[rescue] Slightly OT: Bad Cap Saga

[Curious George]

--- On Tue, 8/19/08, Geoffrey S. Mendelson <gsm at mendelson.com> wrote:

> On Tue, Aug 19, 2008 at 10:02:17AM -0700, Curious George
> wrote:
> 
> >No, different issue entirely.  The device didn't exist -- we
> >wanted them to make one for us.  Since we were unfamiliar with the
> >tradeoffs involved in *that* design process, we couldn't on our
> >own come up with specs that would have ensured an economical design.
> 
> Ok, now that really confuses me. Following your example below:
> (to avoid more confusion, I'll wait for others to read it)
> 
> >By way of an arbitrary (and fictional!) example:  imagine you're
> >asking them to design a thermal printhead assembly. This is
> >little more than a bunch of resistors with digital switches
> >controlling each individually.
> >
> >From the physical (thermal?) standpoint, all you care about is
> >the dimensions of the element and getting a certain amount of heat
> >out of it in a certain time interval.
> >
> >So, a zero-th order approach says "pick an R and specify V to get
> >the desired amount of power dissipation".
> >
> >But, as V goes up, there are consequences to the switching devices
> >(i.e., must have higher breakdown voltages).  As V goes down (because
> >R has been picked as a lower value), there are *other* consequences
> >for the same electronics (e.g., I goes up).
> >
> >What's *their* (manufacturer) process technology like?  What's the
> >sweet spot for the control electronics?
> >
> >What ramifications does this have to the geometry of the resistive
> >devices?  Does it affect the technology used to deposit them on
> >the "printing surface"?  Does it affect reliability?  Does it
> >affect the cycle time of the elements (thermally)? etc.
> >
> >The cost of learning this is far outweighed by the "cost" of
> >outsourcing the problem completely.  So, what you're wanting is
> >to engage in an *engineering* "give-and-take" (i.e., a dynamic
> >design process) with the supplier to figure out how to find that
> >sweet spot.  Obviously, the changes made in the printhead's
> >design will have ramifications to the rest of the system -- things
> >that the supplier will not be aware of (hence, he's in the same
> >boat as you -- except the mirror image problem!).
> 
> Ok, I'm confused. Since it is a device that exists,

No, the example I chose happens to be a device that exists (today).
The actual item in question did NOT exist (at the time).

To put the example in a realistic context:  imagine thermal
printheads did not exist at the time we approached this firm
(who perhaps manufactures thin film devices AND, like many
Japanese firms, also has relationships with firms that do
integrated active components -- e.g., the switching circuitry).

Clearly, it would be silly for *us* to learn how to deposit
thin film resistors on an 11" wide surface every 0.002" *and*
connect those to a (custom?) designed switching circuit capable
of handling all that current (such a printhead would dissipate
about 400W) without anythinig MELTING in the process.  Then, figure
out how to move all that excess heat out of there fast enough
that the head doesn't just go into thermal overload and start
parts of the enclosure, etc.

So, you get IN BED with a trusted supplier and ask them to
come up with a design for you.  But, their design will clearly
affect the rest of *your* design.  You (and they -- their
engineers) know that.  But, their attitude (cultural) is that
they want to meet *your* requirements.  However, your
requirements are a locus of possible requirements that are
affected by *their* pricing (which they can't determine until they
have a design!).

As such, you want to get a feel for what issues/criteria drive
the cost in their design.  Then, see how those issues interact
with your locus of potential solutions.  E.g., if they would like to
run at a high operating voltage, that means your power supply
has to *deliver* that higher voltage.  This could push the design
of the power supply into using a different topology or different
class of components (different ratings, etc.)

I'm saying is the cultural differences seem to result in "us pulling
and them PUSHING" (instead of both sides pulling). So, its hard to
find the sweet spot as they seem to be agreeable to *any* specs
you throw at them.

(hmmm... this may be too subtle of an analogy?)
   
> unless you are building a satellite for a US company, you just
> buy a COTS (commerical off the shelf) unit. I'm not sure about
> SurreySat, but most of the small "bus" manufacturers (people
> who make cheap satellites) do exactly that. 

Huh?  What have satellites got to do with it?  :>

There are lots of things that don't exist that you would want
to rely on some existing company's expertice to deliver.  E.g.,
if I wanted to make candy coated bits of caramel, I'd call the
M&M company instead of trying to figure out a process to do it
myself!

> No, I don't build them I got this info from someone who does and
> he tells me about the public information one can find on the web.
> 
> >I found that American firms are much easier to get into this
> >"give-and-take" sort of specification process.  The Japanese
> >seem to want to "do what you want" (and just adjust their
> >price accordingly)
> 
> Exactly. They adjust the specs to fit yours, but I will bet
> they don't adjust the product. :-)

If the product exists...

> >The cost of replacing/repairing a "bad board" (hence, a
> >"defective product") can easily outweigh the value of the
> >entire sale.
> >
> >I recall one medical device designed many years ago where the
> >cost of the device (DM+DL) was less than half the cost of 
> >sending someone "down the road" (locally) to repair it!
> >
> >For those devices that you can't just place in a cardboard
> >box and *mail* complete replacements, you *really* don't want
> >to have to put someone on a plane, etc. to troubleshoot and
> >repair.
> >
> >(and that's assuming your customer is content with the day
> >or three that this will take)
> 
> Then you don't buy the cheap stuff, you buy the expensive, individually
> spec'ed and test components. For example, if you spec the board to
> work with a "5 nines" 99.999% up time rate at 45c, each component
> is spec'ed at 45c or greater and the board is tested at that temp.

Exactly!  Hence the reason (many posts ago  :> ) of asking if the
"bad cap" issue was *just* an issue of "bad caps" or if something
else was going on.  (because you can avoid bad components)

> Cheaper boards are tested at that temp, no matter what the specs,
> and really cheap boards are built with whatever they can get and
> "statisticaly" tested, i.e. if they test one in> 1,000 and 10 in a row
> don't fail, they are 99.999% ok at that temp. :-)
> 
> >This suggests the failure rate is driven solely by the quality
> >of the components?  (or, that the production quality of all of
> >these houses are virtually identical)?
> 
> Yes, and yes. They use similar materials, similar processes, and buy the
> same components. They also do similar testing.

So, people choose between House A and House B based solely on *price*?
Or, perhaps "established relationships"?

> >I take it you are on the rim?
> 
> Are you kidding? After all of the posts, I've made on this, you haven't
> read my sig once. I'm going to be mean and make you read it this
> time. :-)

I *have* read it.  But, you comments about "over here" in conjunction
with all this stuff about Taiwanese firms *suggests* you are "over
there"!  :<  A .sig can say damn near anything -- for all I know,
you're currently "abroad" looking into a manufacturing problem forwarding
your mail through mendelson.com!  :>
 
> >Yeah -- except they all SNAP together with those frigging
> >flimsy snaps!!  :-/
> 
> Well, that's spec'ed by Dell. Or more likley the consulting firm they
> hired to do time study of their processes and recommend how to minimize
> construction time.

Of course.  It also minimizes DEstruction time!  :<  (one of the
reasons I dislike laptops!)

> >Exactly.  When your (my) reputation stands to win/lose based on the
> >quality of the product, you want some reassurances that your
> >suppliers are as committed to making (not FAKING) a good product
> >as you are.  :<
> 
> Yes. But it also means that you need to test their product.
> As President Reagan said about privatising government functions,
> "trust, but verify".

I thought that was about nukes... <shrug>

In this case, it means its silly to pay them to test.  Instead,
arrange your contract so they suffer for each failure that you
find (which gives them an incentive to improve their quality)
 
> >No, but, IME, the Industry (pick an industry!) is always much
> >smaller than you can imagine.  I.e. "word gets around" by
> >whatever means.  People change jobs and bring their horror
> >stories with them, etc.
> 
> Sure, but to be blunt, rarely does the customer listen to
> them.