Low Pass filteration on accelerometers

[dimondjack]

Is there any sort of filtration in hardware for accelerometers? I've found that a low pass filter is helpful for smoothing out the data (see my posts in software section) and know that previously my other data system had low pass filters for certain sensors (temp, accel, etc.) built in to the circuit.

[brentp]

Hi,

Yes, the digital accelerometer has a built in 1KHz low pass filter that we're using as default. you can read more in the datasheet here:
http://www.kionix.com/sites/default/fil ... ev%203.pdf

At the request of other users we've disabled the software averaging until we come up with a more refined solution, such as performing oversampling and averaging those samples.

[dimondjack]

I agree averaging should be disabled unless it is oversampling. 1kHz doesn't really do a heck of a lot when tire rotations are in the order of 35 Hz (or much less) and Engine RPM is usually under 100Hz (though around 120Hz on the kart). Any motion I would want to measure with my accelerometers would be under a couple of Hz. Are you logging from the SPI output or the analog output? If you were logging from analog you could implement an additional low pass on future hardware versions.

Of course, if we can make sure that the sensor doesn't max out, we can do all of this pretty well in software (my other thread).

I noticed that the sensor has a max reading of 2Gs. When I didn't have the unit well vibration isolated (on a kart, mind you), I was getting G forces over 2Gs in noise. Even with isolators I'm still getting a significant amount of data (~10%) greater than +/- 2Gs. What could be the explanation for this? It is possible I'm pulling 2Gs laterally with the kart, but not very often (at all). Certainly not longitudinally (the data I was looking at). It also seems like I should see a cap at 2Gs if the sensor only reads that high.

[brentp]

The sensor has an internal 12 bit ADC with a specified 819 raw counts per G. So the math shows that it can really go to +/- 2.5G, which is why you're seeing that in your data files. Yes, we're using the SPI port to read the data to save ADC channels on the processor.

Even with using the SPI, a low pass filter can be added to the circuit. We'll add those extra pads to the Rev G board we're working on. In the meantime, it would be possible to patch in some caps using some fine wire if you're inclined to do so. That, and good isolation would help as you've noticed.

Thanks for your testing / feedback work so far! Karts are clearly one of the most challenging environments we'll be seeing, for sure.

[Physic Simulation]

I am looking to this post since I have interests into kart installation and I am physicist.

Is the board, inside the enclosure, suspended ? Like in a old CD player, the internal component is suspended to reduce jogger-generated vibration and keep the laser alignment. I know that the CD player has also memory buffer to temper tracking jump.

If I understand, as per note 2 on page 3 and "Application Design Equations" (page 7), external capacitors would modify frequency bandwidth.

May be suspended/damped board would be suffisant ?

[brentp]

Hi,

The PC board is hard mounted inside the enclosure like most normal pieces of electronic equipment. The best option is to mount it with vibration isolation, and then post process the data as described in this thread.

-Brent

[dimondjack]

I wanted to show everyone my suggested mounting of the RCP using vibration isolators. The pictures were a bit large, so I decided to go for a link to the album:
https://plus.google.com/photos/11759849 ... Pfrst2PowE

You can see that the unit is mounted using four vibration isolators. I probably could have actually gone with smaller ones which I may try in the future. You want to match the vibration isolator to the weight of the unit you are isolating to transmit as little vibration as possible. There is science behind this (you want to calculate the natural frequency of your unit on the isolators.....and stuff) but I've forgotten all of it and generally know that you should know with a smaller isolator if you can get away with it. In this case I didn't want the unit moving too much under accelerative loads, so I went one side larger to a 5/8" damper with a 1/4"-20 thread pitch. This required drilling four holes in the RCP mounts, making sure you drill far enough away from the unit to be able to fit the nuts on (it is tight). You do need to have at least three mounts, as the two slots that come on the RCP will not constrain the unit (need at least 3 points for full constraint!)

The data now looks a LOT better. It still needs post processing, but instead of being unreadable you can now pull very good trends out of it. In addition, you can see that it isn't maxing out the sensor all the time, and that most data points are next to each other. Keep in mind, this data has NO post-processing. (see pictures below - couldn't figure out how to attach them in text)

You can find the vibration isolators at McMaster Carr (www.mcmaster.com) using this link: http://www.mcmaster.com/#catalog/119/1375/=nhn7ik

I chose a male/female one because of ground clearance, but you can choose a male/male or female/female one all the same.

[brentp]

Excellent. Thanks for the update!

I wonder what the data would look like with the unit hard mounted and some hardware low pass filtering enabled.

-Brent

[Physic Simulation]

Nice results !

Once you damped the acc signal, reducing acc's peaks, you could low-pass signal (to have kart acc) and substract this resulting signal from original damped acc to generate high-pass signal (to have high frequencies only).

Then, you choose the "clipping" frequency upon high frequencies time-average tending toward zero.

Am I clear ? Don't know.

Having your undamped and damped date (csv format) I would be please to Excel-play with your data and come back with DSP analysis