Performing a GRR when you get multiple readings over time per part.

[E_Joel]

Hi all,

I'm new here and decided to register because I have an issue with this Gage R&R I need to perform. Hopefully someone could shed some light or offer their expertise.

I'm doing an RR on a load cell that is mounted on an automated machine that measures force to slide.

The way this machine collects data is by what the engineer described as "snapshots." Basically, at a set interval of time(unknown) as the machine pulls the sliding part, it takes a force reading at a specific servo position.

So, when the data is imported into excel it basically shows the numbered "snapshot" (there are about 137 of them) and at each snapshot it gives a force and a servo position reading.

For this study I'm really only doing the analysis on the load cell itself. I'm having a hard time trying to come up with a way to group the data in a useful way. I don't think minitab will let me put multiple readings per part.

Can anyone shed some light on how I might be able to perform this? I'd appreciate any input!

 

[Miner]

I am having a difficult time visualizing your measurement process. Can you attach your data? You may have to make a total of 5 posts before you can attach a file since you are new.

 

[John Predmore]

For this study I'm really only doing the analysis on the load cell itself

I understand resistance to sliding may be jerky, the forces not static and uniform. The same weight slid ten times may not generate the same force profile. Does that variability reflect a deficiency in the load cell? How to validate a measurement device with dynamic force events which are likely not repeatable? I want to offer a new approach. I realize this is not the question you asked.

You describe how the automatic machine works when it is working automatically, including the trigger timing which you stated was unknown. Can the load cell be tested in situ as a standalone measuring device? In simple terms, a load cell measures force. The load cell does not know whether the force is frictional resistance to sliding or a weight hanging on a string. Hang various known weights attached to the load cell and record what the output reads. That will give you a baseline GR&R where the calculations are straightforward.

Obviously, I am making simplifying assumptions. I have not seen your apparatus. I am only offering an idea. You and your engineers have to decide whether this approach might be suitable for your objective and what the tradeoffs are.

Then, once you have quantified the load cell measures static weight as a repeatable linear function, then add complicating factors under controlled conditions, such as kinetic inertia, hysteresis, and response time to a step function in force. See how the ideal function degrades with the addition of each simulated complicating factor as your simulation gets closer to the actual application function.

 

[Bev D]

It would be great to have the data if we could get it.

I have no trouble visualize it however as I have performed (or managed the performance) of thee types of “2 dimensional” measurements hundreds of times. This is the type of situation where we must first understand the data - by graphing it and looking a tit - before we apply any statistical tests.

What I do (did ) si to plot the force on teh Y axis against the measurement position (or time) on the X axis. Then I measure the same thing again and overlay the same graph on the first one. I will do this for several different ‘things’ that are harder or easier than the first one to get some variation. I will then make ‘small multiples ‘ of the paired measurements and LOOK at them.

Are the two ’traces’ very closer to each other within a thing? or are they very different? This is the first pass of assessing measurement error. Are the differences enough to change the outcome? If there are points that are critical (max value, min value, certain points along the path) then you could use those points for a standard R&R study…

not sure if the OP can visualize this. If not I can post an example with pictures. OR we can get their data somehow?

 

[E_Joel]

FTS Final Pass Excel Export.png

• E_Joel
• Aug 11, 2022

Sample of the data output of FTS machine

Hi all, thank your for your swift responses. I realize I may not have had enough technical language to accurately describe the way the machine outputs data. I've attached a media file containing a snippet from the data sheet the machine put out. I want to also specify when I said the time was "unknown" I meant unknown to me, specifically.

If you look at the picture, the P0x_ designation is the "snapshot" number. The S after the underscore is for "servo position" and the F is for "Force". It gives you an average of the forces per zone as you can see on the data. The machine takes ~100 snapshots per cycle.

The issue I'm running into again, is that there is no single force reading for me to put into a traditional crossed RR in minitab. To be frank, I'm not even sure how I would even begin grouping this data in any meaningful way. I'm going to perform the baseline RR on the load cell itself with calibrated weights. Unfortunately for the purposes of the PPAP this is for, they want to see the RR for the machine (load cell) as it relates to our process, so not just how well the load cell RRs on its own, if that makes any sense.

Again, thank you all for your help.

 

[Bev D]

Can you post the excel file itself and not just a picture?

The approach is to start by plotting a 'trace' or a 'run chart' of the force (Y axis) for each servo position (x axis) for each slide. You would need to slide each 'thing' twice and compare the two repeated traces...

Could you tell us what you are sliding?

I can't really see the PNG to do anything with the data.