PDA

View Full Version : Testing in game dive and zoom performance with E_Angle command

Wurkeri
01-16-2009, 05:07 AM
There has been a lot of discussion about measuring the in game angle of flight path on various testing related threads in this board. About 1,5 years ago I got an idea to use the devicelink speed and variometer data to measure true angle of the path. Basicly the idea is very simple; the speed and variometer values can be seen as separate vectors and the angle can be solved from these with basic trigonmetric functions. The devicelink speed value is given as IAS so the only additional calculation needed is IAS/TAS conversion and luckily the UdpGraph program offers functions for that and also the needed trigonometric functions. This turned out to work beautifully and the accuracy of the measurement appeared to be good at steady flight conditions.

Due to good results I contacted LesniHU and asked if it was possible to add such command to his Autopilot program to make a plane to keep wanted flight path during testing. Couple weeks later (Dec 2007) he had done this and the resulting E_Angle command appeared work well at steady flight conditions like climb and low angles of dive. However, in the accelerated conditions there was some unaccuracy caused apparently by sampling rate of the autopilot. LesniHU found solutions for this quickly; setting all speedmult coefficients to 1 in the Autopilot.ini solved the problem.

Last spring (May 2008) Holtzauge posted some dive and zoom results from his C++ simulation and we decided to make an comparison between the simulation an the game to find out if these behave same way. I did most of the testing last summer (while watching Football Euro Cup, it's pretty automatic process once you got settings right) but haven't had really time to compile the data. Anyway, here we go: The first graph shows speed vs time comparison of the P-51D in the case of a 20deg dive from 7000m to 4000m and then zooming up at 20deg, starting speed being 600km/h.

http://img388.imageshack.us/img388/7494/speedcompdg1.th.jpg (http://img388.imageshack.us/my.php?image=speedcompdg1.jpg)

The P-51D in the game and the C++ modeled P-51D show very similar speeds for first 20s. After that there is a separation, which seem to be caused by the drag rise due to compressibility which is modeled in the C++ model but apparently is not modeled in it's full extent in the game. The pull out is at 3g in the C++ model but the autopilot pulls out at much higher g, the highest recorded being over 8g. However, once setled to the zoom climb, the curves are quite similar again.

Next graph shows the acceleration vs Mach number in the dive part of the same case:

http://img83.imageshack.us/img83/2883/accelvsmachpw8.th.jpg (http://img83.imageshack.us/my.php?image=accelvsmachpw8.jpg)

There is consireable scatter in the case of the in game P-51D due to the time the autopilot needs to settle to the steady dive. However, despite scatter, the both datasets show similar trend up to about Mach 0,7 when compressibility start to affect the C++ P-51D.

In addition following graphs show few in game planes compared to the each other. All these are done with the same autopilot settings and tweaking the settings might give slightly different results. The La-5 was chosen for it's low altitude performance and the MiGs because these are basicly the same airframe with low and high altitude engine.

First of these comparess the flight paths of the P-51D, La-5, MiG-3 and MiG-3AM38 in the similar dive as above except that the starting speed is 400km/h (this gives more time for autopilot to settle to the dive):

http://img404.imageshack.us/img404/2245/pathgk6.th.jpg (http://img404.imageshack.us/my.php?image=pathgk6.jpg)

Nothing special here; the other planes show larger wawing than the P-51D. This is not a suprise because the Autpilot settings were originally tweaked for the P-51D. Otherwise the paths are quite similar once the planes settle to the dive or zoom.

Next one shows the separation for the first 60s in the same dive, zero level is the average of the planes:

http://img223.imageshack.us/img223/3223/sepa7xk9.th.jpg (http://img223.imageshack.us/my.php?image=sepa7xk9.jpg)

Then separation in the similar dive except that the starting altitude is 3000m and pull out altitude 500m:

http://img73.imageshack.us/img73/575/sepa3lt1.th.jpg (http://img73.imageshack.us/my.php?image=sepa3lt1.jpg)

And the last graph is an acceleration vs Mach number comparison in the last case, which shows that at slower starting speed there is much less scatter:

http://img234.imageshack.us/img234/8672/accelvsmach2oy9.th.jpg (http://img234.imageshack.us/my.php?image=accelvsmach2oy9.jpg)

All the in game tests were done with following E-Angle settings:

[E_Angle]
pGain=0.03
iGain=0.02
iLimits=1000
dGain=0.5
Spd000Mul=1
Spd100Mul=1
Spd200Mul=1
Spd300Mul=1
Spd400Mul=1
Spd500Mul=1
Spd600Mul=1
Spd700Mul=1
Spd800Mul=1
Spd900Mul=1
Spd1000Mul=1

And a sample script for a dive and zoom from 7000m:

W_RelTime>2
A_Bank=0
R_Slip=0
T_Spd=270
E_Angle=-10 [slow descent to test altitude 7000m from 7500m]
W_Alt<7000
E_Alt=7000
T_Spd=270.3 [400km/h TAS]
W_RelTime>180 [three minutes to settle]
C_WEP
T_Spd=2000
E_Angle=-20
W_Alt<4000
E_Angle=20
W_Alt>99999

edit: Corrected some typos.

JtD
01-16-2009, 10:44 AM
A bit of old news, mate. Still interesting for me. http://forums.ubi.com/groupee_common/emoticons/icon_smile.gif

M_Gunz
01-16-2009, 03:15 PM
Originally posted by Wurkeri:

W_Alt<4000
E_Angle=20
W_Alt>99999

I forget, sorry, is there a relative wait/time command available?
If so then to break the dive to climb transition into say 5 degree steps 1/2 second or so apart may benefit.
Maybe Lesnihu would consider one if not, like wait until now + value?

Tully__
01-17-2009, 02:01 AM
Some of the requirement for steady state conditions will be due to variometer values lagging several seconds behind actual rate of climb changes. This is deliberately modelled into the game as this is the way traditional aneroid variometers actually behave.

M_Gunz
01-17-2009, 07:28 AM
I remember an old pilot telling about not trying to 'fly the needle' back in 98 at the Delphi FSF.
He was talking about the VSI IRL.
Trying to hold a real plane level for long distances I did learn to anticipate. The Lesnihu AI uses something like that but
for changing conditions like dive to climb there is no way I know of to load in a new set of flying data.

Wurkeri
01-17-2009, 03:33 PM
Originally posted by M_Gunz:
I forget, sorry, is there a relative wait/time command available?
If so then to break the dive to climb transition into say 5 degree steps 1/2 second or so apart may benefit.
Maybe Lesnihu would consider one if not, like wait until now + value?

AFAIK there is no relative wait command available and I don't think that steps would help a lot because the delay of the variometer as Tully noted. I have made some tests using E_Pitch command (LesniHU suggested that) in the beginning of the dive and in the pull out, the idea being that the E_Pitch is "faster" command and therefore the plane settles faster to the new condition. However, there was no large improvement, at least with the settings I tried and if I adjusted PID values to give lower g values at pull out, the time to settle increased. Due to the faster nature of the E_Pitch, the steps might work with it but I have not tried that.

Devicelink gives the overload value so it might be possible to add a new command to the Autopilot which could be used to give constant load at pull out as in the Holtzauge's data.

JtD
01-17-2009, 03:37 PM
Originally posted by Wurkeri: Due to the faster nature of the E_Pitch, the steps might work with it but I have not tried that.

I did, didn't help much in my case but was still the best way. I'm using rather "lazy" stick inputs so the E_pitch lags behind as well.

Holtzauge
01-19-2009, 01:38 PM
Nice to see that you got the devicelink working and interesting results Wurkeri. I have been curious to see how the data compares and the new charts.

The speed/time chart I recognize from the first runs you made but it was interesting to see the acceleration/mach chart as well. The results seems to verify the theory that there is basically no drag rise due to compressibility modelled in Il2 and it was quite enlightening to see the unaffected acceleration in IL2 at M>0.70 for the P51.

The data I have used in the C++ code comes in part from high speed dive tests and wind tunnel tests performed by NACA. I also have some data for the Spitfire (from RAE)and Me109 which I have modelled as well. It seems both the Spitfire and Mustang do well in this area, the former due to the quite thin wing profile and the latter due to the clean airframe and laminar wing profile. The Me109 on the other hand, seems to have an earlier drag rise, at least according to data in Hoerner's fluid dynamic drag.

For the Fw190 (both A and D series)I lack data and have extrapolated from other aircraft data. BTW: If anyone has this type of data I would apprecite a PM.

The way things are modelled now, it seems that the performance of the aircraft at high speeds, i.e above M=0.65-0.70 bears little relation to the IRL behaviour. One can only hope that the design team behind the Il2 series will spend development effort on these kind of effects and not spend yet more effort on better grahics if forced to make a choise.

Concerning the separation tests, I was a bit surprised that the La-5 did not do better but OTOH I have not modelled this in the C++ code so I have no basis for a comparison really.

Wurkeri
01-20-2009, 02:38 PM
Originally posted by Holtzauge:
The speed/time chart I recognize from the first runs you made but it was interesting to see the acceleration/mach chart as well.

I choosed that way for the presentation because it shows the similar trends up to certain point well and in addition it shows the scatter due to slow settling clearly which is not really obvious in the speed vs time comparison.

Originally posted by Holtzauge:
Concerning the separation tests, I was a bit surprised that the La-5 did not do better but OTOH I have not modelled this in the C++ code so I have no basis for a comparison really.

It's the La-5, 1942 ie the slowest of them but still one of my favorites in the game. I purposedly selected rather neutral planes for sample comparison. After all the main reasons for this thread were the informing people about the E_Angle and comparison with your simulation.