532
5 CONCLUSIONS
In the paper, a procedure for variable selection for operational cycles identification has been
proposed. The basic idea is simple, as we expect to have cyclic nature of the process, we
should receive some repetitive pattern in the observed data. It is clearly seen in hydraulic oil
pressure data (HydOilP). However, due to extremely harsh condition in the mine, this measurement channel is frequently broken and there is a need to substitute this variable. So, to
find some expected cyclic pattern we decided to use simple tool for dependency analysis in the
data, i.e. autocorrelation function. Similarities in autocorrelation functions (ACF) estimated
for all variables are valuated and based on simple energetic criterion (RMS of residual signal
between HydOilP and others) preselection of variables has been done. After this stage we
obtained 11 variables with similar ACFs. It should be highlighted, that similar ACF does not
mean an exact similarity in the signal—they may be shifted in time and have the same cyclic
nature. So, the next step of processing is needed.
The second phase of the procedure is based on some engineering assumptions related to the
knowledge about the process. It is clear, that the unloading process requires a machine with a
stationary position (no driving speed), and it is (for safety reason) related to the usage of neutral gear (Gear = 0). To unload a haul truck, there is a need to “push out” the ore from the box,
so one might expect some fuel consumption (we assumed it as >10% of max fuel consumption
during the shift). Such simple 3 rules “clear” the remaining channels significantly and it allows
to notice that {breakP and TRNBPS} as well as {FuelUs, ENGTPS,intakeP} are interdependent. Finally, it seems to be obvious: Break Pressure and position of brake pedal are linearly
dependent; similarly as fuel consumption, engine boost pressure and deflection of the gas pedal.
So finally we might conclude that to identify operational regimes of haul truck one need
following variables: speed, gear number and one of {breakP, TRNBPS, FuelUs, ENGTPS,
INTAKEP} or ideally HydOilP. To improve the robustness of the recognition procedure data
fusion approach could be used that will be the next research task for our team. Comparative
study and effectiveness will be also tested on a wider population of signals.
REFERENCES
[1] McBain, J., Timusk, M.: Software Architecture for Condition Monitoring of Mobile Underground
Mining Machinery A framework Extensible to Intelligent Signal Processing and Analysis. In: 2012
IEEE Conference on Prognostics and Health Management, PHM, pp. 1–12 (2012), doi:10.1109/
ICPHM.2012.629954.
Figure 10. Final picture of variable selection.
breakP and TRNBPS
H l i l l i l l l • i1i~IIJI H 1l,
•
·~ 1 ~ 1 1 1 1 . I IIII
-
0
2
4
6
8
10
12
14
x 10 4
FueiUs, ENGTPS, lntakeP
H l l l l i l l l .• iii@IIJ~~~~~
•
Jill] I ; I ~ ~ II l
0
2
4
6
8
10
12
14
x 10 4
HydOIIP
H l l l ul , l . il li ll ill J, ~l i l l
•
: 1111111 :1111 1111 l
0
2
4
6
8
10
12
14
~ ..
x 10 4
5 CONCLUSIONS
In the paper, a procedure for variable selection for operational cycles identification has been
proposed. The basic idea is simple, as we expect to have cyclic nature of the process, we
should receive some repetitive pattern in the observed data. It is clearly seen in hydraulic oil
pressure data (HydOilP). However, due to extremely harsh condition in the mine, this measurement channel is frequently broken and there is a need to substitute this variable. So, to
find some expected cyclic pattern we decided to use simple tool for dependency analysis in the
data, i.e. autocorrelation function. Similarities in autocorrelation functions (ACF) estimated
for all variables are valuated and based on simple energetic criterion (RMS of residual signal
between HydOilP and others) preselection of variables has been done. After this stage we
obtained 11 variables with similar ACFs. It should be highlighted, that similar ACF does not
mean an exact similarity in the signal—they may be shifted in time and have the same cyclic
nature. So, the next step of processing is needed.
The second phase of the procedure is based on some engineering assumptions related to the
knowledge about the process. It is clear, that the unloading process requires a machine with a
stationary position (no driving speed), and it is (for safety reason) related to the usage of neutral gear (Gear = 0). To unload a haul truck, there is a need to “push out” the ore from the box,
so one might expect some fuel consumption (we assumed it as >10% of max fuel consumption
during the shift). Such simple 3 rules “clear” the remaining channels significantly and it allows
to notice that {breakP and TRNBPS} as well as {FuelUs, ENGTPS,intakeP} are interdependent. Finally, it seems to be obvious: Break Pressure and position of brake pedal are linearly
dependent; similarly as fuel consumption, engine boost pressure and deflection of the gas pedal.
So finally we might conclude that to identify operational regimes of haul truck one need
following variables: speed, gear number and one of {breakP, TRNBPS, FuelUs, ENGTPS,
INTAKEP} or ideally HydOilP. To improve the robustness of the recognition procedure data
fusion approach could be used that will be the next research task for our team. Comparative
study and effectiveness will be also tested on a wider population of signals.
REFERENCES
[1] McBain, J., Timusk, M.: Software Architecture for Condition Monitoring of Mobile Underground
Mining Machinery A framework Extensible to Intelligent Signal Processing and Analysis. In: 2012
IEEE Conference on Prognostics and Health Management, PHM, pp. 1–12 (2012), doi:10.1109/
ICPHM.2012.629954.
Figure 10. Final picture of variable selection.
breakP and TRNBPS
H l i l l i l l l • i1i~IIJI H 1l,
•
·~ 1 ~ 1 1 1 1 . I IIII
-
0
2
4
6
8
10
12
14
x 10 4
FueiUs, ENGTPS, lntakeP
H l l l l i l l l .• iii@IIJ~~~~~
•
Jill] I ; I ~ ~ II l
0
2
4
6
8
10
12
14
x 10 4
HydOIIP
H l l l ul , l . il li ll ill J, ~l i l l
•
: 1111111 :1111 1111 l
0
2
4
6
8
10
12
14
~ ..
x 10 4