1
\$\begingroup\$

I'm taking an assembly of three boards to a small production run. There's a main board (n=25) and a satellite sensor board (n=100). The sensor board connects to an adapter board, which simply connects the sensor to the main board through an 8-conductor FFC.

The main board is a USB device, reading signals from a variety of sensors through a harness as well as the FFC, and sending the signals as an HID. I'm thinking about the QC (and have been for a while). The contract assemblers would like to know that the components they're working with are good, of course. Since there's only 25 main boards, I thought testing each one for enumeration on a PC and making sure the signals from the test harness come through isn't that unreasonable -- I figure 2 minutes per unit, max.

For the sensor board, I figure on providing a kitted test platform to test the sensor, with maybe an led bar display. Twenty seconds a unit, max.

The issue is the ZIF FFC connection between boards. I'm using these connectors I'd like to hold down this testing, as I think it will be slow and fairly tedious. I'm thinking about spot-checking about 3-5 units, and going for visual inspection of the solder on the connectors for all the rest (they're fairly large pitch).

Does this make sense? Does it make more sense, with respect to the FFC's, to test the first few off the run then a random sample?

I've had zero problems with the hand-populated prototypes, which have really been run through the mill. I'm using a very reputable local board assembly plant that routinely constructs to mil-spec.

\$\endgroup\$
1
\$\begingroup\$

You don't have to test everything, only those parts that you want to make sure work.

For a large production run you usually make test jigs that carefully test everything. Designing and getting the production test jigs made is usually a larger effort than for the thing you are testing. However, for small volume as you have, you can trade off labor and some uncertainty with test jig creation.

In this case, a test jig for the cable sounds simple. You use a ZIF socket for each end and have a microcontroller run a walking 1s and walking 0s pattern and the like. That will be instantaneous in human terms. This doesn't have to be on a fancy machined custom jig. Test jigs can be built as one-off prototypes, as long a they are physically solid.

Can you test everything together as a single system? Ask yourself what can still be wrong if you put known stimulus into the sensors on one end and see the correct values over the USB at the other end. Obviously a lot has to be working for that all to be correct, although there could possibly still be a few things wrong. For example, the unit may appear to operate correctly, but could be drawing more power than it should and something therefore getting warmer than it should, causing it eventually to fail over time or at higher temperature. Of course drawing too much power is a failure in its own right. In this example the solution could be as easy as using a bench supply with a current readout, and having the production tech verify the current is below some value. However, keep in mind that anytime you are asking a human to do something, there will be a fraction of the time they get it wrong.

Think about it carefully. Would a system test as described above excersize every line in the cable? If so, then you need to do nothing further. If not, you first have to ask yourself why that line is really there. Then figure out a way to test it.

The test strategy needs to be part of the design from the beginning, and producing the test system, even if it is just documentation to the production techs, is something that needs to be considered in the schedule and budgeted for up front. As I said before, usually the test fixture is a larger project than just the thing it is testing. This is one example of where experience matters. If you've gotten to the end and are just now thinking about testing, then you have already screwed up. There simply may not be a way to fix the mess within all constraints of time, money, and test coverage. Get someone who knows what they are doing next time.

 

\$\endgroup\$
  • \$\begingroup\$ Thanks Olin. Testing hasn't been ignored-the test harness and platform for each subsystem are already built and running, and have tested each prototype (except for the output display mentioned above). The signals that pass through the FFC are actually T'ed off and buffered, so I can test the signal handling separately without the FFC in place -- and was designed this way because the test plan was part of the design. I suppose I'm regretting leaving the FFC's out of the plan now, so I'm tweaking. Still well in the black on my Gantt chart-- which is why I'm asking now and not in 30 days. \$\endgroup\$ – Scott Seidman Nov 13 '12 at 15:01
  • \$\begingroup\$ In fact, there's a hard jumper built in to branch into a test mode, where I can test every signal with reference inputs. I'd still like every main board tested for enumeration, though. \$\endgroup\$ – Scott Seidman Nov 13 '12 at 15:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.