Predictability, Probabilty, Trends, Best Guess

[DelphiBob]

Lets assume that I have 1000 widgets. I know that x widgets failed one of 10 tests during testing
I know what test # was failed and I also know where in sequence they failed.

Sample random data

Widget #1 - 10 passed
Widget #11 failed
Widget #12 - 100 passed
Widget #101 - 103 failed
Widget #104 - 110 passed
Widget #111 failed
Widget #112 - 356 passed
Widget #357 failed
Widget #358 passed
Widget #359 failed
Widget #360 - 900 passed
Widget #901 - 910 failed
Widget #911 - 990 passed
Widget #991 - 994 failed
Widget #995 - 1000 passed

I need to determine if there is any Trend(pattern) in the failures, is
there a prediction or probability of when widgets will fail?

Thanks,
Bob Hiller
Lifts for the Disabled LLC

[Patrick_P.]

Hey Bob,

Any functions we have in Measurement Studio for probability and trending are contained in the CWStat control.  I would recommend looking at the help for this control.  You can access the help by going to Start>>All Programs>>National Instruments>>Measurement Studio>>Help. However, generally the formulas for trending and prediction of failures are pretty complex.  I doubt you would be able to implement one with just a single call to one of our statistical functions. 

[DelphiBob]

It does not matter if I have to make 100 calls to your libraries, I just need some sort of crude example to start with. Once again, All I am looking for is a Best Guess. It does not have to be guaranteed.

Bob Hiller

[Patrick_P.]

Hey Bob,

There are a couple of different ways you could approach this.  Using the CWArray.Search function you could search through your array and see which units had failed.  The search function will return an array which includes the indices of the failed units.  You could then process this data.  Another possibility would be to just use the probability of failure.  This would just be the # of units that failed divided by the total # of units tested.  If you need help actually using the CWStat control there is actually an example that demonstrates how to use it.
It is located at "C:\Program Files\National Instruments\MeasurementStudio\VB\Samples\Analysis\Simple Statistics" by default.

Message Edited by Patrick P. on 03-07-2006 07:47 PM

[DelphiBob]

Pat,

Re: Using the CWArray.Search function you could search through your array and see which units had failed. I already know what has failed. I write all part data to a SQL server and then read it back for evaluation at a latter time. 

Re:Another possibility would be to just use the probability of failure.  This would just be the # of units that failed divided by the total # of units tested. This is percentage of failures, not probability of failures.

I may know that 10% of parts failed test #1. What I am trying to determine is if there is any pattern to the failures. Did every 5th part fail, every 12th or was it in a range. Every 10th -15th part there is a failure.

Thanks again,

Bob Hiller

[Patrick_P.]

Hey Bob,

You could keep "bins" for each test.  Increment the bins every time a test position fails.  After several runs you could make a histogram of these bins, and it would tell you which tests were more likely to fail.  If you are looking for a more complicated algorithm, I would recommend to do some research on error trending on the web.  Usually people develop algorithms for their specific applications.