Dealing with Inaccurate System Errors

[babistopher]

Hi all. I have recently taken on a project that involves a device that is instrumented with strain gauges (think of it as a cantilever tube with strain gauges at one end...the instrument details aren't too important for the question). Now, this instrument has SO many errors that I do not know how to deal with it without redesigning the device from scratch. The professor I am working with seems keen to keep the design as is and keeps on telling me that the errors can be accounted for. This is my problem. How can errors like hysteresis, a lack of repeatable zeroes, random gauge sensitivity issues, and literally zero repeatability be accounted for to give meaningful results?

I wanted to see if anyone could guide me a little before talking to the professor as I am relatively inexperienced and want to know a little better about this before I go and talk to him. I am reading up on experimental errors now but nothing seems to help when it comes to this flawed instrument. Thanks in advance.

 

[Danger]

It sounds peculiar to me; I'd think that weeding out inaccuracies in the equipment would be a better approach than trying to compensate for them. Maybe some kind of 'fuzzy logic' circuit?

 

[babistopher]

Hey Danger. What do you mean by weeding out inaccuracies? Could you explain a little bit more? Do you mean identifying them and then documenting them?

 

[Danger]

I think that the randomness that you mentioned rules out documenting them. What I meant was that I agree with redesigning the thing to minimize error at the detection stage.
As for the 'fuzzy' bit, I really don't know much about it, but it seems to work well for stuff like image stabilization in cameras. My very off-the-top-of-my-head thought was that maybe your logic circuits could introduce their own random errors in each of the areas of problem. Due to the double randomness, I figure that you would get some cancellation, which would be your greatest accuracy. Conversely, reinforcement might make the error so large that you can easily identify it and ignore it.

 

[FredGarvin]

I don't see how you can get any kind of accuracy if you can't even get a repeating zero point. I can understand plain old brute force calibration, but to do that you have to have repeatability of some kind.

I would agree that your gut instinct is the correct one.

The question is, how does one introduce that many errors? Did they put the strain gauges in upside down?

 

[stewartcs]

If you have zero repeatability you can't calibrate it. Sounds like something else is wrong. Even if the design was bad, you would probably get repeatable results...although incorrect ones.

Not sure why your professor seems to think you can correct this. It's hard to correct something that is not repeating.

 

[babistopher]

I wish I knew where all of these errors were coming from. There are actually two instruments that have identical designs but I only have had time to take a look at one. I am certain that looking at the second one will help identify, or at least clarify, some of the reasons behind the issues. And hey, maybe they did put them on backwards for that one instrument. Although I highly doubt they would make such a mistake, I will take a closer look tomorrow. And Fred, the fact that there is 0 repeatability was the major source of bafflement when it came to my professor's insistence of simply moving forward with the design.
Also, thanks for that fuzzy logic idea. I will keep it in mind.

 

[babistopher]

Thanks Stewart. Simply getting input on how odd this is reassures me that it's not my lack of experience that is preventing me from finding a solution...or at least cannot be blamed completely for it.