And again, in the May 1993 issue of MOTOR Mike Dale writes . . .
This month, Mike takes us into the lab to show how engineers painstakingly design experiments to prove the reliability of a given product or design.
There's a lot to be said for a good challenge. Some months ago in this column, I reported on a product called Stabilant 22, which is promoted as an electrical contact enhancer. I told you that I had used it in several applications to improve or eliminate electrical intermittencies in connectors and potentiometers.I also mentioned that associates had used it successfully in computer applications in which worn or aged connectors were causing loss of bits and bytes of computer data.
Well, I received a note in the mail recently from one reader who wasn't satisfied with these claims. Ron Greim, of Budget Jaguar in Orange, California, wanted to know if there was any scientific evidence proving the stuff really worked. He also wanted to know if we were endorsing the product. These are both excellent questions that deserve good, honest answers.
The classic definition of an experiment is to hold all variables constant except for the one item you want to study. The trick is to come up with a set of circumstances that can be reliably repeated. Then other factors can be added or left out to see what the final result will be.
The first step in designing an experiment is to establish a goal. In this particular case, our goal was to prove whether Stabilant 22 really improves the quality of intermittent electrical connection.
This first step raised lots of questions:
How do you deliberately make an electrical connection intermittent?
Once you've done that, how do you keep it intermittent from one test to the next?
Do the conditions that cause the intermittency truly represent what might actually be seen on a automobile?
Are these conditions better or worse in cases of high humidity-high vibration or temperature extremes?
What happens when the metal gets old and zinc migrates to the surface of the metal terminals?
Is there some type of failure circumstance that you don't know anything about?
Assuming you've got answers to all these questions, still more question pop up: For instance, how are you going to measure the intermittency, and what types of tests and test equipment will be used? And once you've got the data from the experiment, what are you going to do with it and how are you going to prove that the results are really the right information needed to make a rational decision?
This is all worth mentioning because it shows how difficult it can be to create experiment to prove the reliability of every aspect of a given design. When you replace a car part that went bad, what you are looking at is a failure mode that wasn't tested for, or somehow wasn't found. You see failures in the field because there isn't an army of test engineers in the world big enough to test every part under every possible condition.
There are two basic ways to get around the complexity of an experiment. The first is to limit its size by making the best-judgment "guesstimates" as to what the likely set of failure circumstances might be. The second is to make the experiment as practical (real world) as possible.
Over the last couple of years, one of the buzzwords in the automotive world has been FMEA, for Failure Modes and Effects Analysis. FMEA studies are used to get around the problem of having to test everything. The design engineers on a project take a look at all the things that could go wrong and what the effects might be. They then use this to determine what is worth testing and how the testing ought to be performed.
A second testing method is empirical, or practical, testing. This boils down to driving the car up and down the road long enough for something to break. Warranty failures get all the attention they do because, in the end, the customer is the real "test engineer". To keep the Stabilant 22 test down to a manageable size, several assumptions were made. While these could be argued with, I believe they represent a reasonable test circumstance.
Because it was handy, a piece of wiring harness from a MGB was selected, along with the mating terminal from an instrument panel lamp rheostat. Because Stabilant 22 is to claimed to enhance electrical contact, only the terminals were part of the experiment. The brass of the terminals was naturally dirty and the fit not very tight. This setup was chosen because it represented a connection that could easily be made intermittent. The rheostat and associated wires were set up on a vibration table. The table was programmed to shake the connection in a reproducible cyclic fashion meant to include vibration levels equivalent to a car being parked and hitting railroad tracks while being driven off, all within a 60 second period.
For a signal, we chose a square wave voltage very much like what you'd expect to see from a Hall effect sensor to a computer. The object was to send the voltage through the wire, through the terminals and out the other side where it could be monitored by an oscilloscope. The scope we used has a capture and save mode that grabs and stores what it sees.
The first thing we noticed was that the square wave was getting through the connection well enough at the beginning. It was only after a half-hour or so on the vibration table that we found the connection becoming constantly intermittent (see Fig. 1)
The key point to remember is what this intermittency would mean to the computer. You can see by the extra lines, ragged edges and multiple pulses that the information represented by this square wave was badly garbled.
What drives most technicians crazy regarding intermittent electrical problems is that once this garbage passes, it's entirely possible that the vibration will move the terminal to a new and better position for a while, temporarily ending the disruption.
Fig. 2 illustrates what happened to the waveform after the Stabilant 22 was applied to these terminals. In this case, under these test conditions, the intermittency went away. And an hour's worth of testing on the vibration table couldn't bring it back.
There's no end to the tests we could have done under different circumstances, for longer periods of time or with different types of terminals. Short of having an array of test engineers at our beck and call, all that can be said is that Stabilant 22 worked as claimed in the application in which we used it. While we don't endorse products per se, if we find one that works, we'll mention it.
Probably the most important thing to get out of our little experiment is a feeling for the complexity of testing. There isn't an automotive engineer alive who wants to send anything out into the field that isn't perfect. But with the level of complexity so high and the available time so short, mistakes can and, inevitably, will be made. 0--0