Note: Descriptions are shown in the official language in which they were submitted.
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
1
METHOD AND ARRANGEMENT FOR DETERMINING WEIGHT OF LOAD IN MIN-
ING VEHICLE
[0001] The invention relates to a method of determining the weight
of a load in a mining vehicle, in which method the toad weight is determined
on
the basis of measuring signals obtained from separate measuring means.
[0002] The invention further relates to an arrangement for determin-
ing the weight of a load in a mining vehicle, which arrangement comprises
means for determining the weight of the load.
[0003] Mining vehicles, such as dumpers and wheel loaders, trans
port blasted rock from a blasting location to a dump location. Because this is
a
fairly high-speed operation and the transportation distances are relatively
short, weighing must be done while the vehicle is moving so as not to disturb
production. For the following process, it is, however, necessary to know how
much blasted rock has been transported for further processing. Real-time
weighing information makes it possible to already monitor material flows
inside
the mine, thus facilitating production control and planning. Planning for
preven-
tive maintenance of machines is also made possible by utilising real-time
weighing information.
[0004] A known solution weighs a load by measuring the cylinder
pressure caused by the load in the lifting cylinder that moves the system made
up of lifting arms and a bucket or dump box. The pressure is measured on both
sides of the lifting cylinder several times during a certain measuring period,
and the load in the bucket is calculated on the basis of the average of the ob-
tained pressure differences. The effect of the tilting of the machine and the
po-
sition of the lifting arms or dump box on the pressure difference measured in
the lifting cylinder is compensated by means of compensation coefficients. The
calculation method is linear and load determination is done while the machine
is moving. When calibrating the measuring system, pressure is first measured
with an empty bucket or dump box and then by using a load having a known
weight in the bucket or dump box.
[0005] In a stable state, the obtained measuring values are rela-
tively correct and the load in the vehicle can be determined at an adequate
accuracy. The problem is, however, that due to the quickly driven and short
distances, weighing must be done during the drive, in which case the tilting
of
the vehicle, bumps on the road and several other factors affect the final
result
of the weighing, and in certain situations, a systematic error towards one
direc-
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
2
tion may easily occur. In addition, a problem with the solution is applying a
lin-
ear method to a non-linear system and that even additional measurings used
are not enough to compensate for all errors caused by drive-time measuring in
the level of the pressure signal. One drawback is using a fixed measuring time
when calculating the average of the pressure differences from measuring sig-
nals oscillating at varying period lengths.
[0006] WO publication W099/09379 discloses a method that utilises
a neural network and fuzzy logic to determine the weight of a mining vehicle
load on the basis of measuring signals measured by sensors. Variables to be
measured can be for instance the cylinder pressure of the lifting cylinders of
a
bucket or dump box, the tilting of the vehicle in both longitudinal and
lateral
direction and the position of the lifting arms of the bucket or the position
of the
dump box. The weight of the payload in the vehicle can be determined on the
basis of the measured variables and the dimensions and geometry of the
bucket or dump box mechanics. A non-linear model based on a neural network
and fuzzy logic leads to better weighing results than the linear method de-
scribed above, but the drawbacks of this method are the calibration of the ma-
chine, the large amount of training data required to define a calculation algo-
rithm, and the fact that the calculation algorithm is machine-specific.
[0007] US publication 4,919,222 discloses a method and apparatus
for determining the weight of a load in a loading vehicle. Determining the
load
weight is based on measuring the cylinder pressure of the lifting cylinders of
the bucket and the position of the lifting arms of the bucket when the bucket
is
lifted. A signal representing the load weight is defined on the basis of the
cylin-
der pressure of the lifting cylinders and the position of the lifting arms of
the
bucket and any random pressure variations in the measurements are removed
using curve fitting and averaging. The resulting curve representing the load
weight is interpolated or extrapolated in relation to curves defined during
the
calibration of the apparatus for the purpose of determining the weight of the
load in the bucket. A drawback in the method described in the publication is,
however, that the method is dependent on the lifting rate of the bucket that
needs to be taken into consideration in the method. In addition, when the
track
of the loading vehicle is very bumpy, thus causing the vehicle to tilt quite a
lot,
it is not possible to obtain a sufficiently accurate weighing result.
[0008] FI patent 94,677 discloses a method based on measuring
the deformation of structures for measuring loads directed to structures, espe-
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
3
cially the weight of a load in a vehicle. The method is suitable for
calculating
the load caused by static loads that are practically stationary in relation to
the
structures, but it cannot be used to calculate the load in a moving vehicle.
[0009] It is an object of the present invention to provide a new
method and arrangement for weighing the load of a mining vehicle, with which
method and arrangement weighing can be done at a sufficient accuracy even
when the vehicle is moving.
[0010] The method of the invention is characterized in that a non-
linear Kalman filter is used to determine the weight of the load.
[0011] Further, the arrangement of the invention is characterized in
that the arrangement comprises a calculation unit that is arranged to utilise
a
non-linear Kalman filter.
[0012] The essential idea of the invention is that the weight of a
load in a mining vehicle is determined by a non-linear Kalman filter that esti-
mates the weight of the load in the vehicle, which load weight cannot be di-
rectly measured, by means of measuring signals obtained from measuring
means located in the vehicle.
[0013] The invention provides the advantage that by using a non-
linear Kalman filter, a better estimate can be made on the weight of the
vehicle
load, because to solve a non-linear problem, a non-linear method is used, by
means of which it is also possible to minimise the impact of the noise
included
in the measurements on the estimated load weight. Another advantage is that
the calibration of the method is simple and that the method need not be spe-
cifically trained to identify different masses. Further, the determination of
the
load weight is done faster and more accurately than in the prior art methods.
[0014] The invention is described in more detail in the attached
drawings, in which
Figure 1 is a schematic representation of a dumper used in mines,
to which the method of the invention is applied,
Figure 2 is a schematic representation of a wheel loader used in
mines, to which the method of the invention is applied,
Figure 3 is by way of example a schematic representation of an ap-
plication of a non-linear Kalman filter and an apparatus that can be used to
determine the weight of the load for instance in the dumper of Figure 1, and
Figure 4 is a schematic representation of the operating principle of
the non-linear Kalman filter.
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
4
[0015] Figure 1 is a schematic representation of a dumper having a
body 1 on wheels and a dump box 3 fastened at its rear end by joints 2 to the
body 1. To empty the dump box 3, lifting cylinders 4 are connected between it
and the body 1, and when the dump box 3 is lowered to its down position, its
front end rests on top of supports 5. Further, the dumper has sensors 6 based
on gravitational force to measure the inclination of the body 1 in relation to
the
horizontal both in the longitudinal and lateral direction of the dumper. The
incli-
nation of the dump box 3 in relation to the body 1 can be measured for in-
stance by using angular sensors in the joints 2 or by measuring the volume of
pressure fluid fed into the lifting cylinders 4 and calculating the
inclination of
the dump box 3 on the basis of it and by means of the geometry between the
cylinder 4 fastening points and the joints 2.
[0016] Figure 2 is a schematic representation of a wheel loader hav
ing a body 1 on wheels and a bucket 9 fastened to it on lifting arms 7 through
joints 8, and the bucket turns around joints 10 in relation to the lifting
arms 7. A
separate tilting cylinder 11 tilts the bucket 9 in relation to the lifting
arms 7, and
a lifting cylinder 4 between the lifting arms 7 and the body 1 lifts the
bucket 9.
Further, the wheel loader has in the manner shown in Figure 1 inclination sen-
sors 6 based on gravitational force for measuring the inclination of the wheel
loader in relation to the horizontal on the basis of earth's gravity in both
longi-
tudinal and lateral direction of the wheel loader. The position of the bucket
9 in
the elevation of the body 1 can be defined by using angular sensors in the
joints 8, for instance, and calculating on the basis of the measuring
information
provided by them and using the geometry of the lifting arms 7 the lifting
height
of the bucket 9 when it is turned in the most upright position by means of its
turning cylinder 11. Alternatively, the lifting height can also be defined by
measuring the volume of pressure fluid fed info the cylinder 4, whereby it is
possible to calculate the lifting height on the basis of said volume and the
length of the joints and the cylinder 4.
[0017] Figure 3 is a schematic representation of an apparatus utilis-
ing a non-linear Kalman filter and suitable for determining for instance the
weight of a load transported by a dumper according to Figure 1, with which
apparatus it is possible to measure the load in the dumper when the vehicle is
either moving or stationary, in which case the method and apparatus of the
invention can also be utilised in connection with an automatic filling of the
bucket of a wheel loader to make sure that the bucket is full. For the actual
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
measuring, measuring sensors or measuring means are used, of which two
measuring sensors 2a and 2b are strain gauges, for instance, that are
mounted in a suitable place with respect to the joints 2 of the dump box 3 on
both sides of the dumper body 1. Further, the apparatus comprises sensors 4a
5 and 4b for measuring the pressures of the pressure fluid of the lifting
cylinders
4 on both the side of the lifting cylinders 4 where the pressure fluid is fed
and
the side from which the pressure fluid flows out. By means of these sensors,
the weight of a load can be defined at a sufficient accuracy in a basically
static
situation on a horizontal base.
[0018] Measuring signals from the strain gauges 2a and 2b are for-
warded through amplifiers 12 to a calculation unit 13 that calculates the posi-
tion of the dump box 3 that has been defined as described earlier, and from
the calculation unit 13, the parameter describing the position of the dump box
3
is forwarded to the input of a block 14 implementing the non-linear Kalman fil-
ter. The calculation of the position of the dump box 3 can also be included as
part of the actual Kalman algorithm. The block 14 implementing the non-linear
Kalman filter also receives measuring signals from the pressure sensors 4a
and 4b, the temperature of the pressure fluid from a temperature sensor 4c of
the cylinder 4 and the inclination of the vehicle measured by the inclination
sensors 6. The block 14 can be a microprocessor, signal processor or another
corresponding calculation unit capable of performing pre-programmed func-
tions.
[0019] When weighing the load while the dumper is either stationary
or moving, the operator lifts the dump box 3 in such a manner that it detaches
from the supports 5 shown in Figure 1. An indicator light then lights in front
of
the operator as a sign that only the cylinders 4 and joints 2 support the dump
box 3. After this, the operator presses the button for weighing the load. The
weighing can also start automatically after a certain period of time has
elapsed
since the dump box was lifted. The block 14 implementing the non-linear Kal-
man filter estimates the weight of the load in the vehicle on the basis of the
inclination of the dump box 3 calculated in the calculation unit 13, the meas-
ured cylinder pressures, the temperature of the pressure fluid and the tilting
of
the vehicle. Figure 3 also shows a memory unit 15 for storing for instance the
estimated weight of the load and other values measured, calculated or esti-
mated during the estimation of the load weight. The memory unit 15 also
stores the initial values required by the non-linear Kalman filter for
beginning
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
6
the estimation process and described in the description of the operation of
the
non-linear Kalman filter of Figure 4. When beginning the estimation process,
the initial values are read from the memory unit 15 to the block 14 implement-
ing the non-linear Kalman filter. The memory unit 15 can also be arranged as
part of the calculation unit 14, but for clarity's sake, the memory unit 15 is
shown as a separate component in Figure 3.
[0020] Figure 4 shows on a general level the operation of the non-
linear Kalman filter used in estimating the weight of a load to be weighed.
The
model of the weighing system, which comprises the dump box 3 or bucket 9 of
the mining vehicle, the lifting arms 7 and the lifting cylinders 4 and/or
tilting cyl-
inder 11 to move them, and the measuring means described above, is dy-
namic, non-linear and discretely-timed. The dynamics of the system can be
described by the equation
x(k + 1) = f ~k, x(k)~ + v(k) , (1 )
wherein x(k + 1 ) is the actual state of the system at the time instant k + 1,
f() is
a non-linear function corresponding to the state transition matrix of the
system,
x(k) is the actual state of the system at an earlier time instant k and vector
v(k)
is white process noise with a zero mean value that describes a modelling error
between the actual system and the model made of the system, the modelling
error having the expected value of
Ew(k)~ = 0
and the variance of
ELUk~UJ~T ~=Q(k)~k;
wherein Q(k) is a covariance matrix of the process noise, i.e. model noise,
8k~ is Kronecker's delta, wherein 8k~ = 1 when k = j and otherwise 0, and T de-
scribes the transposition operation of the matrix. For instance, when defining
the weight m of the dumper load, the system model can take into consideration
the high and low pressures Py and Pa of the lifting cylinder of the dump box,
the tilting y of the machine, the position s of the dump box, and the tempera-
ture L of the pressure fluid, e.g. hydraulic oil. A state vector x of the non-
linear
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
7
state model of the weighing system of the vehicle would then comprise six
elements
T
x=[m,Py,Pa~Y~S~L] .
Of these, all others but the actual weight m of the load are measurable vari-
ables. The measurement of the temperature L of the pressure fluid can also be
left out of the above-mentioned measurements without any essential change in
the accuracy of the estimate of the load weight m. The dependency of the load
weight m on said measurements is non-linear, i.e. the function f() describing
the dynamics of the weighing system shown in formula (1 ) is non-linear. In ad-
dition, other factors that are not directly measurable can also be taken into
consideration in the function f() describing the model of the load weight m.
[0021] The estimation of the state of the system and thus also the
weight m of the load using a non-linear Kalman filter is done as follows.
[0022] At the time instant k, the actual state of the system is x(k) 20.
The actual state 21 at the next time instant k + 1 is according to formula (1
)
x(k + 1) = f ~k, x(k)~ + v(k) ,
and the corresponding measurement 22 at the time instant k + 1 is
z(k+1)=h~k+l,x(k+1)~+w(k+1), (2)
wherein the measurement function h() is generally a non-linear function, but
within the scope of this invention, the measurement function h() can also be
linear, and w(k) is white measuring noise with a zero mean value that de
scribes the error summed to the measurements from the measuring devices
and measuring environment. The expected value of the measuring noise w(k)
is
E~w(k~~ = 0
and its variance is
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
E[w(k)R'(J)T ] = R(k)~k;
wherein R(k) is the covariance matrix of the measuring noise.
[0023] The estimate x~k~k) 23 of the actual state x(k) at the time in-
s stant k is an approximation of the conditional expected value of the actual
state,
x~k~k)~ E[x(k~Zk,
formed on the basis of measurements Z'' _ ~z(1),z(2),...,z(k)~ accumulated by
the time instant k. So as to be able to estimate the state of the system at
the
time instant k+1, the non-linearities of the system must be linearized from
the
function f() describing the dynamics of the model close to the state estimate
x(k~k)23 of the time instant k. The Taylor series development is used in the
linearization, and depending on whether only first-order terms are used or
whether second-order terms are also included, either a first or second-order
filter is obtained. Linearization of a non-linear function is also used when
calcu-
lating a measurement prediction z(k+l~k~ 25. By means of the Taylor series
development, the following representation is obtained for a second-order
filter
x(k + 1) = f Ck, x(k~k)~ + fx (Iz~x(k) - x(k~k~~
'r , (3)
+ 2 '~ ei Cx(k)- x(k~k)~T f~ (k~x(k)- x(k~k)~ + KAT+ v(k)
wherein nX is the number of states that in this case is six, e; is an ~~" n~-
dimensional basis vector whose r~" component is one and other components
are zero, KAT describes higher-order terms that in this case can be excluded
and
fx (k) _ ~Oxf (k~ x)T ]T x = x(kIk)
is the Jacobian 29 of the vector f calculated at ~(k~k) 23, and
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
9
.f~ (k~ _~ xOx.f ' (k~ x~~x = xCk~k~ (5)
is a part 29 of the Hesse matrix calculated on the basis of the i~" component
of
the vector f.
[0024] After the linearization, the prediction x~k+l~k) 24 of the state
x(k + l~k~= E j fCk, x(k~k~~~ + E~fx (k~x(k~- x(k~k~~~
1 ~~x l " -~T
+E 2~el~x(k~-x(k~k~~ .f~~x(k~-x(k~k~~
at the time instant k for the time instant k+1 is obtained as a conditional ex-
pected value of equation (3) formed on the basis of the measurements Z~' ac-
cumulated by the time instant k when the terms of a higher than second order
are excluded due to their minor effect. The accuracy of the calculation can,
however, be increased by taking the terms of a higher than second order into
consideration. Because on average, a first-order term has a zero mean value
on the basis of
x(kI k> ~ E[x(k~Zk ] ,
the following is obtained as the state prediction x(k+l~k~ 24 for the time
instant
k+1
x(k+llk~=fCk,x(klk~~+~ '~e~tr~f'~(k~(klk~~, (7)
wherein the tr operation is the sum of the diagonal elements of the square ma-
trix and P(k k~ 28 is the covariance of the state at the time instant k. The
pre-
diction error of the state is obtained by subtracting equation (7) from
equation
(3). By multiplying the thus obtained prediction error by its own
transposition
and by producing a conditional expected value from it in relation to the meas-
urements Zk, the predicted covariance P(k+l~k~ 30 of the state is obtained
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
P(k + l~k) = fX (k>P(k~k~fX (k)T
71x 71x
+ ~ ~ ~ e~ eT tr ~f~ (k)P (k~ k> f~ (k>P (k~ k~]+ Q (k~.
[0025] On the basis of the state prediction calculated in formula (7),
it is possible to calculate at the time instant k a prediction z~Iz+l~k~ 25
for the
5 measurement for the time instant k + 1
-~ n
z(k+l~k~=hCk+l,x(k+l~k~~+~~e~tr~h~(k~+P(k+l~k~,, (9)
wherein ei is i~" raZ-dimensional basis vector, and in the case of this
example,
10 n2 is five, i.e. the number of measurements. On the basis of the actual
meas-
urement z(k + 1) 22 and the measurement prediction z(k + l~k~ 25, it is
possible
to calculate the residual, i.e. innovation, c~(k+1) 26 of the measurement at
the
time instant k + 1
15 ~(k+1)=z(k+1~-z(k+l~k>, ( )
and the related covariance S(k + 1) 31 of the innovation is
S(k+l~k~ = hx(k+l~P(k+l~k)hx(k+1~T
lti Ih
+ 1 ~ ~ e; eT tr ~h~ (k + 1~P (k + l~ k~h~ (k + 1~P (k + 1~ k>]+ R(k~,
2 t=i ~=i
wherein corresponding to formulas (3) to (5)
hx(k+1>=[~xh(k+l,x>T]T x=x(k+l~k~ (12)
and
h~(k+1)=~ Xoxh'(k+l,x>~x=x(k+l~k~. (13)
[0026] The amplification W(k + 1) 32 of the filter can be calculated
from the formula
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
11
W(k+1>=E[~C(k+1)c~(k+1)T Zk,, (14)
wherein X(k + 1) is the prediction error of the state x(k + 1) 21 based on the
information available at the time instant k. The updated estimate of the
state,
i.e. the filtered value x(k + l~k + 1~ 27 of the state at the time instant k +
1 based
on the information available at the time instant k + 1 is
x(k+l~k+1)=x(k+l~k~+W(k+1>v(k+1~ (15)
and the updated covariance P(k + l~k + 1) 33 of the state at the time instant
k + 1 based on the information available at the time instant k + 1 is
P(k + l~k + 1) = P(k + l~k) - W(k + 1)S(k + 1) W(k + 1) T . (16)
[0027] The estimation of the system state, i.e. according to this in-
vention, also the estimation of the load weight m, by means of a Kalman filter
can, in principle, be divided into three parts: predicting the state,
calculating the
amplification of the filter, and calculating the residual of the measurement,
and
on the basis of these, it is possible to calculate an estimate for the system
state, and in this case, especially for the load weight m. The uncertainties
in
the weighing system model and the measuring devices affect through the state
covariance the amplification of the filter, with which the residual of the
meas-
urement is weighted in such a manner that in updating the state estimate, the
information provided by the measurements on the state of the system and the
state calculated on the basis of the system model are taken into account to a
suitable extent, since neither of them alone is completely reliable, i.e.
corre-
sponds to the actual system. The obtained updated values are further used in
forming the estimate of the next time instant. These calculation cycles are re-
peated until the state provided by the filter as its output, i.e. in this case
espe-
cially the weight m of the vehicle load, has settled to a certain level that
thus
corresponds to the estimate of the weight m of the load in the vehicle. The es-
timation can be ended for instance when the variance of the load weight esti-
mate is below a predefined limit value that can be changed, i.e. it is a
parame-
ter of the algorithm. To begin calculation the initial value X(00) of the
state es-
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
12
timate, the state covariance P~0~0~ corresponding to the initial state, and
the
uncertainties of the weighing system model and the measuring devices are
required, all of these being stored in the memory unit 15, from which they are
read to the block 14 implementing the non-linear Kalman filter when weighing
is started. Values set at the factory to the vehicle in question can be used
as
the initial values. The first measurement can also be used as the initial
value
for the states to be measured, in which case the actual estimate calculation
is
started from the second measurement. The reason why the estimated value of
the load weight m does not immediately at the first Kalman filter calculation
cycle give the correct result is due to the fact that the calculation is
started from
the initial value of the state that is not necessarily correct. In addition,
there is
interference in the measuring signals especially at the beginning of the meas-
urement that first must be filtered by the Kalman filter.
[0028] To calibrate the weighing system, the vehicle is loaded with a
test load of known weight. To perform calibration for an empty dump box or
loading vehicle, it is enough to weigh the empty bucket and one known test
load, but it is also possible to use several test loads of different weights.
The
calibration is performed specifically for each machine. Further, the
calibration
can be performed again during the use of the machine to compensate for the
impact of changes caused by aging of the machine or change of components.
In connection with the calibration, the non-linear Kalman filter can also be
used
to estimate the parameters of the non-linear model of the weighing system.
[0029] Correspondingly, in the manner described above, the weigh
ing can be done by means of a wheel loader, in which case the position of the
bucket and other factors can easily be taken into account. In the case of a
wheel loader, it is in principle possible to use the measuring diagram of
Figure
3, in which case the position of the bucket 9 in the elevation of the body 1
and/or the inclination of the lifting arms 7 are taken into account in the
weigh-
ing system model. Thus, the state vector x of the model and the functions rep-
resenting the system dynamics change from what is stated above while the
principle of load weight m estimation remains the same.
[0030] The drawings and the related description are only intended
to illustrate the idea of the invention. The invention may vary in detail
within the
scope of the claims. Thus, the structure of the mining vehicle need not be ex-
actly as described in Figures 1 and 2, but the essential thing is that the
estima-
tion of the load weight is based on estimating by means of a non-linear Kalman
CA 02431584 2003-04-16
WO 02/39073 PCT/FI01/00973
13
filter the states of a non-linear model formed of the weighing system. Special
applications of the Kalman filter, such as a Wiener filter or the like, can be
used
in a corresponding manner to determine the weight of the load in the mining
vehicle.
