dinamically reading load cell

[altelec]

In our application we weigh fruit. We are using a conveyor scale, moving over a chained plastic tray, at rates of 1 to 8-10 units / second. Each plastic tray have two contact point, traveling over load cell surface.
An optical sensor send me a pulse when a tray is in the middle of a load cell. I�m trying to read each fruit weight synchronously with the pulse, but I receive a raw weigh plus noise (mechanical vibration and rumble) and a damped oscillation due to many little shocks over load cell�s cantilever.
I�m using a PCI-6013, 16 bit DAQ. I�ve been logged raw load cell signal and compare it with averaged samples ( 300 @ 3000 Samples/sec), low pass Chebyshev or Butterworth filters, and Basic Averaged DC -RMS Vis� outputs.
All p
rocessed signal look alike.
I need info about a best signal processing and tricks, to get a precise read without decreasing the maximum machine speed.
Thank You,

Alberto Tubía
ALTELEC Ing. Electr.

[Conseils]

Looks like this might get lost, so I hope to help get it going....

I would imagine that the most difficult part would be getting the mechanical engineering / handling correct.

1) Is this your own mechanics ?
2) How are you managing damping of the system?
i.e. When the tray 'hits' the load cell.
3) Are you sure it's not resonant / resonating ?
Have you tested for resonance ?
See damping.
4) Can you correlate actuals with data at lower speeds ?
Post signal data as an example.

Suggestions
===========
I would suggest that you post a picture (It's worth a thousand words) or two and some sample data, then you might get a fuller answer.

[altelec]

1] Not, it�s an existing, old mechanics.
2] We are trying to enhanced it, with following procedures:
Mounting load cells on an isolated platform.
Smoothing rails to avoid try hits, accessing or exiting the load cell zone.
Improving all the whole mechanical trail system. ( gears, chains, sliding)
3] Of course the load cell cantilever ( with a Teflon rail coverture), has a natural damped resonance. It hasn�t any additional mechanical or fluidic damping devices.
4] I�m attaching an improvised LV7.0 Vi with approximately 10Ksamples, with
Strain gauge is single pole, low pass filtered at 1K6 Hz and 100x gain ( hardware).
I�m weighting 73 grams plastic trays
Y scale is in grams, with offset correction, and a gain=13000 (software)
Du
e the lost of some trays, graphics shown sometimes, an unevenly Sync Pulse.
Maximum Sync pulse, is near 6Hz.

Than you very much,

[Conseils]

My initial reaction to looking at the data is

1) It is undersampled
2) Why are the synce pulses different amplitudes
3) The raw data contains little or no information

It appears, although not clear from your comment, that there was an existing system which weighed the trays. If this is the case what I would do is connect in parallel and record the data.

I would send say 100 sets of trays through with known weights in. Say 10 with minimal and 10 with maximal and so on. Then keep a note of what you sent and when.

Now you have a representative data set to create a logged file at say 44100Hz sampling rate and use that as a reference or perhaps use the WAV recorder to create a simple raw record. Most importantly it has no processing
at this stage.

You can then use this to refine your processing technique. Later in the projects life it acts as a system validation check. Some of the ones I have created are the most important parts of the projects I have worked on. From damage limitation (that's what the customer was actually doing when I was called in). To, 'I can prove you have changed something'. To 'Ooooops missed that!'

In summary
==========
I would suggest at this point you don't appear to have valid data. Wether that's a result of hardware or undersampling is impossible to say with the amount of information here.

Put some 'calibration' weights on and make sure the static levels are what you expect from 0 to typical. Then move to dynamic at increasing speeds.

You will no doubt need a calibration section in the design anyway!

[Conseils]

How did you progress ?

[altelec]

1)   Yes, you are right. It isn�t undersampled really, but reviewing the block diagram in my improvised Write Loggin VI , I have found that the �raw� signal was averaged too (300 samples) and I recorded it as is. This is an answer for your item 3} too.
2) Sync pulses have different amplitudes because I�m reading digital Sync pulses with the PCI-6013 A/D converter and I have some �hum� between DGND and AGND. Because I�m only interested in Time info, I don�t care in amplitude modulations at this time

Analyzing LOAD CELL LOGGIN graph, some averaged readings at Sync pulse windows, show �minor� errors around real 73 grams trays, say + / - 10 gm, and many shows �gross� errors around -30/40 gm. At first case I suppose a mix o
f rumble and trays hits, masking reads . Looking at second case, I saw some trays �hanged� on the tracking rod and �robbing� hence 30-40 gm.

Now a technician, is fixing this kind of mechanical issues ( I hope), and engine isn�t operative now. I think that in a couple of weeks, I will takes a �real� raw signal and will send it to you.
Before now, I tested the load cell weighting plastics tray only, but from now on, following votre conseil I will carry groups of 10 trays with known weights.

When I take a static reading I get errors about +- 0,3gm from 0 to 500gm.

I appreciate very much your help.

Thanks a lot.

[Conseils]

I could only work from the data you supplied, so I looked at the graph. What you post, more often than not, determines who and how good the answer is.

I still think that the damping of your measurement system will be critical to effective and accurate measurement as well.

Check out the following
http://www.tedea-huntleigh.com/
They supply accros the world. I think you should get the actual weigh system from them or similar.

If the assistance if usefull, then you could help others by rating the answers.

Good luck.