Can you display the acquired component of the acceleration a a function of time.What about the off set values and their change with time? what is the sampling frequency?

What is the integration method you used? What i the maximum frequency of your vibration? Your sampling frequency must be greater than 2x the maximum frequency.

You have to verify at first that you acquire your signal correctly. Then you must verify that you perform the integration correctly. If these two things are okay you will get correct results

Wish you success.

Dear Marcelo

These devices are very interesting as the accelerometer + gyroscope module can also be used in solar tracking and sun following systems. You have probably seen this very detailed description: 

http://www.cs.unca.edu/~bruce/Fall13/360/IMU_Wk8.pptx

The datasheets on these links will also help

https://www.sparkfun.com/products/10937

https://www.sparkfun.com/products/11028

Dear Marcelo

As Prof. Abdelhalim Zekry said, you have to take the offset into consideration. 

My question is Have you got acceleration values (by converting accelerometer output to acceleration in ms-2) in a right manner? If you are not sure, record it in a microSD card using arduino and view it in a graphical chart (MS Excel, LabVIEW, MATLAB). Then compute velocity by integrating the acceleration values. Here you will see the effects of offset and random noises in the measurement. 

(Before that) As you know, MPU-6050 is an inertial module that has a 16-bit ADC for data conversion. It generates 16-bit signed integer data. Hence, you have to perform signed addition in integration routines. Hope it helps. 

Well, in theory double integration method will give you what you need but in practice double integration accumulates the errors in the measurement as you go so the results are never accurate or reliable to use for distance measurement.

Now in the case of using another sensor in addition to the original sensor this can be used with some sort of filters to reject the measurement errors periodically so it don't accumulate, it is still far from perfect and isn't as good as the worst distance measurement tools but it will do a better job than just double integrating the acceleration of a single accelerometer.

Check this:

http://cache.freescale.com/files/sensors/doc/app_note/AN3397.pdf?fsrch=1&sr=2

Also if you are considering something like a model rocket with large acceleration in one direction this can actually work since a lot of the error sources in such situation are negligible compared to the actual acceleration and distance so it really depends on the situation and the accuracy required.

Dear Marcelo,

Try double integration, but filter the low frequency drift in your data first. Functions like 'filtfilt' in Matlab help you to do this, and avoid phase changes. 

Thank you all for your time. I'll take a look in each suggestion and see if I can solve this issue.

Best regards,

Marcelo Lacerda

If anyone else comes across this posting - here are some useful ressources:

http://www.i2cdevlib.com/forums/topic/312-velocity-from-acceleration

http://www.i2cdevlib.com/forums/topic/312-velocity-from-acceleration

Hi Marcelo, 

Have you solved the above problem? I am interested about this because I am working on a similar project.  If you found any possible solution then please share. 

Thanks

I don't know if it helps in any way, but I know a work where the authors use a simple and noisy accelerometer (no gyro) to detect directional bias for a four-legged robot:

http://cdn.intechopen.com/pdfs-wm/10095.pdf

If you look for linear displacement, I highly doubt that mems accelerometer can help as the range of their measurements is limited and I do not know if the measured acceleration maps with real world acceleration or it is just a relative acceleration.

But if you look for rotation, like sun following system, it perfectly works and you should read about the concept of spherical rotation and how gravity line varies in a rotating object.

It requires a bit of post-processing, but once you acquire the accelerometer data, I recommend integrated it using the trapezoidal rule to approximate velocity. From there, you can use several methods to calculate the displacement from there. It's not impossible, and it is prone to error; however, I did something similar and wrote about it on my blog. I used the ADXL345 and Arduino and it heeded great results (I approximated the forces and distance traveled by an elevator).

If you're interested, check it out:

https://engineersportal.com/blog/2017/9/25/accelerometer-on-an-elevator

Good day I would like to ask how am I able to obtain velocity and displacement from an array of data of accelerometer?

hi,

a simple way is Seismosignal. if you have an academic email, easily can receive a license from the Seismo Softawares Company. the software read the acceleration record and give you the velocity and the displacement history.

be happy

Aziz Hosseinnezhad can that displacement and velocity data be used for control purposes ? For instance, I want to use matlab for control design and I would be reading the data from the accelerometer (which has a matlab toolkit) into Matlab, but I also need the velocity and displacement (3D).

Any suggestions/ideas would be appreciated.

Thanks.

Dear, Daniel,

sure, the software give you the velocity and displacement time history, but if you are using PEER acceleration time history which normally arranged in multiple column format, then you need to study the instruction of the software that guide you how to set the parameters to recall the acceleration time history correctly.

Aziz Hosseinnezhad thanks so much for your ideas.

I will go to the seismosignal website and see what I can get.

Thank you.

You might also find the paper by Romain Delpoux , titled "Acceleration feedback via an algebraic state estimation method". It is the state estimation method that would be relevant to your question. The paper is on ResearchGate at:

Conference Paper Acceleration Feedback via an algebraic state estimation method

First of all you need to calculate the acceleration angle from the Rx Ry and Rz individual terms. so the pitch will be

theta (pitch)=atan2(Rx/sqrt(Ry^2+Rz^2)

theta (roll)= atan2(Ry/sqrt(Rx^2+Rz^2)

so you now have the angles but they will be noisy and you need to add the gyro info. You can do this via a Kalman filter algorithm. Email me for details

maybe it's helpful

There is quite a decent work by Sebastian Madgwick here.. please refer

https://x-io.co.uk/open-source-imu-and-ahrs-algorithms/

I have the time history acceleration data with me. I got the velocity data from the traditional integration method - Trapezoidal. Again I integrated the velocity signal to get displacement. I got displacement data in 10E-12.

My doubt is that, do we get such low displacement ? I cross checked with seismosignal too. Got the same values of velocity but not the displacements. Are these derived displacement values correct ? Please help me.

The vibration signal acquired from an accelerometer usually gives in mV or pC. Based on the sensitivity of the accelerometers (for instance say 100mV/g is sensitivity) one should convert the signal into pure form of acceleration. Then there are different methods to convert acceleration into displacement.

The methods I know are 1. Integration only methods, 2. Use of high pass filter and perform integration and 3. Use of DATS Omega arthemetic integration.

In my point of view method 1 is not correct as the raw vibration signal itself contains noise.

Where as method 2 and 3 gives almost similar results. However, method 3 (Omega) is effective as it integrates the signal in frequency domain not in time domain and it also removes the DC content too. Then convert the signal into time domain once after integration is over.