Function tolerances with component contributions

The function tolerance values given above lack the regard of component tolerances like rail pressure sensor or pump curve. To consider several extreme value combinations tolerance scenarios were defined (Fig. 11). Tolerances can be just max. or min. with regard to nominal, however, gradient deviations are also possible as well as point-to-point (P2P) tolerances i. e. a random distribution.

This approach leads to an algorithm that basically consists of three nested loops: operating points, engine operation modes (like normal operation mode or diesel particulate filter regeneration mode) and tolerance scenarios. The variation of the tolerance scenarios alone (Fig. 11) would lead to billions of loop cycles. Therefore only 300 selected scenarios consisting of different min. and max. component
tolerances were part of the actual simulation run. The relevant tolerance scenarios are marked in grey in Fig. 11.

Possible variations and combinations

Without grad_up grad_down

Max

Min

P2P

RPS

X

X

X

X

X

X

ECU

X

X

X

X

X

X

Pump

X

X

X

X

X

X

Injection quantity

X

X

X

X

Inj. correction funct.

X

X

X

X

Fig. 11: Component tolerance scenarios

Considering the aforementioned component tolerances additional learning value spread is to be expected since errors of the system demand, the sensor as well as errors of the actuator FMU are superimposed. Analyzing these new resulting spreads it turns out that some tolerance scenarios are more severe with respect to certain parameters than others. Histograms that depict the number of scenarios that fall into a specific learning value spread class clarify that situation. The maximum AMC spread has been found for min. gradient pumps (Fig. 12, red circle). This scenario is further investigated.

15

-U

Max. AMC Spread [mmVs]

40

Ј

CD

C 30

 

Pump tol. = gradient down

 

Pump tol. = gradient up

 

EE 20

 

Function tolerances with component contributions

8 10

 

0

 

0

 

Max. AMC Spread [mm3/s]

 

40

Ј 35 re

I 30

5

 

1 30 « 25

 

Pump tol. = ‘

 

10

 

5

 

0

 

0

 

Max. AMC Spread [mm3/s]

 

Max. AMC Spread [mm3/s]

 

Function tolerances with component contributions Function tolerances with component contributions

Fig. 12: Histogram tolerance scenarios

Fig. 13 shows the complementary plot to Fig. 8 this time including component tolerances. Just as expected the number of squares and filled circles significantly increased illustrating that much more learning values lie beyond the thresholds.

MC value within thresholds • MC value < lower threshold ■ MC value > upper threshold

Engine speed [1/min] engine speed [1/min]

Fig. 13: Learning values w/ component tolerances

подпись: 
engine speed [1/min] engine speed [1/min]
fig. 13: learning values w/ component tolerances

MC value within thresholds • MC value < lower threshold ■ MC value > upper threshold

The benefit of the software function AMC is made visible in Fig. 14. The dark cloud of dots shows the spread of the AMC learning values. The light grey dots mark quantity balance scenarios that occur when AMC is not active. It can be seen that the quantity error spread is considerably reduced once the learning function is applied. This results in an improved control performance with an optimum control speed and reduced pressure overshoots.

* MonteCarlo value w/o AMC

* MonteCarlo value w/ AMC

Function tolerances with component contributions

Characteristic parameter

Fig. 14: Spread of AMC learning values compared to scenario w/o AMC

CONCLUSION

The tolerance calculation scheme described in this paper turns out to be a sophisticated tool to provide insight into the residual function tolerances of the learning function AMC. Based on a DoE pump model coupled with the regarded function algorithm and also taking component tolerances into account learning values can be generated for specific operation points. The randomly generated
operation points can be subsequently used for case analysis. The respective tolerance spread of the function becomes visible by producing a large number of learning values. The result proves to be an important part in the context of systems engineering providing information about system performance parameters and tolerances that are composed out of several contributions.

[1] RESULTS AND DISCUSSION

3.1 Investigations at high ambient pressure for a 12-hole injector

A chamber pressure of 1.5 MPa was chosen to model late injection timing, however, the temperature was kept at 20°C to study the influence of atomization at suppressed evaporation. These pressure conditions refer to the end of injection

[2] EXPERIMENTAL SETUP

For a deeper insight into atomization processes, in the first part of the study the macroscopic behaviour of several single component model fuels (n-hexane, n — heptane, n-decane) as well as a 3-component fuel is observed in an optical accessible pressurized and heated injection chamber. In order to assess the global atomization behaviour the liquid spray structures and droplet quantities of a 12-hole solenoid DISI-injector were analysed by Mie imaging and phase Doppler

[3]LVDT Linear Variable Differential Transformer

Добавить комментарий

Ваш e-mail не будет опубликован. Обязательные поля помечены *