Have integrated a 3 axis accelerometer on a microcontroller in order to calculate speed and distance. So far my experimental data is off by a factor of 3. Any ideas of some good algorithms to implement?
Hi,
The intuitive wày is to integrate twice your some example code with the attached link. BUT, the twice integration have a high sensibility to noise and the drift will really high.
So you have to use IMU methods fusion of GPS, gyro and magnetométer to have more accurate method.
https://github.com/clemaitre58/Accelerometer_integration
Hey, that is normal. You have to integrate acceleration data twice, so you also integrate the error. Freescale has published a very good paper on the subject.
https://www.nxp.com/files/sensors/doc/app_note/AN3397.pdf
Best regards,
Götz
I would use IIR filters designed using polynomial regression with an exponentially fading memory. That would give you smoothed acceleration data. A GW indicates, you would need to cascade one integrator to get vel and two integrators to get pos.
For open loop-design, see
https://www.researchgate.net/publication/281103616_Improved_IIR_Low-Pass_Smoothers_and_Differentiators_with_Tunable_Delay
For closed-loop design, see
https://www.researchgate.net/publication/299569076_Optimal_Design_of_Digital_Anticipatory_Servomechanisms_with_Good_Noise_Immunity
Conference Paper Improved IIR Low-Pass Smoothers and Differentiators with Tunable Delay
Conference Paper Optimal Design of Digital Anticipatory Servomechanisms with ...
Hi,
The intuitive wày is to integrate twice your some example code with the attached link. BUT, the twice integration have a high sensibility to noise and the drift will really high.
So you have to use IMU methods fusion of GPS, gyro and magnetométer to have more accurate method.
https://github.com/clemaitre58/Accelerometer_integration
I used an Arduino and ADXL345 to approximate distance traveled by an elevator. As is mentioned above, you have to integrate twice, but there are tricks to smooth the data and ensure accurate results. I wrote an entire article on how I did it on my blog:
https://engineersportal.com/blog/2017/9/25/accelerometer-on-an-elevator
to get velocity and position (distance) from discrete data, one have to integrate the acceleration data twice. first integral gives us velocity, then position. but to have relatively good precision values of the parameters (vel, pos) sampling period (Ts) must be as possible as close to real value of the parameter. in Matlab we can use tic and toc to get it. in this case velocity and distance can be measured relatively real.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Algorithms