jsbsim
jsbsim copied to clipboard
JSBSim-Team
Use gamma to help trim solver?
@bcoconni I'd noticed you periodically having to change the initial theta angle in the initialization file for the Short S23 aircraft in order to ensure that the trim solver found a solution - https://github.com/JSBSim-Team/jsbsim/commit/e0f1449ca56bcfd98dabe8a4d87d580b1bedeeb5
When I was debugging the trim differences I was seeing with the 737 model (negative elevator drag etc.) I noticed fairly different trim results depending on the initial theta value specified. In a much older (many years) version of JSBSm I had specified an initial theta of 4.1 degrees and the trim solution was level flight with a theta of ~ -0.5 degrees.
But with the latest JSBSim it results in a climb of ~2 degrees with a similar value of ~ -0.5 theta degrees. But if I specified an initial gamma value of 0 degrees then the initial value of theta appears to be ignored and I get a trim result very close to the older version of JSBSim.
So I tested out using gamma with Short_S23\reset01.xml and the script scripts\Short_S23_2.xml.
Theta -1.5
Using the currently comitted version of reset01.xml
Trim successful
Trim Results:
Angle of Attack: -1.27 wdot: 8.22e-04 Tolerance: 1e-03 Passed
Throttle: 0.58 udot: -3.43e-04 Tolerance: 1e-03 Passed
Pitch Trim: 0.14 qdot: -3.74e-08 Tolerance: 1e-04 Passed
Roll Angle: -0.04 vdot: 4.54e-12 Tolerance: 1e-03 Passed
Ailerons: -0.01 pdot: -3.68e-08 Tolerance: 1e-04 Passed
Rudder: 0.00 rdot: 1.33e-20 Tolerance: 1e-04 Passed
Trim (Event 4) executed at time: 1.008333
velocities/u-aero-fps = 253.108948
velocities/v-aero-fps = 0.000596
velocities/w-aero-fps = -5.627488
attitude/psi-rad = 3.141593
attitude/theta-rad = -0.048410
attitude/phi-rad = -0.000681
Theta -1.0
Full Trim
Sorry, wdot doesn't appear to be trimmable
Trim failed
Trim Results:
Angle of Attack: -1.27 wdot: -8.35e+00 Tolerance: 1e-03 Failed
Throttle: 0.65 udot: -5.65e-01 Tolerance: 1e-03 Failed
Pitch Trim: 0.14 qdot: 2.86e-01 Tolerance: 1e-04 Failed
Roll Angle: -0.04 vdot: 2.22e-02 Tolerance: 1e-03 Failed
Ailerons: -0.01 pdot: 2.10e-02 Tolerance: 1e-04 Failed
Rudder: 0.00 rdot: -1.44e-04 Tolerance: 1e-04 Failed
FATAL ERROR: JSBSim terminated with an exception.
The message was: Trim Failed
Theta -1.0 Gamma 0.0
Full Trim
Trim successful
Trim Results:
Angle of Attack: -1.26 wdot: 6.85e-04 Tolerance: 1e-03 Passed
Throttle: 0.78 udot: -9.15e-04 Tolerance: 1e-03 Passed
Pitch Trim: 0.13 qdot: 5.02e-08 Tolerance: 1e-04 Passed
Roll Angle: -0.04 vdot: -2.22e-11 Tolerance: 1e-03 Passed
Ailerons: -0.02 pdot: -1.03e-09 Tolerance: 1e-04 Passed
Rudder: 0.00 rdot: 8.79e-20 Tolerance: 1e-04 Passed
Trim (Event 4) executed at time: 1.008333
velocities/u-aero-fps = 253.110318
velocities/v-aero-fps = 0.000583
velocities/w-aero-fps = -5.565552
attitude/psi-rad = 3.141593
attitude/theta-rad = -0.021985
attitude/phi-rad = -0.000705
Theta -10.0 Gamma 0.0
Full Trim
Trim successful
Trim Results:
Angle of Attack: -1.26 wdot: 6.85e-04 Tolerance: 1e-03 Passed
Throttle: 0.78 udot: -9.15e-04 Tolerance: 1e-03 Passed
Pitch Trim: 0.13 qdot: 5.02e-08 Tolerance: 1e-04 Passed
Roll Angle: -0.04 vdot: -2.22e-11 Tolerance: 1e-03 Passed
Ailerons: -0.02 pdot: -1.03e-09 Tolerance: 1e-04 Passed
Rudder: 0.00 rdot: 2.96e-21 Tolerance: 1e-04 Passed
Trim (Event 4) executed at time: 1.008333
velocities/u-aero-fps = 253.110318
velocities/v-aero-fps = 0.000583
velocities/w-aero-fps = -5.565552
attitude/psi-rad = 3.141593
attitude/theta-rad = -0.021985
attitude/phi-rad = -0.000705
Theta 0.0 Gamma 0.0
Theta 0.0 Gamma 0.0
Trim successful
Trim Results:
Angle of Attack: -1.26 wdot: 6.85e-04 Tolerance: 1e-03 Passed
Throttle: 0.78 udot: -9.15e-04 Tolerance: 1e-03 Passed
Pitch Trim: 0.13 qdot: 5.02e-08 Tolerance: 1e-04 Passed
Roll Angle: -0.04 vdot: -2.22e-11 Tolerance: 1e-03 Passed
Ailerons: -0.02 pdot: -1.03e-09 Tolerance: 1e-04 Passed
Rudder: 0.00 rdot: 6.20e-20 Tolerance: 1e-04 Passed
Trim (Event 4) executed at time: 1.008333
velocities/u-aero-fps = 253.110318
velocities/v-aero-fps = 0.000583
velocities/w-aero-fps = -5.565552
attitude/psi-rad = 3.141593
attitude/theta-rad = -0.021985
attitude/phi-rad = -0.000705
I haven't double-checked the trim source code and logic, but it appears based on these handful of tests if we specify the required gamma that the trim solver might not be as brittle/fragile and we won't have to periodically tweak the initial theta value.
I'd noticed you periodically having to change the initial theta angle in the initialization file for the Short S23 aircraft in order to ensure that the trim solver found a solution - e0f1449
Yeah, that's a pain. I have always attributed this to the limited abilities of the default trim solver. As you are pointing out, each time the Short S23 trim fails, I tweak the initial value of theta until it converges again and I check in the process that the converged solution is still the same, or is very close to the previous one.
The mere fact that trim convergence breaks following minor modifications to the code has strenghtened over the time my feeling that the default trim solver is not robust. So each time the Short S23 broke I jumped to the conclusion that the cause was the trim solver brittleness rather than the code modifications I made. I may have been wrong, I don't know.
When I was debugging the trim differences I was seeing with the 737 model (negative elevator drag etc.) I noticed fairly different trim results depending on the initial theta value specified. In a much older (many years) version of JSBSm I had specified an initial theta of 4.1 degrees and the trim solution was level flight with a theta of ~ -0.5 degrees.
But with the latest JSBSim it results in a climb of ~2 degrees with a similar value of ~ -0.5 theta degrees.
That's the kind of result that makes me uneasy about the exact cause of the trim convergence issues. Does the trim solver fail because its numerical algorithms are not robust enough ? Or does it fail because the code modifications I made were incorrect ?
However the 737 should not be affected by the recent changes to the propeller code. So this problem with the 737 might have been existing for quite some time. Have you tried to bisect the code changes to find at which point it has started to converge to a climb of 2 degrees ?
But if I specified an initial gamma value of 0 degrees then the initial value of theta appears to be ignored and I get a trim result very close to the older version of JSBSim.
That's interesting ! Could you please provide an example of an XML file reset*.xml that specifies initial conditions with gamma ? I infer from your examples that you specify both theta and gamma in the initial conditions. By the way, your last 2 examples are both named 'Theta -10.0 Gamma 0.0' but I understand from the console output of the last example that its title should be 'Theta 0.0 Gamma 0.0':
Theta 0.0 Gamma 0.0
Trim successful
Trim Results:
Angle of Attack: -1.26 wdot: 6.85e-04 Tolerance: 1e-03 Passed
[...]
I haven't double-checked the trim source code and logic, but it appears based on these handful of tests if we specify the required gamma that the trim solver might not be as brittle/fragile and we won't have to periodically tweak the initial theta value.
Neither have I. However I suspect the specification of gamma to take over theta specification, the latter being ignored. I will check the code later this week, unless you beat me 😄.
Could you please provide an example of an XML file... I infer from your examples that you specify both theta and gamma in the initial conditions.
Yep, I left theta in and simply tried different values, didn't try removing it completely, and added in gamma.
<?xml version="1.0"?>
<initialize name="reset01">
<!--
This file sets up the aircraft to cruise at 143kt indicated at 3000 ft.
-->
<ubody unit="KTS"> 150.0 </ubody>
<vbody unit="FT/SEC"> 0.0 </vbody>
<wbody unit="FT/SEC"> 0.0 </wbody>
<latitude unit="DEG"> 37.86 </latitude>
<longitude unit="DEG">-122.34 </longitude>
<phi unit="DEG"> 0.0 </phi>
<theta unit="DEG"> 0.0 </theta>
<gamma unit="DEG"> 0.0 </gamma>
<psi unit="DEG"> 180.0 </psi>
<altitude unit="FT"> 3000.0 </altitude>
</initialize>
By the way, your last 2 examples are both named 'Theta -10.0 Gamma 0.0'
Yep, sorry it was a typo.
Have you tried to bisect the code changes to find at which point it has started to converge to a climb of 2 degrees?
I haven't tried that (yet?). The original JSBSim code is from late 2014 that I'm using with my VSS sim and I'm in the process of updating to the latest JSBSim code. I did glance at the FGTrim code from 2014 and the latest and set some breakpoints and stepped through it to see if I could see any obvious differences that may account for the difference I was seeing but didn't spot any at the time. But I didn't spend too much time on it, partly since I had been planning on adding a flight path angle option to the VSS sim anyway.
Did wonder if the code from 2014 was defaulting to a gamma of 0 and using it even if it wasn't specified in the initialization file whereas the latest code wasn't using it at all unless explicitly listed in the initialization file.
Following your findings, I have checked if gamma was already used in our models and there are only a few models that use it in their initial conditions:
https://github.com/JSBSim-Team/jsbsim/blob/647510208c3e88ee9393b68351d591c09c2beecf/aircraft/737/cruise_init.xml#L9
https://github.com/JSBSim-Team/jsbsim/blob/647510208c3e88ee9393b68351d591c09c2beecf/aircraft/737/cruise_steady_turn_init.xml#L9
https://github.com/JSBSim-Team/jsbsim/blob/647510208c3e88ee9393b68351d591c09c2beecf/aircraft/c172x/elevator_doublet_init.xml#L12
and a couple of them use a value for gamma which is not zero.
https://github.com/JSBSim-Team/jsbsim/blob/647510208c3e88ee9393b68351d591c09c2beecf/aircraft/c172p/reset01.xml#L19
https://github.com/JSBSim-Team/jsbsim/blob/647510208c3e88ee9393b68351d591c09c2beecf/aircraft/c172r/reset01.xml#L19
According to the code in FGInitialCondition, setting gamma to 0 is equivalent to setting the vertical component of the velocity to 0.0. This may be equivalent to setting theta to 0 and α to 0 (but I need to understand what does FGInitialCondition::calcThetaBeta exactly do to be sure).
https://github.com/JSBSim-Team/jsbsim/blob/647510208c3e88ee9393b68351d591c09c2beecf/src/initialization/FGInitialCondition.cpp#L284-L306
Regarding your last example 'Theta 0.0 Gamma 0.0', I have tried theta to 0.0 (without specifying gamma at all) and the trim solver converges. So this might be an indication that only the vertical component of the velocity needs to be zero.
I took a quick look with my 737 example, using the following initialization file with gamma set to 0 and with the gamma element removed.
<?xml version="1.0"?>
<initialize name="Sortie_1_-_Task_1-ic">
<ubody unit="KTS"> 194 </ubody>
<vbody unit="KTS"> 0.0 </vbody>
<wbody unit="KTS"> 0.0 </wbody>
<latitude unit="DEG"> -33.9873333333333 </latitude>
<longitude unit="DEG"> 18.6086666666667 </longitude>
<phi unit="DEG"> 0.0 </phi>
<theta unit="DEG"> 4.1 </theta>
<psi unit="DEG"> 0 </psi>
<gamma unit="DEG"> 0.0 </gamma>
<altitude unit="FT"> 5000 </altitude>
<winddir unit="DEG"> 0.0 </winddir>
<vwind unit="FT/SEC"> 0.0 </vwind>
<running> -1 </running>
</initialize>
Setting a breakpoint in FGTrim::DoTrim() before it starts trimming I compared the initial condition values in fgic.
Gamma = 0, Theta 4.1
vUVW_NED: {327.43513999999982, 0.0000000000000000, -3.5527136788005009e-15}
vt: 327.43513999999982
alpha: -3.2550388563675582e-17
Gamma not specified, Theta=4.1
vUVW_NED: {326.59716240070310, 0.0000000000000000, -23.410775694715209}
vt: 327.43513999999993
alpha: 2.1700259042450376e-17
So yes with gamma = 0 the vertical velocity component is practically set to 0.
With gamma not specified it looks like the vertical velocity component is set to vt * sin(theta) which is why I see a trim result in this case which results in a climb.
As you may have noticed, I had to change yet another time the files aircraft/Short_S23/reset01.xml and aircraft/Short_S23/reset02.xml in commit c0b46aa74d19ad79aa91c551e4ed5ad548d55863
I tried adding gamma = 0 but to no avail.