Handheld meter robustness testing

[joeqsmith]

Well, again I fully admit I was biased against Fluke from the start.  Damn you Fluke for making such robust products....

I think we'll only be sad if your Fluke 189 has displaced your BM869s as your go-to meter on the bench. For me, mainly because the 189 was discontinued shortly before I wanted to get one. So damn you Fluke for discontinuing the best meter you ever made...

The Fluke is a very nice meter considering it's age.  It actually does get some use on the bench.  The Brymen BM869s is still king as far as I am concerned.  It's just an all around good meter for electronics.  One day I am sure I will find something I like better.

[joeqsmith]

There is some confusion about the data I show.  Lets start with the four data sets zoomed from 0 to 1 ohm for all 50,000 cycles.  These are absolute values.  The Centech meter drops to 0 at 17,000 where I abort the test. 

[joeqsmith]

Now let's look at the histograms for all four data sets again looking at the same 0 to 1 ohm span.   Let's start with a 1000 bins.    On the far left, the Centech blows off the chart at 0 because I aborted the test.   It seems the confusion is how the Fluke is such a small peak compared with the other data sets.  As I have said, that Flukes standard deviation is very low compared with the other meters.  For me it looks normal but I can fully understand the confusion. 

[joeqsmith]

So let's increase the number of bins from 1000 to 2000.  Notice I have changed the vertical scale of the graph as the peaks are now shorter and wider.  Also notice how the ratio of the peak heights have changed.

[joeqsmith]

Let's go to 4000 bins.  Again I have rescaled the vertical axis.  The peak height of the Kasuntest and Fluke are now very close with one another.  But you can clearly see the Kasuntest is a much wider peak. 

[joeqsmith]

Let go even further to 8000 bins.  Now we see the peak height for the Kasuntest is much lower than the Fluke.   I have zoomed into the horizontal axis to look at the data from 0.3 to 0.6 ohms to provide you with a better idea of what the peak shapes look like.  Not the Centech is there but the resistance is so high there is not much data in this area. 

So to be clear the bin size effects every data set. 

[max666]

The problem I'm still having with the Histograms is that the sum of all the bins, or if you will the area under the curve should be equal the number of samples. So for the Fluke it should be 50 000 also for the Kasuntest and the Baseline, only the CenTech should have a lower total. Since the Fluke has such a low standard deviation it's peak should be much higher to get to the same total of 50 000. Maybe my eyes are deceiving me, and it's just me.

[joeqsmith]

There is also some confusion about the log/log graph.   Again, this graph shows how many cycles until the switches resistance changes by more than a given amount.  The graph shows only the data from 0.005 ohms to 500 ohms for the 50,000 cycles.   So in the case of the Fluke, even with 5mOhms the switch never changed that much in the 50,000 cycles.  This is why it's a flat line.   Now the person asking feels there should be a vertical line going up to the 50,000 but because I am not looking for changes less than 5mOhms, we never see this.   Basically this was because I didn't care about changes this small.  I actually started out just looking at 50mOhms but thought may people would want to see what was happening down in the muck.  The Centech and nameless POS meter both reached 20mOhms right from the start. 

[max666]

And for the Log-Log, yes primary issue for me is that you start plotting from 0.005 Ohm, where as that is already the maximum for the Fluke.
But also I wouldn't continue drawing the horizontal line (at least for the Fluke) after 0.005 Ohm, since the Fluke didn't fail at 50 000 Cycles and didn't actually show a change in resistance of 500 Ohm at 50 000 Cycles.

[joeqsmith]

The problem I'm still having with the Histograms is that the sum of all the bins, or if you will the area under the curve should be equal the number of samples. So for the Fluke it should be 50 000 also for the Kasuntest and the Baseline, only the CenTech should have a lower total. Since the Fluke has such a low standard deviation it's peak should be much higher to get to the same total of 50 000. Maybe my eyes are deceiving me, and it's just me.

No problem.  I want to try and help you understand what you are looking at.  The screen itself has a limited resolution as well.  If we continue to push it so the Fluke filled the screen you would see that those bins are pretty narrow and they all have a lot of data in them.  This will not be the case with the others.  However, part of the problem here may be that you are not actually seeing the bins.  I am drawing a line connecting the peak of each bin which can certainly give a different impression.

[max666]

The problem I'm still having with the Histograms is that the sum of all the bins, or if you will the area under the curve should be equal the number of samples. So for the Fluke it should be 50 000 also for the Kasuntest and the Baseline, only the CenTech should have a lower total. Since the Fluke has such a low standard deviation it's peak should be much higher to get to the same total of 50 000. Maybe my eyes are deceiving me, and it's just me.

No problem.  I want to try and help you understand what you are looking at.  The screen itself has a limited resolution as well.  If we continue to push it so the Fluke filled the screen you would see that those bins are pretty narrow and they all have a lot of data in them.  This will not be the case with the others.  However, part of the problem here may be that you are not actually seeing the bins.  I am drawing a line connecting the peak of each bin which can certainly give a different impression.

That's probably it!
Thank you for being so patient with me, Joe.

[joeqsmith]

And for the Log-Log, yes primary issue for me is that you start plotting from 0.005 Ohm, where as that is already the maximum for the Fluke.
But also I wouldn't continue drawing the horizontal line (at least for the Fluke) after 0.005 Ohm, since the Fluke didn't fail at 50 000 Cycles and didn't actually show a change in resistance of 500 Ohm at 50 000 Cycles.

I figured this was the problem.  So for fun let's push things down even lower into the muck by a factor of 10.   Now we can clearly see there is a limit that we detect some change with the Fluke.   

Now it becomes a question of how good is good enough.  Someone had made the point earlier about how it's a high impedance input and tens of ohms may not be much of a problem except when you start looking at the current input.   I really am not sure that I want 10s of ohms for the power selection as well.   

Anyway, hope some of this helped clear things up.  I can certainly understand that how I view the data may not be the way that you or others would want to see it formatted. 

I did see the post about show them as a percentage which I just plain did not get where they were coming from.  I did respond to them the best I could.  It is difficult at times to know what the person is asking and if I am even addressing their concerns.     

As far as I understand is that you're using the absolute resistance value? If so, than you're not comparing apples with apples. You need to compare the change in value relative to the original start resistance, since the resistance of the tracks/circuits etc of the different meters isn't identical. In software this can easily being done by reading the first value and divide the other readings by the value to get the correct relative value. So this will give you a ratio (or percentage) how much off the meter is after x-amount of cycles.

You understand correct.  While we could plot the resistance as a percentage change,  I don't feel it is something I "need" to do nor do I see any value in displaying the data this way as it tells me nothing.  It could also make a very poor performance switch appear better than it is.

Let's take a simple example using two switches A&B.   Switch A has a DCR of 0.001 ohms.   Switch B has a DCR of 10 ohm.   Switch A say reaches a peak resistance of 0.010 ohms,  while switch B reaches a peak of 20 ohms.   Switch A has an increase of 10X and switch B, 2X.   Looking at a percentage would present Switch B as the better of the two. 

You are also correct in there there are other errors outside the switch contacts.  For the most part, these will remain a constant throughout the test for a given meter.  They make up a small portion of the total resistance.  We are basically talking about the trace lengths from the pads to where the wires are attached, the solder joint and the wires going back to the external meter.   These wires for the most part are a constant from meter to meter.  The same for the solder joint.

[joeqsmith]

I wanted to see of there was by some chance of luck that the Kasuntest would have outperformed the Fluke down in the noise floor.  Looking at the log/log from 0.00005 ohms to 500.   Really pointless with the test setup I am using but here it is anyway.

[max666]

...
Anyway, hope some of this helped clear things up.  I can certainly understand that how I view the data may not be the way that you or others would want to see it formatted. 
...

It certainly helped. And I don't expect this to be a musical request programme, but just to exhaust one more possibilities on how to present the data 

[joeqsmith]

Attached is the picture I show in the last video of one of the ZT102 contacts.  The tip of the wiper had wore through and we ended up with a hollow sort of square.   

[joeqsmith]

I have an old 60's Olympus microscope that I made a custom mount for and it is set on a machinist base.  A real nightmare to look at.  It's decent for working on small circuit boards with a 10-40X zoom.    Shown looking at a 0.01mm/div alignment slide.

[joeqsmith]

I had some lenses as decided to see if I could get some better pictures of the ZT102's contacts.   Tp sort out the focal point, prototyped it with cardboard, paper and masking tape.   

Here are a couple of shots using different lenses.   

[joeqsmith]

It's way too touchy and sensitive to the depth to get a good picture.  But you can clearly see the holes going through the contact.     

[joeqsmith]

Here is with the contact removed and then backlit with and without the top illuminator.   

[floobydust]

It's the rotary switch wipers wearing out? I thought the PCB copper would wear out first, it's surely 1oz (1.4mil) or less.
I guess for low contact resistance, they (cheap chinese DMMs) apply a lot of force - instead of using precision multi-fingers.

[floobydust]

For the ZT-102 it looks like the FR-4 wear generates abrasive (fibreglass dust) and that pretty much quickly kills all metal in the rotary switch.

Many potentiometers, they use three fingers. So you have extra width on the track and 1/3 the contact force over the same contact area. I'm not sure of the relationship between resistance and contact force. I suppose they could do this for multimeters, use multiple fingers and lighter contact pressure. pic from http://www.ladyada.net/wiki/tutorials/learn/arduino/leds.html

Wavetek/Beckman multimeters (HD110, TECH300, 3030 etc.) don't use a wiper, the knob is a rotary cam pressing on many (16!) stationary switches. pic from http://mrmodemhead.com/blog/beckman-industrial-hd110-multimeter
These were not super reliable, the switches got a bit bent up somehow under normal use, and not all had gold plating. I've repaired several bending the arms to align them.

[HalFET]

Might not know much, but that I do. You'd have to stick it in a SEM to see it in detail. But basically you have two rough moon landscapes pushing down on each other. Usually the metal spring should be smoother than the PCB (Electroless nickel is quite porous), but the nickel in the PCB is quite a bit harder. So what you'd expect is that the surface roughness of the pcb is imparted on the contact wiper, however this is where contact force comes into play. Pushing down harder will decrease contact resistance by increasing mutual surface area (by deforming the spring preferably). As you move it around the interlocked sections have a chance of grinding away. So wider contacts with less force will indeed last longer, additionally wider pcb pads less likely to shear off.

What you'd also expect as a result is that high end meters would have better (smoother) plating on the contacts and use the soldermask to assist in holding down the pads on the PCB.

[Fungus]

It's the rotary switch wipers wearing out? I thought the PCB copper would wear out first, it's surely 1oz (1.4mil) or less.

Did you ever ask yourself why the bar on an electric train's overhead contact system doesn't wear out in a couple of days?



It's the same reason the rotary switch contact does.

[HalFET]

Actually, since I am horribly bored in the cleanroom waiting for the vacuum press and laser to do their job, I'll disassemble a DT-830 and put the PCB under the profilometer 

[stj]

It's the rotary switch wipers wearing out? I thought the PCB copper would wear out first, it's surely 1oz (1.4mil) or less.

Did you ever ask yourself why the bar on an electric train's overhead contact system doesn't wear out in a couple of days?



It's the same reason the rotary switch contact does.

but the overhead cables have to be checked and regularly replaced!