[WIP] Improve robustness to bad and lost airspeed data #11846

dagar · 2019-04-14T00:52:03Z

Continuation of #10733

src/modules/commander/Commander.hpp

dagar · 2019-04-14T13:44:55Z

I'm not sure that I like commander taking on a stall speed and load factor calculation, but I don't have an immediate alternative to suggest architecturally.

We already have this accelerated stall logic in the FW position controller, although it was done with the minimum airspeed (to avoid parameterizing a stall speed). Ideally this will all be handled in the same place.

https://github.com/PX4/Firmware/blob/d36b06f779d49d8dc7387bf2b4fa3aade999ada3/src/modules/fw_pos_control_l1/FixedwingPositionControl.cpp#L482-L496

dagar · 2019-04-14T13:49:28Z

src/modules/commander/Commander.cpp

+	// Do actions based on value of COM_ASPD_FS_ACT parameter
+	status.aspd_fault_declared = false;
+	status.aspd_use_inhibit = false;
+	status.aspd_fail_rtl = false;


If these are expected to be set on every iteration what's preventing spamming mavlink messages every iteration?

If any of the fault conditions are triggered, then a fault tie is recorded here: https://github.com/PX4/Firmware/blob/a76726845a2299d7ea4da5c3b77a09f998fffe55/src/modules/commander/Commander.cpp#L4181

All faults have to clear for a minimum time here: https://github.com/PX4/Firmware/blob/a76726845a2299d7ea4da5c3b77a09f998fffe55/src/modules/commander/Commander.cpp#L4212

before fault_cleared can be set to true and _tas_use_inhibit is set to false.

Until this occurs, fault_declared cannot be set to true again which means that once a fault has been declared there is a minimum time of _tas_use_start_delay seconds before it can be cleared and re-declared.

dagar · 2019-04-14T13:50:03Z

src/modules/commander/commander_params.c

+
+/**
+ * RTL delay after bad airspeed measurements are detected if COM_ASPD_FS_ACT is set to 4. Ensure the COM_ASPD_STALL parameter is set correctly before use.


Missing short description.

LorenzMeier · 2019-04-14T16:16:18Z

@RomanBapst Help with review and testing would be appreciated.

Antiheavy · 2019-04-14T20:20:18Z

src/modules/commander/commander_params.c

+ * @group Commander
+ */
+PARAM_DEFINE_INT32(COM_ASPD_FS_ACT, 0);


TAS seems to be the more commonly used Param abbreviation for "airspeed". Should we rename this COM_TAS_FS_ACT?

src/modules/commander/commander_params.c

priseborough · 2019-04-22T01:04:11Z

Looking through the logs supplied by @Antiheavy in #10733 I think there are changes to significantly reduce false positives associated with the use of the innovation check, eg:

Checking the time integral of the exceedance so that transient errors due to gusts can be rejected.
Improve initialisation of wind states. Currently they are initialised to zero with a small uncertainty of 1m/s and take time to learn.
Increasing the EKF2_EAS_NOISE parameter. This is currently set at 1.4 m/s which is too low for windy conditions

However the load factor test on its own is already providing significant improvement in protection against a FOD or water pitot blockage so if EKF checks were be turned off by default, then the change could be rolled in now and the estimator related work rolled in with a second PR.

Extreme wind outside specified vehicle limits. Slow wind state learning contributes to test ratio failures during direction changes.
Windy weather inside vehicle limits. Wind vector converges sufficiently quickly this time.
EKF formed bad initial wind estimate.
Takeoff with pitot tube blocked with moisture:

Antiheavy · 2019-04-22T19:49:18Z

However the load factor test on its own is already providing significant improvement in protection against a FOD or water pitot blockage so if EKF checks were be turned off by default, then the change could be rolled in now and the estimator related work rolled in with a second PR.

This seems like a good idea for getting a basic and already tested safety feature implemented for v1.9

LorenzMeier · 2019-04-23T06:46:02Z

@dagar Could you fix the compile error so we can get this in with checks being off by default?

Bringing stuff in but disabling it and collecting log data is the secret behind successful software development in a lot of industries, in particular in apps and web services.

priseborough · 2019-04-24T00:02:28Z

I've added some changes that reduces the likelihood of false positives for users that want to try the innovation checks, but has them disabled by default so that if the user activates protection they will get the stall ratio test only unless they specifically set an innovation test ratio level.

By default protection and logging is off (COM_ASPD_FS_ACT = 0)

I will now add a patch that logs the required diagnostic data even when COM_ASPD_FS_ACT = 0

dagar · 2019-04-24T02:25:36Z

@dagar Could you fix the compile error so we can get this in with checks being off by default?

Yes I'll take another pass tomorrow.

dagar · 2019-04-29T01:35:15Z

I've added "(Experimental)" to all the parameter names and moved them to the "Developer" category.

Use the time integral of the exceedance to improve rejection of gust induced transients. Previous implementation could trigger above 0.7 * COM_TAS_FS_INNOV so fix scaling so that test ratios below COM_TAS_FS_INNOV cannot trigger. Change to an integer representation for the the threshold parameter and enable innovation test to be disabled by setting the threshold parameter to 0. Improve parameter documentation. The airspeed innovation test is disabled by default.

Always log the consistency checked variable, even when disabled. Revert the innovation threshold parameter back to a float to make its meaning more intuitive vs the estimator_status.tas_test_ratio it is compared against.

dagar · 2019-04-30T06:48:11Z

Merging this as discussed to facilitate more real world testing. Fair warning for any early adopters, I anticipate a bit of parameter naming churn.

dagar · 2019-04-30T07:08:40Z

This is now in master, thanks everyone!

dagar added Admin: Enhancement (improvement) 💡 Fixed Wing 🛩️ labels Apr 14, 2019

dagar mentioned this pull request Apr 14, 2019

Improve robustness to bad and lost airspeed data #10733

Closed

dagar requested review from priseborough, RomanBapst and Antiheavy April 14, 2019 00:56

dagar force-pushed the pr-airSpdFailSafe branch 3 times, most recently from cd16a54 to 1b44483 Compare April 14, 2019 03:03

dagar commented Apr 14, 2019

View reviewed changes

src/modules/commander/Commander.hpp Show resolved Hide resolved

dagar commented Apr 14, 2019

View reviewed changes

dagar force-pushed the pr-airSpdFailSafe branch from 1b44483 to a767268 Compare April 14, 2019 18:40

Antiheavy reviewed Apr 14, 2019

View reviewed changes

src/modules/commander/commander_params.c Show resolved Hide resolved

Antiheavy reviewed Apr 14, 2019

View reviewed changes

src/modules/commander/commander_params.c Show resolved Hide resolved

Antiheavy self-requested a review April 14, 2019 20:33

dagar self-assigned this Apr 24, 2019

dagar added this to the Release v1.9.0 milestone Apr 24, 2019

dagar mentioned this pull request Apr 24, 2019

Pr fw att airspeed scaling #11897

Merged

dagar mentioned this pull request Apr 24, 2019

WIP: FW attitude control limit airspeed scaling changes per iteration #11883

Closed

dagar force-pushed the pr-airSpdFailSafe branch from 2d45562 to fc30f09 Compare April 29, 2019 01:48

priseborough and others added 7 commits April 30, 2019 02:34

WIP: Improve robustness to bad and lost airspeed data

65602a7

minor cleanup

0f509d4

msg: Add logging of airspeed consistency check variable

9fa81d7

Commander: Improve airspeed consistency check logging

ffaf9b7

Always log the consistency checked variable, even when disabled. Revert the innovation threshold parameter back to a float to make its meaning more intuitive vs the estimator_status.tas_test_ratio it is compared against.

mark all new parameters Experimental and add to Developer group

28b005b

rename parameters

2ad6d55

dagar force-pushed the pr-airSpdFailSafe branch from fc30f09 to 2ad6d55 Compare April 30, 2019 06:38

dagar marked this pull request as ready for review April 30, 2019 06:39

commander cleanup _airspeed_fault_type on exit

b8302e7

dagar force-pushed the pr-airSpdFailSafe branch from d4cf777 to b8302e7 Compare April 30, 2019 06:46

dagar merged commit 9bad61b into master Apr 30, 2019

dagar deleted the pr-airSpdFailSafe branch April 30, 2019 07:08

sfuhrer mentioned this pull request May 3, 2019

airspeed architecture discussion #9148

Closed

weekly-digest bot mentioned this pull request May 5, 2019

Weekly Digest (28 April, 2019 - 5 May, 2019) #11965

Closed

sfuhrer mentioned this pull request Jun 28, 2019

Airspeed validation & selection module #12353

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[WIP] Improve robustness to bad and lost airspeed data #11846

[WIP] Improve robustness to bad and lost airspeed data #11846

dagar commented Apr 14, 2019

dagar commented Apr 14, 2019

dagar Apr 14, 2019

priseborough Apr 21, 2019

dagar Apr 14, 2019

LorenzMeier commented Apr 14, 2019

Antiheavy Apr 14, 2019

dagar Apr 29, 2019

priseborough commented Apr 22, 2019

Antiheavy commented Apr 22, 2019

LorenzMeier commented Apr 23, 2019

priseborough commented Apr 24, 2019

dagar commented Apr 24, 2019

dagar commented Apr 29, 2019

dagar commented Apr 30, 2019

dagar commented Apr 30, 2019


		/**
		* RTL delay after bad airspeed measurements are detected if COM_ASPD_FS_ACT is set to 4. Ensure the COM_ASPD_STALL parameter is set correctly before use.

[WIP] Improve robustness to bad and lost airspeed data #11846

[WIP] Improve robustness to bad and lost airspeed data #11846

Conversation

dagar commented Apr 14, 2019

dagar commented Apr 14, 2019

dagar Apr 14, 2019

Choose a reason for hiding this comment

priseborough Apr 21, 2019

Choose a reason for hiding this comment

dagar Apr 14, 2019

Choose a reason for hiding this comment

LorenzMeier commented Apr 14, 2019

Antiheavy Apr 14, 2019

Choose a reason for hiding this comment

dagar Apr 29, 2019

Choose a reason for hiding this comment

priseborough commented Apr 22, 2019

Antiheavy commented Apr 22, 2019

LorenzMeier commented Apr 23, 2019

priseborough commented Apr 24, 2019

dagar commented Apr 24, 2019

dagar commented Apr 29, 2019

dagar commented Apr 30, 2019

dagar commented Apr 30, 2019