Accelerometers

So… I’ve got a project that has fallen into my lap that I think I can do a good job on but it depends on having a quality accelerometer.

I’m talking about mostly the 0-2G range, maybe maybe maybe 3-4G if initial testing is successful and it can be extended to wilder applications. Also is going to need gyroscope so I’ve been looking mostly at 6 axis parts. Space is NOT a concern so running independent accelerometer and gyroscope is totally an option if it makes sense.

The big thing that I’m going to need is the highest sampling speed possible. I’m trying to reconstruct position from acceleration data, hopefully down to the sub-1ft resolution over the course of a couple hundred feet.

I’ve found a couple of parts that seem like they might be up to the task:

Invensense IAM-20680 https://www.invensense.com/wp-content/uploads/2015/10/IAM-20680-Product-Brief-v1.01.pdf

ST LSM6DS33 http://www.st.com/resource/en/datasheet/lsm6ds33.pdf (claims 1600hz sampling rates possible)

Wondering if anyone has used either of these and has any input? Or an alternate part that is known to work well at high sampling rates? I know that noise filtering is going to be a big deal and am prepared for that.

-D

Would any of the one from aftermarket performance ecus for racing data be anywhere near the tolerances you need?

If you are not limited to MEMS parts, you can get other types of accelerometers like piezo ones that can be accurate to ug and go up to several kHz.

Jim, do you know any part numbers or manufacturers to look at? I’ve only ever seen MEMS parts.
-D

Not sure what kind of piezo accel you’re referring to, but most of them are AC devices, which are horrible in his use case (in particular, looking for position.) They do make DC piezos, and in this use case, it is a must. They all work by integrating the acceleration data into a velocity and position curve, but ac based devices tend to accumulate errors (as they bounce above and below the zero point) and will very quickly get out of sync with the reference frame. DC based devices always push the error into a single vector that can easily be factored out.

I would tend to prefer the dc piezo over a mems device as they are more accurate over a much larger range of temperatures and g-loads… but depending on the application, you might just be able to get away with a simple ST Micro device, especially if the motion has a short time frame.

In either case, I would definitely couple it with a gyro, as gyros tend to hold their reference over a longer timeframe and be used to calculate the naturally induced error of the accelerometers. (Gyros are great for long duration duty and horrible at short duration… but accelerometers are the exact opposite.) Getting a chip with the compass integrated will help as well… as that can be used to fix the gyro data as it drifts.

If you really want accuracy, I’d add a GPS as well… that’s basically what I developed when I designed the VirtualHUD Wingman device. (https://www.youtube.com/watch?v=bEfHjxRkD7Y)

Bill

Bill,

I was under the impresion that GPS was too slow to be useful? I also have no real interest in long-term position, just short term position starting from stationary. I’m dealing mostly with (relatively) linear motion along with axial. Trying to observe the first 250-500ft of motion starting from a dead stop in figure 2 seconds or so. If I can get within a foot of actual position at 50-100hz, I’ll be tickled. My interest in the gyro is as much to tell chassis rotation as anything, which provides a different piece of useful data.

thanks!
-D

DC Piezo is what’s used these days in test equipment. Very linear, 50kHz + rate in many cases. Can be expensive if you need top specs.

You really may be able to do this with a mems chip. The advantage to DC piezo from my perspective is the linearity and repeat-ability. No need for calibrations, etc. As far as speed of sampling and signal integrity goes, you’re only limited by your preamp and your ADC.

I’ve never used an IC or ASIC version, so no idea on parts recommendation.

If price and scale are limiting, I think this really could be done with a good mems part. Just try to find one designed for telemetry, acoustics, seismography, etc. I think you may have had bad experiences with accelerometers in the past because the vast majority are now made for consumer cell phone, hard drive, laptop, etc…

Analog devices usually makes good ones and produces for both markets…

L

I don't think it will meet your accuracy requirements, but...

There are integrated autopilots for the drone market that work for
air, ground, or water vehicles. They tend to have accelerometers,
gyros, and compasses built in, with connectors for external GPS
modules.

I'm thinking you can find one with an adequate processor, and write or
modify your own code for it.

There are Arduino based ones (that use the processor, not the Arduino
products), one called PixHawk which I think is a 32-bit arduino style
processor, and some newer Raspberry Pi based ones, including some
recent hobbyist boards that are shields for PiZero.

The ArduPilot Module (APM?) is available dirt cheap as an open source
knockoff from China ($10?). I haven't looked recently, but the
PixHawks were nominally $200 last fall, and I suspect knockoffs are
available as well. I know nothing about sampling rates.

Where are you getting the code to integrate the sensor outputs to get
output position? I'm interested in something like this as well, with
not nearly your accuracy requirements, to keep track of your (biking
or walking) position when you and the GPS are under tree canopy.

Matthew

Matthew: planning on writing code. I’m dealing with things accelerating and moving considerably faster than a human could. my issue is that I need to track acceleration from stationary to ____ over a distance of a couple hundred feet in <2 seconds.

Lorin: I’ve never used an analog output MEMS part, only ever the integrated SPI/I2C interface ones. Never considered that, thanks.

-D

I was thinking about PCB Piezotronics (pcb.com). I co-oped for one of their child companies in college. Though looking at their site, I guess they make mostly AC ones that may or may not work for you depending on your measurement timeframe. Maybe you can also pair up a couple accelerometers to get more extended frequency response.

Cheers,
Jim

Not sure if you are in controlled conditions but there are some pretty accurate laser distance devices.

Also, what about those G meters for car testing?

Laser: considered, but probably not going to work for this particular application.
G meters for car testing: way too coarse. You can do better with any number of accelerometers
pcb.com: will check out, thx

-D

PCB piezo test units are the ones I’ve used.



From: Jim Chen
Sent: Tuesday, January 10, 2017 12:53 PM
To: cincihackerspace@googlegroups.com
Reply To: cincihackerspace@googlegroups.com
Subject: Re: [CHP] Accelerometers

I was thinking about PCB Piezotronics (pcb.com). I co-oped for one of their child companies in college. Though looking at their site, I guess they make mostly AC ones that may or may not work for you depending on your measurement timeframe. Maybe you can also pair up a couple accelerometers to get more extended frequency response.

Cheers,
Jim

I have worked with the bosch motorsports mm5.10 style if I remember the model right. But it was put on a already made data acquisition pack. Stats may be on the bosch motorsports page still but I have no idea if it would work with your coding. Sounds like a bad ass setup your going to have.

I was under the impresion that GPS was too slow to be useful?

The nice thing about GPS data is that it can be time referenced very accurately (remember the signal includes a very very precise time as a part of the data packet) and can be used for incredibly accurate measurements. Our system was designed for sub inch accuracy with no outside augmentation. The benefit of this is that as long as you have correlated data, you can process it after the fact and improve the position data with the multiple datasets each correcting the others… it doesn’t have to be done in real time, I guess is what I’m saying, which vastly increases the accuracy of the data. (Hindsight is 20/20.)

Bill

Thanks Bill!

Unfortunately for my application, latency is the devil. I need absolutely as fast of a response as possible. Being able to post-process and improve accuracy after the fact is no bueno. I’m impressed you were able to get the accurace down to sub inch! That’s really impressive.

-D