pdf download - Software and Computer Technology - TU Delft

Diagnosing the Beam Propeller Movement
of the Frontal Stand 5.6 MBD Implementation 3
into the diagnostic engine. Other variables do not require the addition of a sensor, but require the
implementation of the discretizer defined in Appendix B.5 (e.g., Iset, Vact, etc.). The MBD-3 implementation
is shown in Appendix C.3.6, which lists a modified version of the model in which all
variables of Figure 5.3 are declared, and part of a behavioral rule. This model declares, in addition
to the six variables that were already declared as observables, 8 other variables as being observable,
namely I_mvr, I_to_motor, I_from_motor, real_speed, Vact, Vset, Iact and Iset. These are
chosen, because it is most reasonable that these internals can be measured by means of some sensor
or discretizer. The total amount of possible sensor set-ups are all permutations of theses 8 variables
(e.g., [I_mvr, I_set], [real_speed], [I_mvr, I_to_motor, I_from_motor]).
In order to calculate the best possible sensor set-up, given that the 6 observables of MBD-2 are
already defined, the values of these 6 variables should be known. These values are different for
many fault scenarios, and different values imply other values for the entropy of a particular sensor
set-up. To overcome this problem, the values of these 6 observables are given the observable values
defined by the fault categories. For each fault category, the set of values for the 6 observables of
MBD-2 is added to the model of Appendix C.3.6 (this version of the model shows the addition of
the fault scenario C1-values). Then, the entropy gain for all possible sensor set-ups is calculated
for that fault category. The entropy gain of a particular sensor set-up is obtained by weighting the
outcome of a fault category by the number of real-case fault scenarios that are in that particular
fault category. Ideally, the calculation considers all permutations of the 8 additional observables.
Because there is no tool that automates this extensive search process available, 5 sensor set-ups are
arbitrarily chosen. The outcome, of the examination that is presented below, is a list of these 5
sensor set-ups that is ordered by the ability to increase diagnostic accuracy of the target system.
Table 5.8 shows the entropy gain of the 5 chosen sensor set-ups in case a fault scenario in fault
category C1 occurred. It shows that adding the three additional sensors, I_mvr, I_to_motor and
I_from_motor, have the highest entropy gain, for this particular fault category. It is also interesting
to notice that adding a sensor for the real_speed has equal entropy gain as adding the two sensors
for I_from_motor and Vact. Table 5.9 shows the entropy gain of the same sensor set-ups in case
a fault scenario in fault category C2 occurred. This time adding sensors for I_mvr and I_set has
higher entropy gain than adding the three three sensors. Table 5.10 shows the entropy gain in case
a fault scenario in fault category C3 occurred. Finally, Table 5.11 shows the entropy gain in case a
fault scenario in fault category C4 occurred.
Additional measurements Entropy gain
I mvr, I to motor, I from motor 0.2945
I mvr, I set 0.2589
I from motor, Vact 0.2407
real speed 0.2407
I from motor 0.2407
Table 5.8: Some additional measurements for fault scenarios in fault category C1 (current error)
The entropy gain of a sensor set-up is calculated by weighting the entropy gains of the Tables
5.8, 5.9, 5.10 and 5.11 by the number of times a fault scenario from that category occurred in a
real-case setting. From Table B.2 the number of occurrences of each fault category (for S1-S11 and
S17-S42) is calculated:
• C1: 16 (CURRENT_ERROR = true)
• C2: 7 (SPEED_ERROR = true)
65