Choosing Proper Filters for High-Speed Accelerometer Impact Sensing

[ConnorM]

I’m planning on using the ADXL375 (200g/3200Hz Bandwidth) or ADXL1001 (100g/11,000Hz Bandwidth) to measure the peak accelerations in a crash test dummy’s head during a bicycle/car crash. My goal is to sample at atleast 10,000 S/s using either a Teensy 3.6 or a Rasp Pi3.

The dummy will be mounted on the bike and launched at 20km/hr, then it will be struck by a car from the rear and from the side driving at 30km/hr.

My question is about filter choice after the analog accelerometer that I choose. Should a low or high pass filter be chosen, and where should the corners be?
I would just like some guidance and I’d like to learn how I can go about choosing proper filters for sensors. I have taken a course on electronics that briefly covered Chebychev and other popular filters.

 

[boneh3ad]

What filter you choose depends on what you are trying to resolve. You would certainly want some kind of anti-aliasing low-pass filter, but whether you wanted your cutoff to be lower than Nyquist or a high-pass of any kind depends on your measurement goals, really.

 

[ConnorM]

So for my situation,

- I am looking to measure the acceleration of a crash test dummy’s head during a very brief impact (15milliseconds).

- I want to be able to measure the accelerometer at 10 kSamples/s

- The current accelerometer that I am looking at measures +/-150 g’s with a bandwidth of 2 Hz —> 20 kHz (https://www.digikey.com/product-detail/en/te-connectivity-measurement-specialties/1-1001220-0/MSP1001-ND/279641)

What from these should I be using to determine my filter? I obviously want to be able to look at my acceleration data after and be able to analyze the head acceleration with as little noise from other sources as possible.

 

[boneh3ad]

Is there any reason you want 10 kS/s? That's extremely overkill for your chosen sensor.

 

[ConnorM]

10 kS/s is the specified minimum sample rate that sensors in crash test dummy’s should be sampled at. This is a design project through my school and we are trying to get as close to this sample rate as possible.

Also, 10 kS/s will give us some assurance that the peak acceleration was captured.

 

[boneh3ad]

Actually I misread your last post. Still, look up sampling theory and the Nyquist rate. In order to resolve the full 20 kHz of your sensor you would need to sample at 40 kHz. Since you are so far limited to 10 kHz, it means any content above 5 kHz in your signal will be lost. It also means you should use an anti-aliasing filter with a 5 kHz cutoff. You can look up anti-aliasing to check that out.

 

[ConnorM]

Oh thank you I have been reading about Nyquist Theory and you finally clued me into where the filtering aspect comes into play. This sensor can output 20 kHz but I can cut this to 5 kHz via a LPF, so I can sample at 10 kS/s for a full reconstruction of the analog wave.

So this would this ensure a smoother digital output? Are there any considerations I should make that would give me the clearest possible digital reconstruction?

 

[boneh3ad]

You might also consider the bit depth of your ADC.

 

[skeptic2]

Do you have an idea of how much deflection you expect the accelerometer to experience on impact?

 

[skeptic2]

It's always good to do a sanity check before you start building so that you know that it's going to work. So let's throw a few numbers around and see what happens.

In the first example with the bike going 20 km/hr, the car 30 km/hr and approaching from behind, the difference in speed is 10 km/hr. You are expecting the impact to ocurr in less than 15 ms with a peak acceleration of less than 150 G's. Converting everything to meters and seconds we get v = 2.78 m/s. Since I can only calculate average acceleration instead of peak, I'm going to use 75 G's instead of 150. v = a*t so t = 0.00378 s or 3.78 ms. (I multiplied 75 G's by 9.8 to get acceleration in m/s^2 of 735.) Using these numbers the bicyclist's head would move 735*0.00378^2/2 or 0.525 cm during impact? Does that seem reasonable?

Using the side collision at 30 km/hr the numbers we get are t = 11.3 ms, a = 75 G's and the distance the head moves during impact is 4.72 cm. Is this what you're expecting?