noise accuracy dual axis accelerometer

[heang]

Hi

I am doing calibration on accelerometer of analog providing dual axis acceleration to measure tilt angle of sensor.

Based on spec sheet from analog, the sensor should be in linear equation V=Ka+B

V=Voltage output

K=Linear sensitivity

a=acceleration of axis

B=Bias voltage.

However, i turn the sensor with known angle, i get the voltage. Plotting the graph of Voltage versus acceleration, i am not able to get linear curve of V=Ka +B, but nonlinear graph.

How can i determine the non linear error and resolution from the newly plotted graph in labview?i have collected data. and i wish to do regression analysis. How i can do it in labview?

After getting the nonlinear regression, how can i know my new model (new equation) has improved error and resolution?

Can someone ple help

your kindness is Much appreciated

Regards

[tbob]

If the spec sheet says that the accelerometer is linear, and you are reading non-linear, I would say that your measurements are faulty.  Check your setup.  Are you using shielded wiring for connecting the accelerometer output to your voltmeter?  Using short wires too?  Is your voltmeter sensitive enough to read the low voltage outputs?  What is the resolution?  What about the supply voltage to supply power to the accelerometer?  Is it steady with no ripple?  Are you moving in one axis only?  Don't try to correct your setup errors by defining a new non-linear formula.  Instead, fix your setup.  It is also possible that you have a faulty accelerometer.

[heang]

Actually i am measuring dual axis acceleration.

The accelerometer has 2 voltage output.

Vx is output of Ax

Vy is output of Ay

If the accelerometer is moved in plane, it has linear function which obeys the spec sheet.Moving direction of x, Ax will change. but Ay no change, n vice versa.

But once it is tilted, Ax and Ay change simultaneously.Tilted angle is defined as Arctan of (Ax/Ay).

It has cross effect of each axis acceleration, meaning that Ax is influenced by Ay.

The resolution of sensor 0.03 degree.Thanks for your prompt reply.

Regards

[tbob · Accepted solution]

OK, this description makes sense.  Probably nothing wrong with your setup.  So you want to create an equation using the data points that you have aquired when tilting in both X and Y directions?  Does the Arctan function not serve the purpose?  Divide all Ax values by all Ay values and feed the array into the ATAN function.  I've used three linear points in the following example and the output is linear.

[heang]

Thanks

Theoretically angle =arc tan (ax/ay) is correct.

 i have sets of data showing the nonlinearity of Voltage output Vx which is influenced by Vy.

under the effect of Vy, Vx =Kgsin (angle)+B is no longer valid.

g sin angle= Ax

B= bias voltage

with effect of Vy,can i make assumption that  Vx=Kgsin (angle)+ B+m1*Vy^2+m2*Vy+P?

m1,m2,P=coefficient obtained from nonlinear regression

My assumption of Vy is quadratic function.

If my assumption is valid, how can i get the improved error on my new model?

Also, how can i determine the noise/resolution if i am only provided data which shows sinusoidal function, with voltage output and acceleration?It is static calibration.

Regards

[tbob]

I would think that a good accelerometer would not have one axis affect the other.  However, I cannot answer your question.  My experience with math is long gone.  Hopefully someone else can chime in here.  Maybe you could plot the measured outputs versus the theoretical outputs, take the differences and plot an error curve based on the results.