## Problems for designers in tolerance stackup analysis

Hi all,

I am writing a program for tolerance stackup analysis, and I wanted to know kind of problems designers face with respect to existing softwares currently.

For example, I found out that using GD&T and monte carlo simulations, many ASME 14.5 rules cannot be incorporated. Also the major problems with these packages are that the results depend on the expertise of the user, not just the GD&T specifications.

So what are other problems that need to be addressed.

Thanks a lot!
 Hello, I work for a high volume manufacturer, so yield prediction is important. Most guesses can be made RSS, but final work must be based on something more substantial, like Monti-Carlo. The single largest difficulty we face is having component data. Distributions are hard to come by, and we've spent a lot of time sampling and characterizing. A tool to aid with the statistics of sampling would make things easier. Best Regards, Mike
 Hey Mike, Thanks a lot for replying! I guess the best way for yield prediction is Monte Carlo. I will surely incorporate that in my program. Can you please elaborate the difficulty you are specifying regarding component data? I mean can you give me an example of the problem you are stating, regarding sampling? As far as distributions are concerned, I guess what we do now is try out various combinations to see which one fits the best. Another alternative is there could be a code which looks at all the data and decide the distribution from its own. Mostly there is a standard machine learning algorithm for it. I'll find out more about this. Thanks again! Regards, Abhi

## Problems for designers in tolerance stackup analysis

Yes,

Time and again, I've written MathCad routines or Excel files to find fits. When designing filters, this is especially handy. I also have a Nelder-Meade optimizer which calls upon a Spice package to solve complex relationships.

However, the most fundamental, day-to-day tools are based upon statistical modeling of the components. Given insight into the probability density function for any given parameter, the Monti-Carlo method can give a fair approximation of yield data.

In reality, it has two shortcomings:(1) suppliers are resistant to admitting distributions, and (2) errors in the modeling typically give biased results such that a correction often needs to be made.

Collecting a database of component PDFs is the most expensive aspect of using these techniques.

- Mike

 Tags gd&t, monte carlo, tolerance analysis