Beruflich Dokumente
Kultur Dokumente
ScienceDirect
journal homepage: www.elsevier.com/locate/issn/15375110
Research Paper
article info
Traction efficiency of farm tractors barely reaches 50% in field operations (Renius et al.,
Article history:
1985). On the other hand, modern trends in agriculture show growth of the global tractor
markets and at the same time increased demands for greenhouse gas emission reduction
as well as energy efficiency due to increasing fuel costs. Engine power of farm tractors is
3 September 2014
growing at 1.8 kW per year reaching today about 500 kW for the highest traction class
machines. The problem of effective use of energy has become crucial. Existing slip control
Published online
approaches for farm tractors do not fulfil this requirement due to fixed reference set-point.
This paper suggests an optimal control scheme which extends a conventional slip
Keywords:
Slip control
Optimal control
and net traction ratio and adaptively adjusts the set-point under changing soil conditions.
The proposed methodology can be mainly implemented in farm tractors equipped with
Traction efficiency
hydraulic or electrical infinitely variable transmissions (IVT) with use of the drive torque
Traction parameters
feedback.
2014 IAgrE. Published by Elsevier Ltd. All rights reserved.
1.
Introduction
1.1.
In this section, the main factors contributing to traction efficiency are discussed. First, the wheel dynamics are briefly
described. The corresponding force diagram is given in Fig. 1.
The soil reaction force Fz acts against the axle load Fz,axle and
* Corresponding author.
E-mail addresses: osinenko@inbox.ru, osinenko@ast.mw.tu-dresden.de (P.V. Osinenko), geissler@ast.mw.tu-dresden.de (M. Geissler),
herlitzius@ast.mw.tu-dresden.de (T. Herlitzius).
1
Graduate student.
2
Scientific staff member.
3
Chairman.
http://dx.doi.org/10.1016/j.biosystemseng.2014.09.009
1537-5110/ 2014 IAgrE. Published by Elsevier Ltd. All rights reserved.
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
Nomenclature
ht
k
m
uw
r
az
bt
Fh
Fz
Jw
m
Md
mw
rd
s
v
vw
Traction efficiency
Net traction ratio
Horizontal force coefficient
Wheel revolution speed, rad s1
Rolling resistance coefficient
Wheel vertical acceleration, m s2
Tyre section width, m
Horizontal force, N
Normal force, N
Wheel moment of inertia around lateral axis,
kg m2
Vehicle mass, kg
Drive torque, Nm
Wheel mass, kg
Tyre dynamic rolling radius, m
Slip
Vehicle travelling velocity, m s1
Wheel travelling velocity, m s1
Fh
;
Fz
(2)
ri
Frr;i
Frr;e
;r
Fz e
Fz
(3)
k m re ;
(4)
jvj
;
rd juw j
s 1
rd juw j
;
jvj
if v rd uw ;
if v > rd uw :
(5)
21
(1)
k
1 s:
kr
(6)
Usually, the traction parameters k,r and the traction efficiency ht are considered as functions of slip. Some characteristic curves for different soil types are illustrated in Fig. 2.
The curves of the net traction ratio are shown without bias at
zero for simplicity. Details of zero-slip conditions have been
described by Schreiber and Kutzbach (2007).
It can be seen that, in general, maxima of ht(s) as well as
maximum achievable traction effort, characterised by k, are
different for different soil types.
1.2.
Improvement of traction
22
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
2.
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
u_ w;j
1
Md;j rd;j Fz;j mj ri;j ;
Jw;j
j 14v_
1
m
n
X
mk Fz;k
x_ f x; u; Q;
_ 0171 ;
Q
y x;
(9)
n
X
k1
23
!
re;k Fz;k Fd ;
(7)
k1
typically used in the three point hitch control and is performed by, for example, magnetoelastic sensors or strain
gauges installed in load pins. Usually, farm tractors have front
suspension and some have rear suspension as well which
allows wheel load to be measured using pressure sensors and,
possibly, induction sensors to measure the stroke displacement. Gyroscopes, yaw rate sensors, accelerometers or other
relatively cheap measurement devices may be additionally
used to improve estimation. If measurement of the rear wheel
load is not available, it can be computed using the force diagram in Fig. 3.
Using D'Alembert's principle for the sum of torques around
D0 in Fig. 3 yields:
Fz;r
1
Fg maz l lr Fz;f ld l Fd;x hd max hCG
ld
2
y :
Jyy m ld lr h2CG 4
(10)
24
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
(11)
*
Fz
p ;
bt =2r0
2p$105 $pt
(12)
Md Jw u_ w
ri :
rd Fz
(13)
n
X
!
mk Fz;k Fd
k1
v_
:
g
(14)
n
1 X
m Fz;k re :
mg k1 k
(15)
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
3.
3.1.
The suggested methodology of this paper extends a slip control algorithm, realised either by the three point hitch, TCS or
some other method, by incorporating an optimality condition
depending on two factors: the traction efficiency and performance. The modification is made in the form of a supervisor
which estimates the traction parameters online using the
drive torque feedback and measurement signals from sensors
which are often available. The estimated values are utilised in
estimation of the net traction ratio characteristic curve. The
optimality functional is formulated in terms of this curve, the
corresponding traction efficiency curve and one parameter to
balance these two factors. Uniqueness of a maximum of the
functional is shown. The computed optimal drive train slip
set-point is transmitted to the slip control method. The latter
together with the supervisor constitute the suggested optimal
traction control. This strategy is not used to predict the
optimal operating conditions or to define the machine and/or
tyre dimensioning as it is performed in traction prediction.
The goal of the approach is to change the set-point adaptively
during the field operation. To summarise, the suggested
optimal traction control includes the following steps:
1. obtain machine parameters (wheel radius, dimensioning
etc.) and operation strategy (efficiency or productivity),
2. perform measurements,
3. estimate the traction parameters,
4. estimate the net traction ratio characteristic curve,
5. compute the optimum of slip,
6. perform slip control with the computed optimal set-point
7. check soil condition change.
The optimisation in step 5 is one-dimensional and has
polynomial time complexity which indicates that the algorithm is efficient (Cobham, 1965). Supposing the optimum is
located within the unit interval and given the tolerance of 1/n
for some natural number n (that is, the outcome of the algorithm and the actual optimum will differ at most by 1/n), the
worst-case time complexity is O(n). The estimation process in
step 4 cannot be unambiguously performed with classical
model identification approaches from the current operating
point and generally requires some curve fitting algorithm
from a set of estimated points. Such a procedure may
comprise multidimensional optimisation which might be
computationally expensive. On the other hand, some parts of
the estimation can be carried out offline and the obtained
parameters can be used further online without considerable
hardware requirements. A variant of a such method is
currently used in the suggested optimal traction control and
comprises a set of 15 parameters obtained offline from typical
net traction ratio characteristic curves. The proposed algorithm is able to estimate the curve given one slip-k tuple. The
set of 15 parameters is built-in and not used as an input. This
is different from several traction prediction algorithms where
the user defines some empiric or measured wheel and soil
parameters with which the characteristic curve can be
25
obtained. Instead, the user only defines the strategy via one
parameter ranging from zero to one which corresponds to
emphasising traction efficiency or performance. The only
purpose of the parameter set used in step 4 is to reduce
computational load and to make the algorithm appropriate for
conventional microcontrollers. The same goal may be achieved by tuning the tolerance of the method in step 7 where
the soil condition changes are detected. Increasing a
threshold, beyond which changes in soil conditions are indicated, allows for more sparse optimum calculations and less
computational load. The details of the soil condition change
checking are described in Section 3.3.
3.2.
curve
(16)
(17)
where a0, c0, b0, c1, b1 are the unknown parameters. With such
a formula, appropriate accuracy of approximation can still be
achieved and different behaviour in the low- and in the highslip range can be captured. On the other hand, it can provide
the necessary convexity property for the optimisation problem to guarantee uniqueness of solution. Details will be discussed in the next section. The resulting characteristic curve
k(s) is equal to k0 (s)re.
The idea of the cheap algorithm is to provide parameters
q (a0,c0,b0,c1,b1)T of a k0 -curve given one user-defined point
(s,k0 ), i.e. to find q q(s,k0 ). For this purpose, a set of simplified
characteristic curves which roughly classify soil conditions
26
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
2
c1 expb1 sk2 ;
minimise kk a0 c0 expb0 s
(18)
subject to
a0 ; c0 ; b0 ; c1 ; b1 0;
where 2 denotes Euclidean norm. For a discrete set of N
points of a given k0 -curve, the objective amounts to:
N
X
2
k0j a0 c0 exp b0 sj c1 exp b1 sj
:
(19)
NRMSE
1
maxj k0j minj k0j
v
2
uP
u N
t j1 k0j k0 a0 ;c0 ;b0 ;c1 ;b1 ;s
N
(20)
0
ai;0 ai;1 k0 ai;2 k02 ;
i 1:::5;
(21)
where ai,0, ai,1, ai,2 are the subparameters. In this case, fitting
was done using polynomial approximation by means of
Vandermonde matrix for each parameter:
0
1 k01
B
0
B
Vi B 1 k 2
@
1 k07
0 2 1
k
10 2 C
k2 C
C;
A
0 2
k7
i 15:
(22)
i 15;
(23)
j1
This
problem
is
non-convex
since
(a0 c0 exp(b0s) c1 exp(b1s)) is a non-convex function for
arbitrary (a0, c0, b0, c1, b1), therefore, only a local solution is
possible. Nevertheless, a global solution is not crucial at this
stage. Satisfactory accuracy can be achieved by changing
initial conditions and running the optimisation algorithm
repeatedly. For all given curves, the normalised root-meansquare error (NRMSE) of fitting.
i 15;
(24)
provided appropriate accuracy. Here, bi,0, bi,1, bi,2 are the subparameters. Approximation NRMSE of parameters a0, c0, b0, c1,
27
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
b1 were 0.07, 0.15, 1.14, 0.05 and 0.18% respectively. The last
step is to determine curve index k for the given point (s,k0 ).
This is performed using the algorithm in Fig. 7. If a relevant
curve index is not found, the algorithm returns flag
UPD_CON 0 and UPD_CON 1 otherwise. In the latter case,
the set-point will not be updated. Further explanation is given
in the next section. All numerical procedures were performed
in MATLABR2010a on a platform with AMD AthlonProcessor/2148 Mhz and 1 Gb RAM.
The worst-case time complexity of the algorithm in Fig. 7 is
O(n). The k0 -curves estimated with use of this methodology are
without bias at zero which, in fact, corresponds to the external
rolling resistance coefficient re (see Schreiber & Kutzbach,
2007 for detail). Therefore, an offset should be performed
before estimation, namely, the current operating point (s,k)
should be set to (s,k re) as input to the algorithm in Fig. 7.
3.3.
ks; q; re
1 s:
ks; q; re re ri
(26)
(27)
The optimisation problem can be scalarised and reformulated in the following way:
maximise over s sks; q; re 1 sht s; q; re ; ri ;
subject to
0 s s1 ;
(28)
where s 01 is a user-defined parameter which characterises the operation strategy ranging from maximal traction
efficiency to maximal productivity.
Theorem 1. Consider optimisation problem (28) together with (26),
(27). Let the following conditions hold:
maximise over s
maximise over s
subject to
(25)
(29)
(30)
ks
;
ks r
(31)
(32)
s/
hs
ks0 ks r k0 sks
2
ks r
rks0
ks r2
(33)
It can be seen that hs0 > 0 for any sss~ since ks0 > 0.
The second derivative is:
2 !
ks0
2
:
ks r2
ks r3
00
00
hs r
ks
(34)
00
28
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
gs hs1 s;
(35)
00
gs hs 1 s 2hs :
(36)
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
29
Fig. 9 e Traction efficiency, net traction ratio, drive train (thick solid lines), traction power (thin solid lines) and power losses
(dashed lines) as functions of slip for three soil conditions.
section width etc., can be programmed into ROM of a microcontroller since they are changed rarely, e.g. when the vehicle
is equipped with other wheels. Information about the tyre air
pressure must be provided by the operator, i.e. driver, or by
means of sensors.
In the case where a relevant characteristic curve is not
found (UPD_CON 0), the algorithm does not update the setpoint and slip control is performed with the previously
computed reference. This step is needed to process a failure in
Fig. 7 and lasts for STD_T seconds after which the algorithm
tries to find a curve again. The parameter STD_T can be
adjusted. Tuning parameter Dk is used to detect noticeable
changes in soil conditions. It can be adjusted by the user.
Lower values would mean more frequent computation of setpoints and make the control system more sensitive to changes
in soil conditions and vice versa.
3.4.
Simulation results
Fig. 10 e Computed optimal slip (dashed lines), traction power (solid black lines) and power losses (solid grey lines) as
functions of the user-defined strategy s.
30
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
estimated by filtering the velocity and wheel speed measurements using a 4th-order Butterworth low-pass filter with cutoff frequency equal to 5 Hz and taking discrete derivatives
using the following discrete transfer function:
Wz
z1
;
Ts z
Fig. 11 roughly correspond to the transient phases in estimated re and k. After the soil conditions stabilse, the updating
of the set-points stops.
The results of k-curve approximation for soils I, II and III are
shown in Fig. 13. NRMSE for all three cases is below 1.3%.
Optimal traction control was performed with s 0.25. Results
for soil I are shown in Table 1.
It is seen that with optimal traction control, the traction
efficiency is almost the same as for conventional traction
control, while the net traction ratio and traction power are
23% and 43% higher respectively. The productivity is 8%
higher. The growth of the power losses is about 44%, which is
about as high as growth of the traction power. With no control, operating at full drive train power, the traction power
grew 19%, while the power losses were 33% higher. For some
practical purposes, such excessive growth of power losses
might be unreasonable since the increase of traction power is
only one half the increase in loss. Therefore, the value of slip
at about 13e14 % can be recognised as optimal for soil I.
Table 2 contains results for soil II. In this case, even the
maximum of the traction efficiency is not achieved with
conventional traction control. Optimal control showed 2.4
times the traction power than with 10% slip. The growth of
power losses is less, i.e. twice as high as with conventional
traction control. The productivity is 56% higher. Working at
10% slip is unreasonable and the tractor simply does not
achieve effective drive train power. However, further increase
of Pdrive becomes unprofitable since the traction power grows
42% higher, while the power losses are 61% higher.
31
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
ht , %
0.3
72.5
0.37
0.4
Ptr ,
kW
Ploss ,
kW
Productivity,
ha h1
33
12.5
3.38
72.4
47.1
18
3.65
70
56
24
3.81
ht , %
0.155
50
12
12
2.07
0.27
56
28.6
24.3
3.24
0.33
51
40.8
39.2
3.67
Ptr ,
kW
Ploss ,
kW
Productivity,
ha h1
3.5.
ht , %
Ptr ,
kW
Ploss ,
kW
Productivity,
ha h1
1.71
0.1
46
0.195
53.5
17
14.8
2.54
0.23
41
32.8
47.2
3.11
32
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
4.
Conclusions
Acknowledgement
The authors would like to thank A.Gunther, the AST, and K.
benack, the Chair for Control Engineering of TU Dresden,
Ro
for their valuable suggestions and comments. Special thanks
ll for the major ideas in developing the algorithm of
go to H. Do
the net traction ratio characteristic curve approximation. The
research was conducted in the framework of the agriculture
electrification project at the AST of TU Dresden.
references
b i o s y s t e m s e n g i n e e r i n g 1 2 9 ( 2 0 1 5 ) 2 0 e3 3
33