Investigation: What is your way to ananlyze test data?

[Engineer Chandler]

Dear guys,

In order to improve our performance of R&D, and production line, I'd like to ask all of you to talk about what information you gather for test data, how you effectively stored it and what kind of information you expect to get after analyzing it.

The information I gather for test data is:

measurements for each test item; duration for each test item; error information when error happens; duration for the whole test.

I stored test data into a plain text file for each test.

I use a tool to gather data from all those text file into a table. And then I can see which test has failed and why it failed. This tool also calculates the max and min values for each measurement test item.

I feel these are not enough for a ATE system. So long to listen to your opinions.

[Graziano]

Hi!
   I think that the best way to analyze data is by means of graphs.  You can plot test result, in a convenient way, you can process data someway before plotting.... it depends upon your needs.  Of course, this is good  for a human-reader! A computer will work on raw data, so you'd better find an algorithm, or an indicator, that can be meanful in your test.

   Hope it helps....

graziano

[Graziano]

One more issue....
 
About storing data!  I think it's worth to store data in a portable format.  This doesn't mean that you have to use a proprietary mechanism (e.g. save data in .xls files).  Many times, if you format properly data (i.e. points separed by tabs, different lines for different vectors, and other simple stuffs like these), your data will be readable to many third party softwares, like Matlab (by MathWorks), Origin (by OriginLab) or Excel or different ones.  I can suggest you to check that data you saved in files are properly imported in these suites.

Have a nice day!

graziano

[Engineer Chandler]

Thanks, Those are really great ideas. They remind me a lot.

I'd like to put data into a database file. And I can use the advantage of ODBC to ananlyze the data.

I am gonna design my new ATE SW now.

[daijoubu]

In my previous work, I have maintained run charts for the first pass yield and run-rate for each test system.  I maintained a pareto chart comparing the yield for each individual test item.

I would also suggest moving the data into a database.  Text files are okay until you get large amounts of data...  Databases allow better management of the data, quicker filtering, easier maintenance at the cost of not being as easy to browse.

Message Edited by daijoubu on 03-28-2006 08:03 AM

[oldvaxguy]

Analyzing Test data.  Wow that's an involved subject.   There are a number of methods to store data for analysis. 

First you need to write a requirements lsit.  Here are a few questions that always need to be asked.

How much data you wish to store?  (   100k or 100 meg )

How Fast do you need to collect the data?  (  1 sample second or 10,000 samples per second ?)

What are you going to use to analyize it with?  ( Excel , Matlab, home brew... ?)

How long are you going to collect data?   ( 1 second  or 36 hours worth? ) 

Can you miss a sample or two and not have a problem?

These are just starter questions. 

If you answer these then you will be well along the way to solve your problem.

The old vax guy

[daijoubu]

I guess I can answer a few of these questions based on the production test systems I've built...

---

@Old vax guy wrote:

Analyzing Test data.  Wow that's an involved subject.   There are a number of methods to store data for analysis. 

First you need to write a requirements lsit.  Here are a few questions that always need to be asked.

How much data you wish to store?  (   100k or 100 meg )
Unlimited.... 

How Fast do you need to collect the data?  (  1 sample second or 10,000 samples per second ?)
From 1 sample per second to 1 sample per minute depending on the speed of the product under test.  On rare instances, I will pull in 1000 points from the oscilloscope.

What are you going to use to analyize it with?  ( Excel , Matlab, home brew... ?)
All of the above. 

How long are you going to collect data?   ( 1 second  or 36 hours worth? )
16 hours a day, 6 days a week for the next 5 to 20 years.

Can you miss a sample or two and not have a problem?
No.

These are just starter questions. 

If you answer these then you will be well along the way to solve your problem.

The old vax guy

---