Integration drift of logged channels.

[pdelrivo]

Hello everyone, my name is Pablo. I'm a Mechanical Engineering student and I'm currently working on a University project.

I am analyzing data from a skid pad test, and to gain insight on the vehicle behaviour, i'm trying to create some math channels to study slips (CG slip, individual wheel slip). I'm doing it this way because so far I have no data from Slip Angle sensors. I'm mainly working with the following (at least for this part of the analysis):
- Linear Acelerometer @ 50 Hz
- Gyro sensor (yaw velocity) @ 200 Hz
- GPS @ 20 Hz

So, for example, if I try to obtain a plot of the longitudinal speed (Vx) by integrating the longitudinal acceleration, the results "drifts" away, or moves all over the place. In other words, the resulting channel diverges. The same happens when trying to obtain lateral speed (Vy), or any other parameter.

So far I have tried using a low pass filter on the channel before integration and a high pass filter on the channel resulting from integration. Drift can be eliminated by the high pass filter, but the cut off frecuency has a big effect on the actual magnitude of the result. In fact, the impact is so big that I'm nowhere near trusting the result.

As far as I'm concerned, the logged data is of quite good quality, and makes all the sense in the world. Having said that, my knowledge and experience in signal processing are almost non existent.

Is there any other technique that I can try to obtain a trustworthy integrated channel?

[David Ferguson]

I think achieving good integrations from real-world noisy data is a real challenge.

I would suggest you set up some tests, where you know the answers, then work on scaling / filtering the data until the calculated results correlate with the measured values.

For example. If you have fitted an accurate wheel speed sensor (calibrate it with the GPS speed, but having both will help you understand the time shift typically found with GPS data -- calculating the speed and reporting it takes time), then the integration of your linear accelerometer should be able to equal the speed change from the starting point. And when holding a steady speed, the integration should result in close to 0 change in speed.

So if you setup an IR beacon on your skid pad, with your wheel speed sensor, you could determine the distance and average speed of a lap. Integrating the yaw rate channel to find change in yaw angle should produce close to 360 deg for one lap.

Good Luck!

[pdelrivo]

David, thank you very much for your response.

I do actually have a calculation for yaw angle and it does indeed result in nearly exactly 360 degrees of rotation. I think in this case, the integration works better because of the higher quality of the logged data. As stated before, we record the Gyro sensor at 200 Hz and, although I'm not a 100% certain, I think the sensor itself is more precise.

Although I do understand your proposal, and I agree with it, I don't think I'll be able to perform that in the short term. We do have wheel speed sensors but they are very rudimentary, really low quality data because we are running only one magnet per wheel, so it's really choppy. In the long term we might be able to perform some tests with a better setup.

Meanwhile I'm looking at Kalman filters.

Anyways, I'm open to further suggestions.

Thanks in advance.