Determine acceleretion sensitivity for accelerometer

Hello,

I am trying to determine the sensitivity of an accelerometer based on recordings from another accelerometer. I have previously been making recordings with an old wired accelerometer, but am now wanting to go wireless and want to ensure then the new accelerometer is good enough to replace the old.

The new accelerometer is an ADXL354 (analog) from Analog Devices (ADXL354 datasheet), coupled to a Shimmer IMU device (Shimmer IMU spec sheet) via the Shimmer PROTO3 mini expansion board (PROTO3 spec sheet). I do not know the specs of the old accelerometer.

I have conducted a shaker test with the ADXL354 and the old accelerometer. I keep the amplitude of the shaker constant at an arbitary value, and sweep in frequencies like 20-300-500-800-1000 Hz, then turn up the amp a little an sweep back down. (fig 1) The sample frequency of the old accelerometer is 5000Hz, while the ADXL354 is limited to 2048Hz by the Shimmer IMU. I know the old accelerometer outputs in volts, and if I understand correctly the ADXL354 outputs in mV as per the datasheet specifying the ADXL354's (typical) sensitivity as 400 mV/g?

The x-axis are not aligned as I cannot start the recordings for both accelerometers simultaneously, but I have panned the recordings to match in time.

My problem occur when I want to convert the accelerometer output from V to g. I have been given a factor (0.0176*1.1) to multiply the old accelerometer recording with to convert its output to g. I then want to convert the ADXL354 output to g and I divide by 400 as per its sensitivity. However, when I do this the acceleration for each accelerometer does not match. (fig 2)

Summed up my question is; how can I correctly match the measured acceleration of a new accelerometer to an old one to ensure that the new accelerometer is sensitive/good enough to make recordings of equal quality as the old accelerometer.

I have attached the data I have recorded for the two accelerometers and the script I use for plotting if anyone is interested.

Edit: ops, at first try I didn't get the old acc data attached as is was larger than 5MB. I have now divided it into two tables which should be merged when running the script.

[Mathieu NOE]

hello

your time plots are a bit confusing. If you ammply always the same input signal on your shaker, I'd would expect both accels signals to be similar , only difference would be output amplitude in regards to different sensivity.

is the old accel still available or do you mean you have only the recording left for comparison purpose ?

you can acquire only one channel ? or 2 channels ?

if you have 2 channels , why not oberlay the two traces at each frequency, record their amplitude and do the ratio (knowing the old accel sensivity it's pretty easy).

if you have only one channel, you have to do it sequentially, but do not change anything regarding test signal amplitude or shaker amplifier gain !

[Dams]

Hello Mathieu NOE,

I have applied the same acceleration to both accelerometers since I had both accelerometers mounted on the shaker at the same time. The reason the time axes do not match is because I cannot start the recordings simultaneously. I agree with you that the signals should be very similar and only differ in the amplitude as the accelerometers have different sensitivity. From what I understand the old acc has higher sensitivity as the range for its output is -4000 to 4000 mV (at its' highest), while the ADXL354 is only -200 to 200 (at its' highest) when experiencing the same amount of acceleration from the shaker.

I still have access to the old accelerometer and can perform additional tests with both the old acc and the ADXL354.

I do not understand what you mean by acquiring one or two channels. I am only interested in movement in one direction (up and down) so I only need one axis, eg. one channel from the accelerometers.

[Mathieu NOE]

hello Dams

why bother with multiple frequency tests ? are you triying to compute the sensivity of your new accel vs frequency ?

if not , you can test it with only one sinus (industrial accels sensivity are given at only one specific frequnecy (100 or 1000 Hz)

at least, that would give you a plot with only one amplitude and not these bizarre plots where the sahpe are evn not the same (keeps me thinking there is something wrong in your test procedure)

if you know the old accel sensivity (let say 3 V/g) and the new accel gives 200 mV amplitude (when the old one gives 4000 mV , then the new accel sensivity is 3*200/4000 = 0.15 V/g (so the inverse is engineering factor = 1/sensivity = 6.7 g/V)

hope it helps

[Dams]

Hello Mathieu NOE,

Thank you for your answers.

I am trying to determine if the sensitivity of the new acc (ADXL354) is equally good as the old, also when at different frequencies.

I have made a new test where I keep the frequency constant and only increase amplitude, as I also wish to know how the ADXL354 holds up against the old at different amplitudes. The frequency I have set constant at 300Hz, since the datasheet for the ADXL354 does not mention any specific frequency for the sensitivity.

After increasing the amplitude a couple of steps the old acc saturates at -4000/4000 mV. The ADXL354 saturates only later at around -500/500 mV. When converted to acceleration I get the following plots:

I do not know the sensitivity of the old acc but I do for the ADXL354 which is 400 mV/g. Following your suggestion I should be able to find the sensitivity for the old acc. Thus I find at two different amplitudes (same frequency) the old acc to give 3697 mV when the ADXL354 gives 85 mV. Then 400*3697/85 = 17398 mV/g = 17.398 V/g which inversed is 1/17.398 = 0.0575 g/V.

I do the same but at a different amplitude (still same frequency) and find the old acc gives 575 mV while ADXL354 gives 25.55 mV. Then 400*575/25.55 = 9002 mV/g = 9V/g, inversed is 1/9 = 0.1111 g/V.

I understand that these two values should be equal since the sensitivity of the accelerometers should remain the same across these tested amplitudes and the set frequency, but they are not, and I do not understand why.

Next, I will make a new test where I set the frequency to only 100 or 1000Hz to keep it simple, following your suggestion.

[Mathieu NOE]

hello

the ADXL specs says the bandwith is 0 to 1500 Hz. The mems accels do measure static (gravity), so make sure you have not issue by reading static + dynamic data on the ADXL plot and compare with (for example) a piezo accel that have typically bandwith between 5 Hz to 5 kHz at least (NB here NO dc value can be measured)

so depending of accel technology (mems / piezo / voice coil) look at static performance also if your acquisition is DC or AC coupled

[Sylvain]

Hi,

I am confused with your experimental methodology, as you have no measurement reference, and you did not precise if the ramp up / ramp down of the amplifier is repeatible. Therefore you may want to mount both accelerometers on the same shaker and record the transfer function between the two (looking at your graph, I can see you did the shaking tests for for different time period).

Hopefully you will get a flat response corresponding to a sensitivity difference.

Another aspect to consider is the background noise. record your signals in absence of motion, and plot both time signals and spectrum.

[Dams]

Hello,

I am sorry if I have not been clear on the matter but every test have been conducted with both accelerometers mounted on the same shaker at the same time. The reason for the difference in registred recording times is because I cannot start the recording for the accelerometers at the same time since they run from two different programs (LabScribe and Shimmer Consensys).

I have conducted a third test using a different 'old' accelerometer for comparison. This accelerometer is static (DC) and I can now correctly compare the two accelerometers. The problem was as you suggested, Mathieu NOE, that I was trying to compare an AC to a DC accelerometer measurement.

Thank you for you help and time.

Edit: Mathieu NOE, I wold like to accept your answer as the answer to my question but I cannot since you have only commented on my question. Would you copy your comment from November 3rd and post it as an answer so I can accept it?

[Mathieu NOE]

hello

will do !

hope you get through the testing activities !

all the best

Answer 1

[Mathieu NOE]

hello

the ADXL specs says the bandwith is 0 to 1500 Hz. The mems accels do measure static (gravity), so make sure you have not issue by reading static + dynamic data on the ADXL plot and compare with (for example) a piezo accel that have typically bandwith between 5 Hz to 5 kHz at least (NB here NO dc value can be measured)

so depending of accel technology (mems / piezo / voice coil) look at static performance also if your acquisition is DC or AC coupled