| ??? 06/06/10 14:01 Read: times |
#176447 - Yes but... Responding to: ???'s previous message |
Erik said:
the only way to make sure something works is DESIGN. Neil said:
For example I have a test program. To test it I read all the results in.
Then set the test parameter to fail that number High, then low. Done. Another Test Runs in to a load Resistor. What if it is open? I added a test to insure the voltage drops. If your stuff is digital go no go, then checking both states says it is good. Analog is the same if the value changes from the idle range to the good range happiness. But it is your fixture. You have to look at what it does. Can a bad component or open wire cause a false pass. And what can you do to detect that? Example I have a 10V and 20V input test. There is no reasonable failure that will allow the test to pass with bad test fixture. It is always case by case. optimal is a test fixture tester. But where do you stop. Per said:
Most digital signals the tester picks up are possible to control by the microcontroller. So the microcontroller can set the signal to both states and the tester tests that it is low when expected to be low and high when expected to be high.
For analog signals, it's good if the tester can multiplex between the test point and some references. A problem is when measuring a voltage that is expected to be 0V - do you measure 0V or do you measure air? If the test point has a reasonable reistance, then you can add a pull-up resistor to the test jig, and have the jig measure the value with and without the pull-up. If you measure a strong signal, the pull-up will affect little. If you measure "air", the pull-up will let you measure the pull-up voltage instead. In the end, loopback or programmable stimuli, should be able to cover almost all parts of the test jig and tested measurement points. Todays very fast and cheap micrcontrollers allows many test points to be handled concurrently without need for expensive instruments, so having an extra step or two for each test point need not affect the total testing time too much. All very well said. Might I add that you need to also protect against a fault on a DUT that could damage the Test Jig. And most important is good documentation so that you can repair the test jig or even make a new one. (This is always an area of concern when making a one-of since a quick fix could not make it into the documentation). Another hint for computer controlled test jigs would be to parameterize everything so that you can tweak ranges without having to re-compile the software for the test jig and to also partition each sub-test so that it can be done individually and repeatedly for a particular DUT that is causing problems. BUT ISO annoyingly and unfortunately correctly says "How do you know that you have covered everything?" What they want is a validation and that means that every test jig (and we probably have 100 or more) must go through some form of FMEA (Failure Mode and Effects Analysis) despite the fact that we have never had any product that passed the test as a result of a faulty test jig. Validation would best be done by a 3rd party as is the case of any form of review but in our organization QA does not appear to have the knowledge to do this, so it means that a "minor" NCR will probably take up most most of the year being unproductive. Yet the all work must continue. Even creating samples as was originally suggested by ISO is a significant effort with ongoing maintenance implications. I guess I am just looking for a silver bullet <sigh>! If I may modify my original question- has anyone come up against ISO on this subject? -Aubrey |
| Topic | Author | Date |
| test the test jig | 01/01/70 00:00 | |
| only ISO believes ... | 01/01/70 00:00 | |
| "Calibration" | 01/01/70 00:00 | |
| ISO9000, concepts and lies | 01/01/70 00:00 | |
| doghouse and flagpole | 01/01/70 00:00 | |
| I agree... | 01/01/70 00:00 | |
| Use loopback and stimuli | 01/01/70 00:00 | |
| "Design For Test" (DFT) | 01/01/70 00:00 | |
| You can not ask us | 01/01/70 00:00 | |
| Yes but... | 01/01/70 00:00 | |
| No silver bullet | 01/01/70 00:00 | |
| Calibrations... NIST traceability.... fun | 01/01/70 00:00 | |
| That is what traceable means | 01/01/70 00:00 | |
| Should be few steps to a national reference | 01/01/70 00:00 | |
| Article on the topic | 01/01/70 00:00 | |
Test the test harness | 01/01/70 00:00 |