Closed Loop Test - Data Analysis

[Xplode]

When testing the dynamic behaviour of a vehicle, you have two options:

- Open loop test:
The whole test procedure is precisely determined (speed,steering wheel angle build up,..)
The driver is not part of the information loop and so there is no feedback.

- Closed loop test:
The test driver has to follow a road or trajectory, taking corners at different speeds, with unspecified steering wheel angles.
The driver introduces feedback in the information loop.

If you acquire data from such a closed loop test, you get a whole cloud of data,
which has to be filtered and curve fitted to achieve results similar to those from an open loop test.

If there is anyone who has experience analysing and processing this kind of closed loop data,
I would like to discuss the different, possible approaches and the quality of the results that can be achieved.


Thank you!

Greetings

Xplode

 

[GregLocock]

As a matter of interest what sort of manouevre / data are you talking about?



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Xplode]

In examining the possibilities of closed loop data analysis,
the focus is on steering and the appropriate manoeuvres:
constant radius turn, constant velocity sweeping steer manoeuvre for different lateral acceleration levels,..

The closed loop data is acquired from test runs on a handling track, performed at different paces.

The aim is to filter the closed loop data in such a way (limited speed range or lateral acceleration, only using the start of a steering manoeuvre,..) that similar results can be achieved as with the open loop tests.

Greetings

Xplode

 

[GregLocock]

Oh. The way I do it is to use closed loop runs, record what the driver does, play that into my model, change the model until it performs the same as the real car, then run the open loop tests on my model.

There are a few test we run open loop in the real world, you need a lot of real world.



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Xplode]

I'm sorry, I forgot to mention the data channels.

The data acquisition system measures:
- vx
- ax
- ay
- steering wheel angle and torque
- body motions: pitch, yaw and roll
- slip angle
- ...

Currently I was examining the results of steering wheel angle vs. lateral acceleration, as an example.

Greetings

Xplode

 

[Xplode]

I also thought about building a model and using simulations to match the model to the real world vehicle behaviour,
but I was still thinking about the necessary complexity of the model.

Do you use a full multi-body-model or a simpler, mathematical model?

Have you compared your simulation results of the open loop tests with real world open loop data?
Are these results satisfying or is there still a significant difference?

How do you measure the driver actions? Steering wheel angle, brake pressure, wheel torque,..?

For the closed loop tests, do you use a simple, easy-to-install data acquisition system (Microsat, VBox,..) or a higher quality measurement system?

Cheers

Xplode

 

[GregLocock]

I use ADAMS. I drive that model via recorded vehicle speed, and steering wheel angle, inputs. That is there is an internal speed controller in the model. I don't use braking.

I rarely correlate to open loop data, but yes, I have done so. We really don't run many open loop tests. We have a steering robot to run them.

Mostly the results are as good as the tire data. Sometimes the agreement is very good straight out of the box, other times it never works.





Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Bribyk]

If you want to look at the results quickly and want to have something to compare your model to, check out LapSim software, available free on Bosch's website (it's tricky to find on there though).

 

[cibachrome]

A closed loop test will reveal useful information about a driver and about a vehicle in moment control instead of displacement control. Depending on the level of "maneuvering severity" a number of useful numerical techniques are applicable. Just driving a vehicle straight (or trying to) can produce useful cross correlation metrics. System Identification techniques can show you the complexity of your system, and a simple Bode analysis can show you problems when your coherence goes on vacation. For those of you who are so sure that linear analysis is inappropriate to non-linear systems, you should re-think your methodology. If your test involves a fixed path with sideslip limits (for example: the Consumer Reports Avoidance Maneuver or CRAM Test), You can compare the actual path with the ideal path and produce a score based on path error. If you are at a race track, cross plots of steer moment or angle vs. yawrate or sideslip ought to show relevant difference in test conditions. What is necessary is a good model of the data so you can fit it to this model and score the RMS error. Having GPS position is handy for racing, too. Elevation on the path will show you how different drivers use change in elevation to 'surf' the course corners. If you run the course at controlled speed increments instead of best lap time, there will be enough data to produce a multiple input, multiple output system model which can capture the essence of you vehicle at various stages of the course.

Remember that the vehicle has no idea where it is, so steer angle inputs are easy to do in simulation and track testing, but the vehicle operates in moment control: It searches around for tire and suspension/chassis forces and moments to balance the moment in the steering shaft. Where it goes to get the job done is controlled caos. Yeah, really good cars go where they are pointed, but there's only a few of them...

And, Captain Peacock, I am unanimous in my opinion on this.

 

[GregLocock]

One of the classiest proofs that your measurement system is good is to integrate your yaw velocity reading and your vehicle speed readings, to provide a path prediction. If you were driving around doing a constant radius test (ie increasing speed on a constant radius curve) you should get some nice round circles that overlay. Of course if you've got GPS as well then you can do that for any manouevre, but we didn't used to have GPS.

I suppose working back from that you could use GPS to provide your yaw velocity data. Hmm.

I need to use a non linear model because my linear range work is largely calibration of the model, so that for the non linear range (ultimate handling, rollover, etc) I know the model itself is good, and all the problems are in the tire models.











Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Xplode]

First of all, thank you Bribyk for the useful tool!

I can conclude the simulation process is very helpful, as long as you have an appropriate and good model of the data.
Are there any system identification tools or techniques that are particularly useful or which result in high correlation? I have a basic knowledge on the identification process, but I'm not really into the more complex theories..

Is there anyone who has performed a 'simple' dataprocessing (filtering, statistical analysis,..) and curve fitting on such closed loop data? With any satisfying results?

I have performed a small test, with medium quality closed loop data from the handling track. I made a selection of data points, based on the range for speed and lateral acceleration, according to the procedure describing the open loop test.
As an example, I examined the cross-plot steering wheel angle vs. lateral acceleration.
A polynomial fit through the selection resulted in a similar curve as the open loop results.
As there was only data available from a limited number of runs, this explains the differences between open/closed loop results.

Has anyone performed similar dataprocessing?
Would a larger number of dataruns increase the accuracy of the closed loop results? Or is it hard to expect any decent correlation using this method?

Xplode

 

[GregLocock]

It should work fine. Expect to take between 3 and 6 runs for each test. Quality control is very important.



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.