wheel topography [1]. The following
issues therefore should be considered during the modeling of grinding
forces:
As illustrated in Fig. 1(a), hundreds of discrete moving grains would
simultaneously interact with the workpiece surface and thousands
continuously in the grinding process [2]. Thus, the normal grinding
force at a certain moment should be the vector superposition of all
the normal force components induced by all the cutting grains at this
moment.
Tangential forces are usually determined by both normal forces and
friction coecients. However, the friction coecients would be
varied for dierent grains depending on dierent grain-workpiece
contact stages [15], including (i) rubbing where elastic or plastic
deformations happen on the workpiece surface, (ii) plowing where
pile-ups are formed and (iii) cutting where the workpiece material is
removed [3] (see Fig. 1b). Therefore, prior to the calculation of
grinding forces, it would be necessary to determine which contact
stage each grain is currently experiencing for a certain grinding
moment.
The bigger issue is, each grain protrusion height is also dierent due
to the varied grain sizes and distributions, which would further
result in diverse grain trajectories and dierent grain penetration
depths. Dierent grains therefore would have dierent grain
engagement, disengagement and stage transition points when they
pass the workpiece surface (see Fig. 1c). Hence, it would be essential
to decide the total number of contact stages each grain would totally
experience and the locations of the engagement, disengagement and
stage transition points for each individual grain.
3. Modeling of grinding forces
3.1. Flowchart
The uniqueness of the proposed grinding force modeling is to
calculate the grinding forces for every grain at every grinding moment
considering random wheel topography and transient grain-workpiece
contact stages and then superimpose the forces.
The modeling methodology contains four main steps (see step (i)
(iv) in Fig. 2): (i) modeling of random grinding wheel topography, (ii)
analyzing the cutting path of every grain, (iii) determining current
grain-workpiece contact stage for every grain at every grinding
moment, and (iv) calculating the rubbing, plowing and cutting forces
separately and superposing them.
One note here is that, three groups of independent input variables,
including (i) machining parameters, (ii) grinding wheel specications,
and (iii) workpiece properties, are carefully selected in the proposed
modeling method because each group would independently inuence
the desired output grinding forces:
Machining parameters including cut depths, wheel speeds, and
workpiece feed rates directly determine the motion of each grain
in grinding, and further decide the undeformed chip thickness for
each grain, therefore nally inuence not only the general features
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 1. (a) Schematics of the grinding process where hundreds of grains simultaneously interacts with the workpiece surface and thousands continuously, (b) three possible grain-
workpiece contact stages [3] including (i) rubbing where elastic/plastic deformations happen on the workpiece surface, (ii) plowing where pile-ups are formed and (iii) cutting where the
workpiece material is removed, and (c) varied grain engagement, disengagement and stage transition points for each individual grain due to diverse grain trajectories and dierent grain
penetration depths.

but also the detailed information of the output grinding forces; sharpness could be approximately set to be 2% for most grinding wheels
Grinding wheel specications including wheel diameters, abrasive (Fig. 3a).
numbers and structure numbers directly aect the grinding wheel For grain sizes, Ref. [54] concluded that the diameter of the grains
topography in terms of grain protrusion heights and grain densities dg could be expressed as:
(grain numbers within a unit area), and further decide the each
grain penetration into the workpiece, and nally inuence the dg = dgmean + , [ /2, /2], dgmean = (dg max + dg min )/2,
desired output grinding forces; = dg max dg min (1)
Workpiece properties including elastic modulus, hardness, and
Poisson ratios determine the material responses of the workpiece where dgmean is equal to 15.2/ M (M is the grain number according to
when interacting with engaging grains, and therefore inuence the ANSI B74.13-1982), and dgmax, gmin can be obtained according to Ref.
calculation of the rubbing, plowing and cutting forces for each grain [54].
(see Sections 3.5.13.5.3) and the resultant grinding forces (see For grain locations, as seen in Fig. 3(b), the modeling process is
Sections 4.5.4). similar to Ref. [55], including: (1) discretising the wheel surface into
cubes whose side length is Lcube , (2) arranging each grain at the cube
Given that the proposed model is hoped to be applicable to wide center symbolised by (xc , yc , zc ), and (3) adding a ranged 3D random
grinding conditions and dierent kinds of workpiece materials, the vector (xran , yran , zran ) to (xc , yc , zc ). The parameters xran , yran , zran are set to
above three groups of variables are selected as the input of the proposed be smaller than 0.5Lcube to avoid the grain overlap and the parameter
Lcube could be obtained by
model.
6Vwheel
Lcube = 3
ng (2)
3.2. Modeling of random grinding wheel topography
where the grain number ng could be determined by
As grains are the cutting elements in the grinding process, the wheel
ng ng +1
topography modeling therefore is conducted by the modeling of: (i) 1 1
[dg(i )]3 Vwheel & [dg(i )]3 Vwheel (ng = 1, 2, ... ),
grain shapes, (ii) grain sizes, and (iii) grain locations. 6 6
i =1 i =1
For grain shapes, according to the experimental observation re-
(3)
ported in Ref. [52], the abrasive grains could be considered to be
shaped like the cross section of a frustum of a cone with the included and the grain volume rate could be obtained by = 1.5(37 N )%.
angle of 60 deg (Fig. 3a), so that the wheel dressing process prior to the Fig. 3(c) presents an example of the established grinding wheel
grinding process and the wheel wear during the grinding process could topography.
be considered. The wear ats could be sized by the grain sharpness Unlike any previous grinding wheel models, the additional con-
(dened as the percent ratio of the at diameter to the grain diameter dg sideration of the proposed methodology is that, the modeled wheel
[3]), and the experimental measurements in Ref. [53] showed the topography has been evaluated before it has been used. This is because,

5
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 2. Flowchart of the proposed detailed modeling of grinding forces including four main steps: (i) modeling of random grinding wheel topography, (ii) analyzing the cutting path of
every grain, (iii) determining current grain-workpiece contact stage for every grain at every grinding moment, and (iv) calculating the rubbing, plowing and cutting forces separately and
superposing them.

although the above modeling procedures could output random numer-
ical wheel topography, the dierence between the numerical and
realistic wheel topographies would probably still exist; and if the
dierence is big, the accuracy of the proposed model would be largely
limited. To address this, the wheel topography utilised in the following
calculation is obtained by: (i) iteratively running the above modeling
procedures and then (ii) measuring and comparing the roughness Ra
and waviness WA values (according to ISO 4287:1998) of the three
cross-section proles randomly extracted from the numerical and
realistic wheel topographies (see Fig. 4), until the dierence of Ra

Fig. 3. Modeling procedures of the random grinding wheel topography: (a) grains shaped like the cross section of a frustum of a cone with the included angle of 60 deg (out of scale), (b)
the determination of grain locations including three steps: (1) discretising the wheel surface into cubes whose side length is Lcube , (2) arranging each grain at the cube center symbolised
by (xc , yc , zc ) , and (3) adding a ranged 3D random vector (xran , yran , zran ) to (xc , yc , zc ) , and (c) the established grinding wheel topography.

6
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 4. (a) Illustration of the measurement coordinate when comparing the roughness Ra and waviness WA values between the numerical and realistic wheel topographies, and (b) one
typical comparison result of cross-section proles extracted from the numerical and realistic wheel topography.

and WA values between the numerical and realistic wheel topographies x (i ) = d (i ) sin (i ) d (i ) vw (i )
local
is lower than 10%. vs
,
z (i ) = d (i )[1 cos (i )]
local (4)
3.3. Kinematics analysis of cutting grains considering grain protrusions where the sign + refers to the up grinding process while
corresponds to the down grinding.
Although the trajectory of a single grain (see the curve OiAi Bi in Given that the grain interval is very small relative to the wheel
Fig. 5) has been derived by famous grinding researchers like Malkin diameter [2,3], it would therefore be reasonable to assume the wheel
et al. [1] and Rowe et al. [2], the grain protrusion (see h(i ) in Fig. 5) has rotation (i ) from the grain i engagement to the disengagement as a very
been rarely considered. In realistic grinding, however, each grain passes small angle, i.e.
the workpiece surface along dierent grain trajectories because of the 1 (i ) 2
varied grain protrusions induced by the random wheel topography. sin (i ) (i ); 1 cos (i ) [ ] .
2 (5)
Given that the proposed methodology is based on the analysis of
dierent kinds of grain-workpiece interactions and each grain-work- Eq. (4) therefore could be transformed into:
piece contact information should be available, the new kinematics (i ) 2
[xlocal ]
(i )
analysis of cutting grains considering grain protrusions has been zlocal = .
(i )
2d (1 vw / vs )2 (6)
performed in this section, which contains two steps: (i) analysis of
single cutting grain trajectory in the local coordinate, and (ii) analysis Taking grain protrusion heights into consideration, the distance
of multiple cutting grain trajectory in the global coordinate. from the cutting point to the wheel center d (i ) could be expressed as:

d (i ) = h(i ) + 0.5ds. (7)

3.3.1. Kinematics analysis of single cutting grain in the local coordinate
To give a uniform cutting path expression for grains with dierent Therefore, the trajectory of a single grain in the local coordinate
protrusions, the local coordinate system xiOz is established for each system xiOz
i i
(see the curve OiAi Bi in Fig. 5) could be nally expressed as
i i
grain where the origin Oi is xed at the lowest point as seen in Fig. 5. (i ) [x (i ) ]2
In the local coordinate system xiOz , the parametric curve of the zlocal = (i) local
i i [2h + ds](1 vw / vs )2 .
cutting path of the grain i (see the curve OiAi Bi in Fig. 5) in terms of the y (i ) = y (i ) + y (i )
local c ran (8)
parameter wheel rotation (i ) then could be expressed as:

3.3.2. Kinematics analysis of multiple cutting grains in the global coordinate

To express every grain path in a unied coordinate, the global
coordinate system (see the coordinate system xOz in Fig. 6) is created.
The x coordinate of the origin point O (denoted as xg in Fig. 6) is set to
be the x coordinate of the cutting point where the grains rst contact
with the workpiece, while the z coordinate (denoted as zg in Fig. 6) is
set to be the same as the maximal grain protrusion hmax to allow
machined surface topography to be positive in the z direction.
According to Ref. [3], hmax could be approximately considered to be
40% of the average grain diameter dgmean .
Therefore the cutting paths of multiple cutting grains (i.e. the curve
OiAOi +1BOi +2COi +3D in Fig. 6) in the global coordinate xOz could be
expressed by a function set P , i.e.
(i )
[xglobal xg]2
zglobal
(i )
= (i ) + [h(i ) 0.4dgmean] (i = 1, 2, 3, ... ng )
P= [2h + ds](1 vw / vs )2 .
y (i ) = y (i ) + y (i )
global c ran

(9)
Fig. 5. Schematics of the cutting path of a single cutting grain (i.e. the curve OiAi Bi ) in the
local coordinate system xiOz
i i (out of scale).

7
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 6. Schematics of the cutting path of multiple cutting grains (i.e. the curve OiAOi +1BOi +2COi +3D ) in the global coordinate system xOz (out of scale).

3.4. Determination of the grain-workpiece contact stages determined according Ref. [56]):

It is well-accepted that there are three possible grain-workpiece When hm(i ) > hcutting
(i )
(see Fig. 7b), the grain would totally experience
contact statues in grinding [1,2] (i.e. rubbing, plowing and cutting), but all the three possible contact stages including rubbing, plowing and
few researchers have established a scheme to quantitatively determine cutting.
which status a certain grain is currently experiencing at a certain When hplowing
(i )
< hm(i ) < hcutting
(i )
(see Fig. 7c), the grain would totally
grinding moment. In this section, a new strategy for the determination experience two possible contact stages including rubbing and
of the grain-workpiece contact stages has been suggested as follows. plowing.
Dierent grain protrusion heights (see h(i ) in Fig. 7a) results in When hm(i ) < hplowing
(i )
(see Fig. 7d), the grain would only experience
varied maximal undeformed chip thicknesses (see hm(i ) in Fig. 7a), and the rubbing stage.
the relationship between these two parameters could be expressed as:
With regard to the contact stage that the grain i is currently experien-
hm(i ) = hm max [dg max h(i )], (10) cing at the grinding moment t , it could be decided by the relationship
between the instantaneous chip thickness for the grain i (denoted as
where, according to Ref. [1], hmmax can be expressed as: ht(i ) ) and hplowing
(i ) (i )
and hcutting :
6 vw ap 1/2 1/2
hm max = [ ( )( ) ] ,
Crc vs ds (11) When ht(i ) < hplowing
(i )
, the grain i is currently experiencing the rubbing
stage;
and the grain protrusion height h(i ) could be obtained by When hplowing
(i )
< ht(i ) < hcutting
(i )
, the grain i is currently experiencing the
plowing stage;
h(i ) = zc(i ) + zran
(i )
+ 0.5dg(i ). (12) When ht(i ) > hcutting
(i )
, the grain i is currently experiencing the cutting
According to Ref. [54], the contact stages that the grain i would stage,
totally experience could be determined by the relationship between the
maximal chip thickness hm(i ) and the critical chip thicknesses for plowing where the instantaneous chip thickness for the grain i (denoted as
(i )
and cutting stages (separately denoted as hplowing (i )
and hcutting and could be ht(i ) ) could be obtained by the following equation (see Fig. 7a):

(i )
Fig. 7. (a) Front view of the modeled grinding wheel topography, and the possible grain-workpiece contact stages when (b) h m(i ) > hcutting
(i ) (i )
, (c) hplowing < h m(i ) < hcutting
(i )
and (d) h m(i )<h plowing .

8
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

ht(i ) = zglobal
(i )
[dg max hm max + hm(i )]. (i ) (i ) 1/2
(13) 2 L 2 2 L (i ) L
p(i ) = [ (i ) ] sin1[ 2(i ) ] [ 2(i ) ]2 1 ,
dg dg
g
d

(18)

3.5. Modeling of grinding forces where the average bearing width L 2(i ) of the grains perpendicular to the
grinding direction (see Fig. 8b) could be expressed as:
After the availability of each grain-workpiece contact stage, the
3 (i )
grinding forces could be calculated for each grain from the grain L 2(i ) = [L1(i ) + 4%dg(i )]/2 = ht + 2%dg(i ).
3 (19)
engagement to the disengagement.
Therefore the normal plowing force could be obtained by:
3.5.1. Rubbing forces (i )
Fnp = Ftp(i) / p(i)
The rubbing (or friction) during the grinding process happened at
3
the interfaces between the wear ats of the wheel and the machined Hs[ 3
ht(i) + 2%dg(i)]ht(i)
= .
surface [1] and could be characterised by the elastic and plastic 3 (i ) 3 (i ) 3 (i ) 1/2
2
h + 2 % dg(i ) ht + 2 % dg(i ) h + 2 % dg(i )
[ 3 t 3 t
deformations of the workpiece (Fig. 8a). This grain-workpiece interac- ( i )
]2 sin1[ 3 ] [ ]2 1

dg dg(i ) dg(i )
tion would be similar to the contact between a round-tip cone and an
elastoplastic half innite plane, the normal rubbing force therefore
(20)
could be expressed as Eq. (14) according to Ref. [57].

Fnr(i ) = 23/2HB[dg(i )]3/2 [ht(i )]1/2 . (14)

Given that the wear at area during the rubbing process would 3.5.3. Cutting forces
increase and the elastic, elastoplastic, or plastic workpiece deformation The cutting stage is considered to be related with the formation of
might happen, therefore the friction coecient r (or the ratio of the chips and the removal of the workpiece material.
normal rubbing force to the tangential one) is assumed to be variable Based on Ref. [60], the normal forces induced by the moving
and could be expressed as [16]: indenter during the cutting stage could be represented as:
A1(i )
r(i ) = C1 + C2. (i ) 28
Fnr(i ) (15) Fnc = [ht(i )]2 HB.
3 (21)
where the wear at area A1(i ) (see Fig. 8a) could be obtained by According to Ref. [12], the friction coecient during the cutting
2
A1(i ) =[1%dg(i )] and the coecients C1 and C2 could be obtained stage could be expressed as:
according to Ref. [16]. Hence, the tangential rubbing force could be

obtained by: c(i ) =
4 tan 60 (22)

[1%dg(i )]2

Ftr(i ) = r(i ) Fnr(i ) = {23/2HB[dg(i )]3/2 [ht(i )]1/2 }C + C . Therefore the tangential cutting force could be:
1 ( i ) 2
Fnr (16)
28
Ftc(i ) = c(i ) Fnc
(i )
= [ht(i )]2 HB .
3 4 tan 60 (23)

3.5.2. Plowing forces

As seen in Fig. 8(b), the plowing stage could be featured by the
small amount of accumulated materials in front of the rake face of the
grains and the ridge at the sides of the grinding path.
According to Ref. [58], the resistance of the workpiece materials
towards plowing by a frustum of a cone could be dened as the scratch
hardness (denoted as Hs ), and the tangential plowing force therefore
could be expressed as:
3 (i )
Ftp(i ) = HsA2 = Hs[ ht + 2%dg(i )]ht(i ).
3 (17)
Ref. [59] gave the expression of the plowing friction coecient as
3.5.4. Superposition of single-grain grinding forces
The proposed force modeling is based on the grain-workpiece
interactions, therefore the resultant normal and tangential grinding
forces could be separately obtained by making the vector superposition
of the grinding forces of each grain.
As seen in Fig. 9, the normal and tangential grinding forces in the
grinding process are usually dened in the vertical and horizontal
directions, therefore the following expressions of the resultant normal
and tangential grinding forces could be obtained by:

Fig. 8. Schematics of the grinding force modeling for the grain-workpiece contact stages including (a) rubbing, (b) plowing and (c) cutting stages.

9
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 9. Schematics of the superposition of single-grain grinding forces to obtain resultant normal and tangential grinding forces.

Table 1 4. Experimental methodology for model validation

Chemical composition of the AISI 1055 workpiece.
To validate the proposed modeling, the grinding experiments are
Element Fe Mn C S P
performed in this study.
wt(%) 98.41% 0.6% 0.5% < 0.05% < 0.04%
4.1. Workpiece and grinding tools

The AISI 1055 workpiece with the dimension of 100 mm (length)

Table 2
50 mm (width)60 mm (height) is utilised in the experiments, and the
Material properties of the AISI 1055 workpiece.
chemical composition and the material properties are outlined in
Material (kg m3) E (GPa) HB (MPa) Hs (MPa) Table 1 and Table 2 respectively. The polishing operation (using the
#400 diamond abrasives, Shanghai Yuli Industry and Trade Co., Ltd.) is
AISI 1055 7840 190 0.27 197 92 performed on the workpiece surface before the experiments to avoid
any inuence of previous scratches until the workpiece surface rough-
ness reaches the Ra value of 0.8 m.
The utilised aluminum oxide grinding wheel is dressed by a dressing
wheel before the experiments to restore the wheel cutting ability and
z (i ) (i )
zwheel
Fn = ng Fnr(i ), np, nc cos [arctan global (i ) the detailed information of these two wheels and the dressing para-
(i ) ] + Ftr , tp, tc
i =1
x (i )
global x wheel meters are separately presented in Table 3 and Table 4.

(i )
zglobal (i )
zwheel
sin [arctan (i) (i ) ]
4.2. Experiment setup
xglobal x wheel
.
(i )
zglobal (i )
zwheel
The experiment setup is illustrated in Fig. 10. The workpiece is
Fn = i =1 Ftr , tp, tc sin [arctan x (i) x (i) ] Fnr , np, nc
ng (i ) (i )

global wheel mounted on the jig which is then mounted on the dynamometer. The
(i )
zglobal (i )
zwheel dynamometer is xed on the electro-magnetic worktable of the surface
cos [arctan (i)
(i ) ] grinder (Shenyang Machine Tool (Group) Co., Ltd., M7150). The used
xglobal x wheel (24) grinding parameters are given in Table 5. No cutting uids are used.
Thus, the specic normal and tangential grinding forces could be
expressed as: 4.3. Grinding force measurement
Fn = Fn / b
. The normal and tangential grinding forces are measured by using
Ft = Ft / b (25) the system including the multi-component dynamometer (9257, Kistler
Co., Ltd.), the charge amplier (UT4208, uTekL Co., Ltd.), the A/D data
acquisition card (USB-6001, National Instruments Co., Ltd.), and the PC

Table 3
Detailed information of the utilised diamond grinding wheel and the dresser wheel.

Item Shape Abrasive Bond Size (mm) M (#) N Diamond concentration

Grinding wheel disk-type aluminum vitrified 300 40 30 150 5

Dressing wheel disk-type diamond metal 30 12 5 50 4 100%

10
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Table 4 the grain interval (denoted as s in Fig. 11a) is uniform, the minimal
The utilised dressing parameters. sampling rate fmin should satisfy fmin 2vs / s according to the sampling
Dresser Dressing Grinding Dressing ratio Dressing Dressing
theory [61] where s 1/ C . According to the 3D grinding wheel
wheel wheel depth mode topography (see Fig. 11b) measured by the prolometer (Talysurf PGI
speed (m/ speed (m/s) (m) Optics, Taylor Hobson Co., Ltd.), it could nd C 12 grains/mm2 for
s) the grinding wheel used in the experiments. The sampling rate there-
Dressing 9 15 0.6 15 (15 Up
fore in the experiments is set to be 200 kHz.
wheel times) dressing To evaluate and minimise the measure error, a set of standard
weights are gradually and carefully loaded on the dynamometer
(increasing 1 kg each time) and the captured signal is presented in
Fig. 12. It could nd that, the signals are scattered within the range of
0.5 N around the theoretical values, indicating the accuracy of the
measurement system.
The raw force signals might probably contain a substantial amount
of noise, although hardware lters have been set in the dynamometer
and the acquisition card. Therefore digital lters based on the cross-
section function are designed and utilised in this study strictly accord-
ing to the instruction given in Refs. [9] and [62]. The adjustment of the
lter parameters is iteratively performed until the force signal uctua-
tion during the non-grinding period is close to 1 N (see Fig. 13), which
could correspond to the signal uctuation shown in Fig. 12. The
denoised force signals during the steady grinding period then will be
employed in the following validation as the experimental grinding
forces.

5. Model verication

Fig. 10. Experiment setup to validate the proposed model of grinding forces. 5.1. Comparisons with previous models

Before the comparison with the experiments, the initial evaluation

Table 5 of the proposed methodology is rstly performed by comparing the
The used grinding parameters. forces obtained by the proposed methodology and by other previous
models. Specic average grinding forces (i.e. average forces per wheel
Group No.
width) are employed as the comparative item, because previous models
1 2 3 could only predict 1D force values (i.e. forces without any information
in the time domain).
Depth of cut ap (m) 30 80 150 To provide a comprehensive comparison, dierent kinds of previous
Wheel speed vs (m/s) 26.9 force models including empirical, semi-analytical, and analytical mod-
Workpiece feed vw (m/min) 10 els (catogerised based on the literature review in the Section 1) from
Grinding model Down grinding
1970s to 2010s are utilised. In order to make the results comparable,
the same machining parameters, workpiece material properties and
(E93z, ThinkCeter Co., Ltd). grinding wheel specications (see Table 6) are employed as the inputs
One note here is the selection of the sampling rate during the of the proposed model. Forces predicted by the proposed method is
grinding force measurements. Because the proposed modeling method calculated according to the procedures given in Section 4. To minimise
aims to depict detailed information of the grinding forces, therefore the errors induced by random wheel topography, three runs of each
sampling rate should be as high as possible to recognise the engagement calculation are performed and standard error bars are illustrated to
and the disengagement of small numbers of individual grains. Assuming show the stability of the results. One note here is that, specic material
removal rates (i.e. removed volume per unit of time per unit of wheel

Fig. 11. (a) Illustration of the selection of the sampling rate during the grinding force measurement and (b) the measured 3D grinding wheel topography of the used grinding wheel
(approximately having 48 peaks (P1~P48) within the 2 2 mm area).

11
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 12. The captured raw signals when standard weights are gradually and carefully loaded on the dynamometer (increasing 1 kg each time).

Fig. 13. The denoised signals where the force signal uctuation during the non-grinding period is close to 1 N (corresponding to the signal uctuation shown in Fig. 12) and the forces
during the steady grinding period then will be used in the model validation.

Table 6
Previous grinding force models and corresponding parameters used in the comparison.

Ref. No. Model type Machining parameters Workpiece properties Wheel specification

[4] 1 Empirical vs=10, 15, 20, 25, 30, 35, 40 m/s E = 190 GPa Grit size: 150#
vw=0.04, 0.08, 0.1, 0.15, 0.2 m/s = 0.27 Grit type: aluminum
ap = 5, 15, 25, 35, 50 m HB = 197 MPa
[14] 2 Semi-analytical vs=30 m/s E = 220 GPa Grit size: 60#
vw = 0. 1, 0. 2, 0. 3 m/s = 0.3 Grit type: aluminum
ap=2, 5, 10, 15 m HB = 230 MPa
[63] 3 Semi-analytical vs=36 m/s E = 170 GPa Grit size: 80#
vw = 0. 3 m/s = 0.29 Grit type: aluminum
ap = 10, 20, 30, 40 m HB = 152 MPa
[16] 4 Analytical vs=12 ,24, 30 m/s E = 130 GPa Grit size: 120#
vw = 0. 2 m/s = 0.27 Grit type: aluminum
ap=10, 20, 30 m HB = 92 MPa
[22] 5 Analytical vs=27 m/s E = 49.6 GPa Grit size: 80#
vw = 0. 5 m/s = 0.29 Grit type: aluminum
ap = 10, 12, 14, 16 m HB = 38.2 MPa

12
H.N. Li et al.
Fig. 14. Comparison of specic (a) normal and (b) tangential grinding forces calculated by the previous models given in Table 6 and by the proposed methodology under dierent specic
material removal rates ( refers to relative error between the forces predicted by the proposed method Fpro and the forces reported in the previous studies Fpre , i.e.
= abs(Fpro Fpre )/Fpre ).
width [13]), rather than a certain machining parameter, are employed
as the horizontal axis in the results in order to demonstrate all the
comparisons in one diagram.
The comparison results for normal and tangential forces are
separately presented in Fig. 14(a) and (b). It could nd that, for
dierent values of specic material removal rates, the general
tendency of forces predicted by the proposed method is consistent
with those calculated by previous models, i.e. forces are increased
with increasing removal rates. Moreover, among all the 22 compar-
isons, most predicted forces are close to the ones reported in previous
publications, and only the relative errors of 7 sets for normal forces
and 6 sets for tangential forces are more than 10%, proving the
feasibility and accuracy of the proposed methodology. Besides, the
proposed method also shows a high applicability when wide ranges of
specic normal and tangential forces (058 N/mm for normal forces
13
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
and 028 N/mm for tangential forces) are compared and a large range
of workpiece properties, machining parameters and wheel specica-
tions are used in the trials.
From this comparison, the advance of the proposed method could
also be observed: all the previous models could only predict the 1D
forces (forces without any information in the time domain), while the
proposed model could obtain grinding forces with rich information in
the full time domain (or full force history). Moreover, because the
proposed method is based on the analysis of single-grit forces, force
details at the grain scale (e.g. values of rubbing, plowing, and cutting
force components, numbers of grains that are experiencing rubbing,
plowing, and cutting stages at a certain moment) can also be accessible,
which could enable the studies of grinding force from macro scale to
micro scale. More details about the detailed comparison of predicted
and experimental grinding forces are provided in Section 6.2.2.
H.N. Li et al.
Fig. 15. The validation of the proposed model including (i) general comparison and (ii) detailed comparison of grinding force signals, where the general comparison is quantied by the
maximal, minimal and average values of the upper envelope, lower envelope and the moving average curve, and the detailed comparison is quantied by the spike amplication and the
spike interval.
5.2. Comparisons with experimental results
To further validate the proposed method, experiments and corre-
sponding calculations with the same workpiece properties, grinding
wheel specication and machining parameters are also performed and
compared. The validation is conducted from the following two aspects:
(i) general comparison and (ii) detailed comparison of grinding forces.
As seen in Fig. 15, the general comparison is evaluated by
the maximal, minimal and average values of the Upper Envelope
(UE) of the grinding force signals;
the maximal, minimal and average values of the Lower Envelope
(LE) of the grinding force signals;
the maximal, minimal and average values of the Moving Average
Curve (MAC) of the grinding force signals,
where UE, LE and MAC could be are obtained by correspond-
ingly using the algorithms proposed in Refs. [6466].
The unique ambition of the proposed method is to depict
grinding forces including detailed information at the grain scale.
Therefore the detailed comparison is focused on the characteristics
of each single spike signal, which could be quantied by the
parameters including:
the maximal, minimal and average values of the amplication of the
spike signal (see A1~A5 in Fig. 15);
the maximal, minimal and average values of the spike interval (see
S1~S4 in Fig. 15)
where the spike signals could be dened by the peak point pi that
satises pn > pn1 and pn > pn+1 (see p1~p5 in Fig. 15).
5.2.1. General comparisons between experimental and numerical grinding
forces
The general comparison result is presented in Fig. 16. It can be seen
that, the upper envelope, the lower envelope and the moving average
curves of experimental and numerical grinding forces are close to each
other, indicating the feasibility and accuracy of the proposed modeling
method to a large extent.
The general comparison is quantied as presented in Table 7. It
could nd that, among 54 sets of comparisons, the relative errors of
only ten sets (marked red in Table 7) are beyond 10%, indicating that
the proposed model could, to a large extent, describe the general
14
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
features of realistic grinding forces under dierent machining condi-
tions. The average percentage error of all the comparisons is 6.73%,
which is close to those presented in other previous studies (e.g. Refs.
[43,50]), and even smaller than those given in Refs. [15] and [45],
indicating the possibility to accurately model grinding forces based on
dierent stages of grain-workpiece interactions.
5.2.2. Detailed comparisons between experimental and numerical grinding
forces
As seen in Section 4, the unique ambition of the proposed
methodology is to model grinding forces with detailed information
(e.g. rubbing, plowing and cutting force components) in the full time-
domain. Therefore in this section, the detailed comparisons between
experimental and numerical grinding forces are presented in terms of
the amplications and intervals of the spikes in grinding force signals.
It could nd from Fig. 17 that, the characteristics of the spike signals
predicted by the proposed methodology are comparable to those
captured in experiments, although the values of the spike amplications
and intervals at a certain moment might be dierent due to the intrinsic
dierence between the random realistic and numerical grinding wheel
topography. Besides, the quantitative comparison of the spike force
signals (see Table 8) also indicates that, the values of the parameters A
and S are in a good agreement with other previous studies or theoretical
values:
For the parameter A, Durgumahanti et al. [15] experimentally found
that the specic normal and tangential grinding forces were
approximately within the ranges of 1822 N/mm and 1214 N/
mm respectively when vw was 6 m/min, vw was 20 m/s and ap was
30 m. These ranges could match well with the numerical grinding
forces given in Table 8;
Moreover, according to the measured grain density of the
utilised wheel topography (see Fig. 11b), the normal and tangential
grinding forces for each single grain could also be obtained, which
are within the ranges of 5.16.3 N/grain and 3.44 N/grain. These
ranges are also in a good agreement with those measured in the
single-grit trials reported in Refs. [67,68].
The parameter S corresponds to the time duration from the rst
grain engagement (or disengagement) to the next grain engagement
(or disengagement). The grain interval could be approximately
considered to be 0.289 mm (see Fig. 11b, equal to (1/12) mm)
H.N. Li et al.
Fig. 16. General comparison of experimental and numerical grinding forces with vw of 10 m/min, vs of 26.9 m/s and ap of 30, 80, and 150 m: (a) normal and (b) tangential forces when
ap is 30 m, (c) normal and (d) tangential forces when ap is 80 m, and (e) normal and (f) tangential forces when ap is 150 m (NS: numerical signal, ES: experimental signal, UE: upper
envelope, LE: lower envelope, MA: moving average).
when assuming the interval is uniform, while the grain speed could
be assumed to be the same as the wheel-workpiece relative speed of
27.3 m/s. The theoretical value of the parameter S therefore should
be around 10 s, which is also in the good consistency with the
numerical results given in Table 8.
It could also notice from Table 8 that, among all the 36 sets of the
comparisons, 25 sets showed the relative error of more than 10% and
the biggest error is even more than 200%. This big dierence might
probably because the parameters A and S are both highly-localised
micro-scale parameters and are largely inuenced by the realistic and
numerical wheel topography. These two topography, however, are hard
to be the same, although the numerical topography is modeled based on
the stochastic nature of the realistic one. In fact, it might be too
demanding to require an analytical methodology to predict realistic
15
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
grinding forces with the exactly same spike amplication and interval
at the exactly same grinding moment. Even so, the proposed model is
believed to be meaningful to provide a new way to model grinding
forces with detailed information at the grain scale with the considera-
tion of dierent contact stages of the grain-workpiece micro interac-
tions.
6. Eects of grinding conditions on rubbing, plowing, and cutting
forces Model applications
After verifying the feasibility and accuracy of the proposed model,
one example of the model applications has been given in this section to
show the advance of the proposed methodology, which is investigating
the eects of grinding conditions (including depths of cut, wheel
speeds, workpiece feed speeds and grinding wheel abrasive sizes) on
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Table 7
General comparison of the experimental and numerical grinding forces with vw of 10 m/min, vs of 26.9 m/s and ap of 30, 80, and 150 m (NS: numerical signal, ES: experimental signal,
Err.: error rate, max.: maximal value, min.: minimal value, aver.: average value).

ap Upper envelope curve (N/mm) Lower envelope curve (N/mm) Moving average curve (N/mm)

max. min. aver. max. min. aver. max. min. aver.

30 Fn NS 23.71 21.93 22.82 10.65 8.59 9.62 16.48 14.67 15.575

ES 24.75 23.29 24.02 10.76 8.96 9.86 17.88 16.18 17.03
Err. (%) 4.39 6.20 5.26 1.03 4.31 2.49 8.50 10.29 9.34

Ft NS 15.17 14.41 14.79 5.28 4.53 4.91 10.69 10.19 10.44

ES 16.32 15.48 15.89 6.06 4.72 5.39 10.54 9.65 10.10
Err. (%) 7.45 7.50 7.47 14.77 4.19 9.89 1.42 5.60 3.42

80 Fn NS 36.52 35.71 36.14 17.49 16.28 16.89 28.15 27.48 27.82

ES 36.13 34.91 35.52 16.91 16.04 16.48 27.11 26.01 26.56
Err. (%) 1.11 2.29 1.69 3.43 1.50 2.49 3.84 5.65 4.73

Ft NS 25.16 23.69 24.43 10.32 9.59 9.96 17.93 16.16 17.05

ES 23.52 20.98 22.25 8.79 7.51 8.15 16.38 14.64 15.51
Err. (%) 6.97 12.92 9.78 17.41 27.70 22.15 9.46 10.38 9.90

150 Fn NS 71.57 67.98 69.78 32.41 29.7 31.06 51.48 50.1 50.79
ES 75.02 72.41 73.72 30.78 29.71 30.25 53.54 51.08 52.31
Err. (%) 4.82 6.52 5.65 5.30 0.03 2.68 4.00 1.96 2.99

Ft NS 45.76 43.54 44.65 25.19 23.42 24.31 35.26 33.43 34.35

ES 48.89 47.21 48.05 22.81 20.98 21.90 36.16 34.01 35.09
Err. (%) 6.84 8.43 7.61 10.43 11.63 11.01 2.55 1.73 2.15

each component of grinding forces (rubbing, plowing, and cutting
forces). Some new understandings on grinding forces, which would be
dicult for other previous models to obtain, have been gained in this
section.
It could generally comment from the results in Fig. 18 that, large
depths of cut and low wheel speeds would lead to big normal and
tangential grinding forces while the eects of the changes of workpiece
feed rates seem to be limited. This might because grinding forces are
closely related with chip thickness hm [1], which according to Refs. [2,3]
could be expressed as
4 vw ap 1/2 1/2
hm = [ ( )( ) ] ,
Cr vs de
therefore large cut depths and low wheel speeds easily increase chip
thickness and thus increase the forces. Workpiece feed rates, on the
other hand, are usually 100-1000 times smaller [13] relative to wheel
speeds, therefore the impacts of workpiece rates on forces are not
obvious.
Meanwhile, it could generally nd from Fig. 18 that, grinding
wheels with small abrasive numbers (corresponding to large abrasive
grain diameters) easily result in low grinding forces when machining
parameters are kept constant.
The detailed analysis of the eects of grinding conditions from
Fig. 18 are as follows:
Rubbing, plowing and cutting grain-workpiece interactions simulta-
neously exist in grinding, although wide grinding conditions (ap :
10~50 m; vw : 4~12 m/min; vs : 13.5~31.3 m/s; abrasive sizes:
#36~#180) have been employed; This can be treated as the
analytical evidences of wheel-workpiece interactions in realistic
grinding process: nearly hundreds of grains interacting with the
workpiece surface at a high speed. A part of these grains are
rubbing, during which the workpiece is sheared by grains at the
grain-workpiece interface with no permanent plastic deformation
presents; Another part of these grains are plowing, during which the
ridges to either side of the grain motions are generated without any
material removal; Other grains are cutting, during which the chips
are formed and the materials are removed; the above three interac-
tions simultaneously happen and none of them is missing during the
whole grinding process.
The contribution of each component (i.e. the ratio of rubbing,
plowing and cutting forces to resultant grinding forces) is varied
within certain ranges (Fig. 18af): the rubbing and cutting contribu-
tions vary separately within 17.9~68.9% and 19.9~68.7% while
the plowing contribution is relatively stable, within a narrow range
of 9.1~15.6%. This implies that, although there are three well-
accepted grain-workpiece interactions in grinding, the number of
plowing grains accounts for only a small portion while the dominant
grain-workpiece interactions are rubbing and cutting. This means,
the proposed method would be useful to lower grinding temperature
(26)
by optimising grinding conditions so as to achieve the minimum
number of grains that take part in rubbing (because heat generated
at the shear plane during the rubbing regime is regarded as the
major cause of high grinding temperature [13]).
For the eects of cut depths, it could nd that, shallow-cut grinding
process is dominated by grain-workpiece friction because the
contribution rate of rubbing forces are beyond 50% (see Fig. 18a
and b). This could be explained by the situation presented in
(i )
Fig. 7(c) where hm(i )<h plowing and therefore among all the engaging
grains most of them would only experience the rubbing stage from
the grain engagement to disengagement.
The above observation could also explain the size eect in
grinding (i.e. the specic energy becomes higher when cut depths
are decreased [69,70]): with small cut depths, more grains only
experience rubbing stages during which more input energy would be
transformed into friction-induced heat and therefore much more
energy is required to remove the same volume of materials than
those when most grains are in plowing or cutting stages.
With increased cut depths, the contribution of rubbing decreases
and, as expected, cutting contribution increases to a large extent;
however, both the decrease and increase speeds become slower.
Therefore it might predict that when ap becomes bigger than 50 m
the contribution of rubbing, plowing and cutting will probably keep
relatively constant. This might probably because the grain protru-
sions are limited by grain sizes, thus the dierence between the
highest and lowest cutting edges on the wheel surface are within

16
H.N. Li et al.
Fig. 17. Detailed comparison of experimental and numerical grinding forces with vw of 10 m/min, vs of 26.9 m/s and ap of 30, 80, and 150 m: (a) normal and (b) tangential forces when
ap is 30 m, (c) normal and (d) tangential forces when ap is 80 m, and (e) normal and (f) tangential forces when ap is 150 m.
a certain range. Considering increased cut depths easily lead to poor
surface nish because of the increased chip thicknesses [1], the cut
depth of 50 m would probably be a rational grinding depth for AISI
1055 steel workpiece because both relatively high cutting eciency
and good-quality surfaces could be obtained.
The eects of wheel speeds are also observable (see Fig. 18c and d):
fast wheel speeds lead to less rubbing forces and more cutting forces.
However, among all the results presented in Fig. 18, the cutting
contribution is no more than 69%, indicating a great potential of the
machining eciency of the existing grinding techniques. Random
wheel topography might probably be the key obstacle for grinding
development because of the dierence of grain protrusions: even
though the grains with large protrusion heights could reach ecient
cutting stages, the grains with small protrusions might could only
experience rubbing stage. Hence, the grinding tools with more
uniform grain protrusions are expected to be promising.
Besides, Fig. 18(c) and (d) also provide explanations of the
17
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
superiority of high and ultra-high speed grinding technique (e.g.
reduced wheel wear, low grinding temperature and improved sur-
face quality): with increasing wheel speeds, more grains take part in
cutting stages rather than rubbing stages, therefore less friction and
friction-induced heat, together with more workpiece materials that
are removed as chips rather than left on the workpiece surface as
debris or ridges, are achieved;
Unlike cut depths and wheel speeds, the eects of workpiece feed
rates on each component force seems to be ambiguous (see Fig. 18e
and f). The contribution of each component is relatively stable,
although normal and tangential grinding forces are increased to a
small extent when workpiece feed rates are increased. Therefore
high workpiece feed speeds should be employed to achieve high
material removal rates.
The eects of grinding wheel abrasive numbers, which, to the best
knowledge of the authors, has been rarely researched before, are
presented in Fig. 18(g) and (h). It could observe that, for the
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Table 8
Detailed comparison of the experimental and numerical grinding forces with vw of 10 m/min, vs of 26.9 m/s and ap of 30, 80 and 150 m (NS: numerical signal, ES: experimental signal,
Err: error rate, max.: maximal value, min.: minimal value, aver.: average value).

ap Spike amplification A (N/mm) Spike interval S (s)

max. min. aver. max. min. aver.

30 Fn NS 12.51 1.75 7.13 9.11 6.55 7.83

ES 12.81 0.56 6.69 7.18 5.68 6.43
Err. (%) 2.40 212.50 6.66 26.88 15.32 21.77

Ft NS 10.32 1.87 6.10 8.74 6.21 7.48

ES 10.01 0.76 5.39 10.77 6.97 8.87
Err. (%) 3.10 146.05 13.18 23.23 12.24 18.66

80 Fn NS 22.32 2.21 12.27 10.26 5.38 7.82

ES 21.07 1.02 11.05 8.39 6.91 7.65
Err. (%) 5.93 116.67 11.05 22.29 28.44 2.22

Ft NS (N) 14.32 1.85 8.09 9.79 6.03 7.91

ES (N) 13.98 0.94 7.46 7.96 5.07 6.52
Err. (%) 2.43 96.81 8.38 22.99 18.93 21.41

150 Fn NS (N) 39.82 5.23 22.53 10.34 6.26 8.30

ES (N) 38.05 3.98 21.02 8.35 5.56 6.96
Err. (%) 4.65 31.41 7.19 23.83 12.59 19.34

Ft NS (N) 20.19 4.09 12.14 8.95 7.39 8.17

ES (N) 16.54 2.57 9.56 10.23 7.47 8.85
Err. (%) 22.07 59.14 27.05 14.30 1.08 8.32

grinding process conducted by grinding wheels with large abrasive transferable to study other grinding issues (e.g. grinding heat,
numbers, high proportions of engaging grains that experience machined surface topography, grinding chatter) at micro (or grain)
rubbing stage. scale;

The eects of abrasive numbers seem to be large, even larger than
that of any machining parameters analyzed above, because the max-
imum and minimum contributions of cutting and rubbing forces are all
obtained in Fig. 18(g) and (h) by changing wheel abrasive numbers. It
might comment that, coarse-grained grinding wheels could achieve
higher cutting grain ratios than ne-grained wheels and therefore lead
to lower grinding forces, although the machined surface quality might
be degraded due to increased chip thicknesses. The careful selection of
grinding wheels with a proper abrasive numbers is of great importance
when reduced grinding forces are the main focus. If grinding wheel
service life and dressing cost are not the main consideration, grinding
wheels with small abrasive numbers (corresponding to large abrasive
grain diameters) are recommended.
7. Conclusions
In this paper, a novel detailed modeling methodology of grinding
forces has been analytically established, experimentally validated and
utilised to study a specic issue that previous methods can not address.
The contribution of this paper and some new ndings obtained by the
proposed method might include:
For the methodology:
A new methodology to predict grinding forces with detailed
information (e.g. three components including rubbing, plowing
and cutting forces) has been proposed, in which (i) a new strategy
to determine each grain-workpiece interaction at each grinding
moment and (ii) a novel derivation of grinding kinematics consider-
ing dierent grain protrusions have been suggested. Except for
grinding forces, the proposed methodology also enable the avail-
ability of other grinding process details at the grain scale (e.g. the
ratios of grains that are experiencing rubbing, plowing and cutting
stages to the total engaging grain number). Because it is based on
realistic grinding process, the proposed method is anticipated to be
For the results:
Unlike the prediction of average [1,4,6,7,1114] and maximal
forces [1725] in previous studies, numerical grinding forces with
detailed information (e.g. rubbing, plowing and cutting force
components) in the full time-domain, to the best knowledge of
authors, have been obtained for the rst time;
Moreover, validation experiments have proved that, the ob-
tained numerical forces are, to a large extent, similar to realistic
grinding forces in terms of (i) general features of the force signals
including upper envelopes, lower envelopes and moving average
curves (average error of 6.73%), and (ii) detailed features of the
spike signals including spike amplications and intervals (average
error of 14.3%)
To show the advance of the proposed methodology, the eects of
grinding conditions (including depths of cut, wheel speeds, work-
piece feed speeds and grinding wheel abrasive sizes) on each
component of grinding forces (rubbing, plowing, and cutting forces)
have been analyzed. Some new ndings have been gained as
follows:
Rubbing, plowing and cutting grain-workpiece interactions si-
multaneously exist in grinding and none of them is missing at any
grinding moments; Among the three possible stages, rubbing and
cutting are the two dominant ones while the plowing contribu-
tion is relatively limited;
Grinding wheel abrasive numbers have the largest impacts on
rubbing, plowing and cutting contributions and therefore should
be properly selected before grinding; The employment of grind-
ing wheels with large abrasive numbers easily leads to high
proportions of rubbing grains. Small abrasive numbers (corre-
sponding to large abrasive grain diameters) therefore are re-
commended if wheel service life and dressing cost are not taken
into consideration (because both the high cutting grain ratios and
low grinding forces could be achieved);

18
H.N. Li et al. International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx

Fig. 18. Eects of grinding conditions on grinding forces and each component including rubbing, plowing and cutting forces: eects of depths of cut on (a) normal and (b) tangential
grinding forces, eects of wheel speeds on (c) normal and (d) tangential grinding forces, eects of workpiece feed rates on (e) normal and (f) tangential grinding forces, and eects of
grinding wheel abrasive numbers on (g) normal and (h) tangential grinding forces (ap : depth of cut, vs : wheel speed, vw : workpiece feed rate; dgmean : average abrasive grain diameter;
workpiece material: AISI 1055 steel).

Depths of cut have the second largest impacts, and the dominant
force component might probably be changed from rubbing to
cutting by employing dierent cut depths; Shallow cut depths
easily result in a large rubbing contribution, which could also
explain the size eects in grinding (i.e. the specic energy
becomes higher when cut depths are decreased [69,70]); The
cut depth of 50 m would probably be a rational grinding depth
for AISI 1055 steel workpiece (because both relatively high
cutting eciency and good-quality surfaces could be obtained).
The eects of wheels speeds are also observable: fast wheel
speeds lead to less rubbing forces and more cutting forces, while
the eects of workpiece feed rates seems to be limited. Therefore
high wheel speeds and large workpiece feed rates are recom-
mended to achieve high material removal rates;
Although wide grinding conditions have been employed, the
cutting contribution is no more than 69%, indicating a great
potential of the machining eciency of the existing grinding
techniques. The random wheel topography might probably be
the key obstacle, and therefore the grinding tools with more
uniform grain protrusions as presented in Refs. [7173] are

19
H.N. Li et al.
expected to be promising.
The proposed methodology therefore are believed to be able to not
only enhance the existing understandings of grinding forces from macro
to micro scale, but also guide industrial grinding process to a large
extent (e.g. optimisation of grinding parameters to simultaneously
achieve machining eciency and avoid high grinding temperature
induced by high portions of rubbing grains), especially under the trend
of miniaturization and microfabrication where grinding details at the
grain scale are highly needed to optimise the micro grinding tool
eciency and micro-grinding accuracy.
Acknowledgements
The authors acknowledge the support from the National Natural
Science Foundation of China (No. 51275084).
References
[1] Malkin S, Guo C. Grinding technology: theory and application of machining with
abrasives. USA: McGraw-Hill; 2008.
[2] Rowe WB. Principles of modern grinding technology. Boston: William Andrew
Publishing; 2009.
[3] Marinescu ID, Rowe WB, Dimitrov B, Ohmori H. Tribology of abrasive machining
processes. Netherland: Elsevier; 2012.
[4] Werner G, Koenig W. Inuence of work material on grinding forces. Ann CIRP
1978;27:2438.
[5] Bttler E. Konzept und technologische Grundlagen zum Aufbau eines
Informationszentrums fr Schnittwerte fr die Schleifbearbeitung [Ph.D. thesis].
Germany: RWTH Aachen; 1979.
[6] Tnsho HK, Peters J, Inasaki I, Paul T. Modelling and simulation of grinding
processes. CIRP Ann Manuf Technol 1992;41:67788.
[7] Mishra VK, Salonitis K. Empirical estimation of grinding specic forces and energy
based on a modied Werner grinding model. Procedia CIRP 2013;8:28792.
[8] Fuh K-H, Wang S-B. Force modeling and forecasting in creep feed grinding using
improved BP neural network. Int J Mach Tools Manuf 1997;37:116778.
[9] Liu Q, Chen X, Wang Y, Gindy N. Empirical modelling of grinding force based on
multivariate analysis. J Mater Process Technol 2008;203:42030.
[10] Guo M, Li B, Ding Z, Liang SY. Empirical modeling of dynamic grinding force based
on process analysis. Int J Adv Manuf Technol 2016:111.
[11] Malkin S, Cook NH. The wear of grinding wheels: part 1 attritious wear. J Eng Ind
1971;93:11208.
[12] Li L, Fu J, Peklenik J. A study of grinding force mathematical model. CIRP Ann
Manuf Technol 1980;29:2459.
[13] Yao C, Wang T, Xiao W, Huang X, Ren J. Experimental study on grinding force and
grinding temperature of Aermet 100 steel in surface grinding. J Mater Process
Technol 2014;214:21919.
[14] Younis M, Sadek M, El-Wardani T. A new approach to development of a grinding
force model. J Eng Ind 1987;109:30613.
[15] Durgumahanti UP, Singh V, Rao PV. A new model for grinding force prediction and
analysis. Int J Mach Tools Manuf 2010;50:23140.
[16] Tang J, Du J, Chen Y. Modeling and experimental study of grinding forces in surface
grinding. J Mater Process Technol 2009;209:284754.
[17] Xie Y, Williams J. The prediction of friction and wear when a soft surface slides
against a harder rough surface. Wear 1996;196:2134.
[18] Badger J, Torrance A. A comparison of two models to predict grinding forces from
wheel surface topography. Int J Mach Tools Manuf 2000;40:1099120.
[19] Salonitis K, Stavropoulos P, Kolios A. External grind-hardening forces modelling
and experimentation. Int J Adv Manuf Technol 2014;70:52330.
[20] Hecker RL, Ramoneda IM, Liang SY. Analysis of wheel topography and grit force for
grinding process modeling. J Manuf Process 2003;5:1323.
[21] Shaw. M, Fundamentals of grinding. In: Proceedings of international grinding
conference new developments in grinding, Pittsburgh; 1972
[22] Hecker RL, Liang SY, Wu XJ, Xia P, Jin DGW. Grinding force and power modeling
based on chip thickness analysis. Int J Adv Manuf Technol 2007;33:44959.
[23] Chang H-C, Wang J-JJ. A stochastic grinding force model considering random grit
distribution. Int J Mach Tools Manuf 2008;48:133544.
[24] Park HW, Liang SY. Force modeling of microscale grinding process incorporating
thermal eects. Int J Adv Manuf Technol 2009;44:47686.
[25] Park HW, Liang SY. Force modeling of micro-grinding incorporating crystal-
lographic eects. Int J Mach Tools Manuf 2008;48:165867.
[26] Cao J, Wu Y, Li J, Zhang Q. A grinding force model for ultrasonic assisted internal
grinding (UAIG) of SiC ceramics. Int J Adv Manuf Technol 2015;81:87585.
[27] Jianhua Z, Hui L, Minglu Z, Yan Z, Liying W. Study on force modeling considering
size eect in ultrasonic-assisted micro-end grinding of silica glass and Al2O3
ceramic. Int J Adv Manuf Technol 2016:120.
[28] Johnson GR, Cook WH. A constitutive model and data for metals subjected to large
strains, high strain rates and high temperatures. In: Proceedings of the 7th
international symposium on ballistics; 1983.
[29] Azizi A, Mohamadyari M. Modeling and analysis of grinding forces based on the
20
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
single grit scratch. Int J Adv Manuf Technol 2015;78:122331.
[30] Xie Y, Williams J. The generation of worn surfaces by the repeated interaction of
parallel grooves. Wear 1993;162:86472.
[31] Holtermann R, Menzel A, Schumann S, Biermann D, Siebrecht T, Kersting P.
Modelling and simulation of internal traverse grinding: bridging meso-and macro-
scale simulations. Prod Eng 2015;9:45163.
[32] Holtermann R, Schumann S, Menzel A, Biermann D. Modelling, simulation and
experimental investigation of chip formation in internal traverse grinding. Prod Eng
2013;7:25163.
[33] Denkena B, Deichmueller M, Krger M, Popp KM, C. C, Schroeder A, Wiedemann S.
Geometrical analysis of the complex contact area for modeling the local distribution
of process forces in tool grinding. In: Proceedings of the 1st international conference
on process machine interaction; 2008.
[34] de Payrebrune K, Krger M, Deichmueller M, Denkena B. On contact modeling of
workpiece and grinding wheel with nonlinear elements. In: Proceedings of the 7th
euromech solid mechanics conference; 2009.
[35] Siebrecht T, Biermann D, Ludwig H, Rausch S, Kersting P, Blum H, Rademacher A.
Simulation of grinding processes using nite element analysis and geometric
simulation of individual grains. Prod Eng 2014;8:34553.
[36] Siebrecht T, Rausch S, Kersting P, Biermann D. Grinding process simulation of free-
formed WC-Co hard material coated surfaces on machining centers using poisson-
disk sampled dexel representations. CIRP J Manuf Sci Technol 2014;7:16875.
[37] Warnecke G, Zitt U. Kinematic simulation for analyzing and predicting high-
performance grinding processes. CIRP Ann Manuf Technol 1998;47:26570.
[38] Zitt U. Modellierung und simulation von hochleistungsschleifprozessen.
Kaiserslautern, Germany: University of Kaiserslautern; 1999.
[39] Aurich J, Braun O, Warnecke G, Cronjger L. Development of a superabrasive
grinding wheel with dened grain structure using kinematic simulation. CIRP Ann
Manuf Technol 2003;52:27580.
[40] Aurich JC, Herzenstiel P, Sudermann H, Magg T. High-performance dry grinding
using a grinding wheel with a dened grain pattern. CIRP Ann Manuf Technol
2008;57:35762.
[41] Braun O. Konzept zur gestaltung und anwendung deniert gesetzter
CBNSchleifscheiben. Kaiserslautern, Germany: University of Kaiserslautern; 2008.
[42] Herzenstiel P, Ching RC, Ricker S, Menzel A, Steinmann P, Aurich J. Interaction of
process and machine during high-performance grinding: towards a comprehensive
simulation concept. Int J Manuf Technol Manag 2007;12:15570.
[43] Perveen A, Rahman M, Wong Y. Modeling and simulation of cutting forces
generated during vertical micro-grinding. Int J Adv Manuf Technol
2014;71:153948.
[44] Tahvilian AM, Hazel B, Raeian F, Liu Z, Champliaud H. Force model for impact
cutting grinding with a exible robotic tool holder. Int J Adv Manuf Technol
2014:115.
[45] Xiao X, Zheng K, Liao W, Meng H. Study on cutting force model in ultrasonic
vibration assisted side grinding of zirconia ceramics. Int J Mach Tools Manuf
2016;104:5867.
[46] Li H, Yu T, Zhu L, Wang W. Analysis of loads on grinding wheel binder in grinding
process: insights from discontinuum-hypothesis-based grinding simulation. Int J
Adv Manuf Technol 2015;78:194360.
[47] Yu T, Li H, Wang W. Experimental investigation on grinding characteristics of
optical glass BK7: with special emphasis on the eects of machining parameters. Int
J Adv Manuf Technol 2016;82:140519.
[48] Li H, Yu T, Zhu L, Wang W. Modeling and simulation of grinding wheel by discrete
element method and experimental validation. Int J Adv Manuf Technol
2015;81:192138.
[49] Chen J, Fang Q, Wang C, Du J, Liu F. Theoretical study on brittleductile transition
behavior in elliptical ultrasonic assisted grinding of hard brittle materials. Precis
Eng 2016.
[50] Wu C, Li B, Yang J, Liang SY. Prediction of grinding force for brittle materials
considering co-existing of ductility and brittleness. Int J Adv Manuf Technol
2016:19.
[51] Li HN, Axinte D. Textured grinding wheels: a review. Int J Mach Tools Manuf
2016;109:835.
[52] Knig W, Lortz W. Properties of cutting edges related to chip formation in grinding.
Ann CIRP 1975;24:2315.
[53] Jackson MJ, Hitchiner MP. High performance grinding and advanced cutting tools.
USA: Springer Science & Business Media; 2012.
[54] Hou ZB, Komanduri R. On the mechanics of the grinding processPart I. Stochastic
nature of the grinding process. Int J Mach Tools Manuf 2003;43:157993.
[55] Liu Y, Warkentin A, Bauer R, Gong Y. Investigation of dierent grain shapes and
dressing to predict surface roughness in grinding using kinematic simulations.
Precis Eng 2013;37:75864.
[56] Wang D, Ge P, Bi W, Jiang J. Grain trajectory and grain workpiece contact analyses
for modeling of grinding force and energy partition. Int J Adv Manuf Technol
2014;70:211123.
[57] Popov V. Contact mechanics and friction: physical principles and applications. USA:
Springer Science & Business Media; 2010.
[58] De Vathaire M, Delamare F, Felder E. An upper bound model of ploughing by a
pyramidal indenter. Wear 1981;66:5564.
[59] Sin H, Saka N, Suh N. Abrasive wear mechanisms and the grit size eect. Wear
1979;55:16390.
[60] Lawn B, Wilshaw R. Indentation fracture: principles and applications. J Mater Sci
1975;10:104981.
[61] Ewens WJ. The sampling theory of selectively neutral alleles. Theor Popul Biol
1972;3:87112.
[62] Liu Q. Pattern recognition of grinding defects and assessment strategies of grinding
H.N. Li et al.
[Ph.D. thesis]. Nottingham, UK: The University of Nottingham; 2005.
[63] Brach K, Pai DM, Ratterman E, Shaw M. Grinding forces and energy. J Eng Ind
1988;110:2531.
[64] Lartillot O, Toiviainen P, Eerola T. A matlab toolbox for music information
retrieval. USA: Springer; 2008.
[65] Lartillot O, Toiviainen P. A Matlab toolbox for musical feature extraction from
audio. In: Proceedings of International conference on digital audio eects; 2007.
[66] Leeb SB, Shaw SR, Kirtley JL. Jr, Transient event detection in spectral envelope
estimates for nonintrusive load monitoring. IEEE Trans Power Deliv
1995;10:120010.
[67] Ohbuchi Y, Matsuo T. Force and chip formation in single-grit orthogonal cutting
with shaped CBN and diamond grains. CIRP Ann Manuf Technol 1991;40:32730.
[68] Matsuo T, Toyoura S, Oshima E, Ohbuchi Y. Eect of grain shape on cutting force in
superabrasive single-grit tests. CIRP Ann Manuf Technol 1989;38:3236.
21
International Journal of Mechanical Sciences xx (xxxx) xxxxxxxx
[69] Vollertsen F, Biermann D, Hansen HN, Jawahir I, Kuzman K. Size eects in
manufacturing of metallic components. CIRP Ann Manuf Technol
2009;58:56687.
[70] Cooper WL, Lavine AS. Grinding process size eect and kinematics numerical
analysis. J Manuf Sci Eng 2000;122:5969.
[71] Pacella M, Axinte D, Butler-Smith P, Shipway P, Daine M, Wort C. An assessment of
the wear characteristics of microcutting arrays produced from polycrystalline
diamond and cubic boron nitride composites. J Manuf Sci Eng 2016;138:116.
[72] Butler-Smith PW, Axinte DA, Daine M. Ordered diamond micro-arrays for ultra-
precision grinding an evaluation in Ti6Al4V. Int J Mach Tools Manuf
2011;51:5466.
[73] Butler-Smith PW, Axinte DA, Daine M. Preferentially oriented diamond micro-
arrays: a laser patterning technique and preliminary evaluation of their cutting
forces and wear characteristics. Int J Mach Tools Manuf 2009;49:117584.