In the movie Disclosure character Tom Sanders, played by Michael Douglas, is head of Manufacturing for a start-up company with a revolutionary new technology, which is about to be merged with a publishing company. Without spoiling things too much, with the production line seemingly beset by production problems, he is clearly being blamed – and forced out – in part by the very aggressive new head of the company, played by Demi Moore, someone with whom he has worked before (and with whom he used to be romantically involved).
Alone and seemingly in a hopeless situation, he receives an email saying “Fix the problem.” A clue to his course of action.
An excellent book I read some years ago introduced me to the concept of multi-var analysis; a production problem solving technique which I’ve used successfully (for example, here). One of the things the book stressed was that – in the best of all possible worlds, of course – one can only say a problem is solved when you know what to do to turn it on and off. That is to say, you’ve identified to root cause of the failure (using multi-var, Ishikawa diagrams, etc., as a part of a formal problem-solving process) and taken deliberate action to solve it and prevent its recurrence (i.e., more than tweaking things and then putting a sign up saying “For G-d’s sake don’t touch this dial!”).
An example of identifying the root cause and coming up with a solution comes from a glue pouring operation at Ford. Hot-melt glue was being dispensed into the groove of a standard tongue-in-groove glue joint, with the mating part being pressed in at the next station downstream. Unfortunately the glue was not cooperating, not always being dispensed into the bottom of the glue groove, but sometimes being poured on the sides of the groove, on the edge of the groove “cup”, and even oozing over to drip down the outside. This was creating a high scrap rate as in the next station a lens assembly with a decorated bezel insert was pressed into the body with the glue. With a bad glue pour the headlight would fail the leak test, requiring disassembly to recover the valuable components such as the reflector inside and hardware on the back… not to mention material scrapped and an assembly cycle wasted.
I was asked to look into an inspection system to on-the-fly inspect the glue pour to kick out a lamp body with a bad glue pour. With the glue being black and the body also being black, but the glue being a hot melt, an infrared (IR) camera was the obvious choice. We invited a couple of IR camera makers to come in, set up equipment to take live pictures, and quote a system. But this would only contain the problem, not solve it – and with the quotations coming in, do so quite expensively.
The logical question which I then asked was “Why are we getting bad glue pours?” In a theoretically perfect world, everything should be fine. The answer was “Variation”; the next Why? was “Why are we getting variation?” followed quickly by “Where is the variation?”
There were three places where we could be seeing variation. The first was in the glue dispensing system (robot + dispensing machine). But the robot had a manufacturer-stated repeatability in its path to within a fraction of a millimeter. The dispensing system was likewise very repeatable in shot size and flow rate. Verification of these was not just a matter of taking the manufacturers’ word for it, but was a part of the machine acceptance protocol. Any variation in this part of the system was miniscule. The one program (remember this) was spot-on repeatable.
The next place to look was at the fixtures, of which there were somewhere around 40. There, too, the fixture acceptance protocol required that we examine fixture-to-fixture variation as a part of final approval of fixtures. ANOVA testing of the fixtures during acceptance verified that the fixtures varied within a few percent of each other; fantastic Gage R&R numbers.
Last up were the parts themselves. Since this was a high-volume molded body, how many mold cavities were there? Three. It’s a given that any product with multiple cavities will have cavity-to-cavity variations. On top of that the bodies were made from polypropylene, which among molders is often nicknamed “polywarpylene” for its propensity to warp and distort from its ideal, molded shape. Some dimensional checks verified that the positions of the grooves from cavity to cavity floated by up to half of the groove width itself.
Recall from above there was one robot program? This one program was a compromise program pouring into three different grooves whose position varied from piece to piece. Since the variation was the result of an inherent property of the material, it couldn’t be eliminated.
So I proposed the following:
- Develop custom programs for each cavity (requiring the purchase of extra memory for the robot).
- Put a bar code into each cavity identifying it, and an optical scanner in the station upstream from the dispensing area to pass the cavity information to the robot.
- The robot, using a cavity-custom program, would adapt to the variation between the cavities and eliminate the problems arising from the compromise program.
Unfortunately my plan was not implemented, and instead the plant elected to live with the scrap rate and troubleshoot it as they had always done before – reducing, but never truly fixing, the problem. But this could have been fixed for a NPV cost far, far less than the scrap costs the plant incurred through the life of the product.
Lesson: When faced with variability that is inherent in the system and which cannot be engineered out, think about how you can adapt to it to neutralize its effect.
Part II will look at a package perforation issue I was asked to investigate as a part of a contract position I had at a medical device company.
© 2013, David Hunt, PE