FIT MTBF amp predictedactual failure rate calculations

[scott jones]

I need to learn how to calculate FIT, MTBF and predicted/actual failure rates for electronic printed circuit boards. They can have as few as 100 to a maximum of 500 components installed. Where could I go to learn this? Also, is there industry standard literature about this. I need to learn this information and am looking for a book and/or website location to go to. If there is software available to assist in these calculations, all the better. I am entry level and wish to progress to novice.
Thanks

[tarik]

If you are testing the whole electronic PCB and collect data for failure times and suspensions (units that survived the test) you can use the Weibull++ software to analyze your data and calculate the metrics you need.
For more info about Weibull++ and life data analysis check the following links:
http://reliasoft.com/Weibull/index.htm
http://www.weibull.com/lifedatawebcontents.htm

If you colleted data for each component, analyze the data with Weibull ++ and then build a Reliability Block Diagram RBD with the BlockSim software that describes the configuration of your system (for example, 100 components in series), analyze the system and calculate the metrics you need.
For more info about BlockSim and System Reliability Analysis check the following links:
http://reliasoft.com/BlockSim/index.html
http://www.weibull.com/systemrelwebcontents.htm


Since you wish to learn more about this area, I suggest you read the following material:
Blueprint for Implementing a Comprehensive Reliability Engineering Program: http://www.weibull.com/Articles/RelIntro/index.htm

[Peter Dopheide]

How do you get on if reliability is so good that there have not been enough failures to get any handle of failure rate.
Our product lines are mainly electronic and we get virtually no failures which are due to componenet failure. A few have poor solder joints but they are fixed before shipping.

[tarik]

If you have very no failure data you can use the exponential (if you are willing to assume a constant failure rate) distribution to model the suspension times of the units that didn't fail or you can use the 1-parameter Weibull distribution (you have to assume a beta value from prior knowledge). For more details you can read the following material: http://www.weibull.com/LifeDataWeb/weibull_probability_density_function.htm.

The other approach is to use standard-based prediction that bases reliability estimate on databases of failure rate information about different common components. For more info visit: http://www.reliasoft.com/predict/index.htm. This approach also assumes constant failure rate.

[ROHAYA]

Regarding FIT. How to calculate FIT? Is it applicable for operating life test and any non-bias test?

[tarik]

Additions to Peter’s question:
The other approaches are to use the Bayesian anlaysis, in which you can use the previous knowledge about the model parameters in addition to the little information you collected from test to estimate reliability. This new approach will become available in the next version of Weibull++, which will become available very soon http://www.reliasoft.com/Weibull/version7.htm.

Another good approach is to use field data (customer returns, repairs..) http://www.weibull.com/Articles/RelIntro/Field_Data.htm

[tarik]

The FIT or Failure Rate general equation can be found at http://www.weibull.com/LifeDataWeb/basic_statistical_definitions.htm. For distribution specific forms of the failure rate you can check the online reference for life data analysis http://www.weibull.com/lifedatawebcontents.htm and check the formulas for Failure Rates depending on the distribution you are assuming.
Note that Weibull++ offers easy ways to calculate the Failure Rate (for a certain time value) in the QCP consule and also plots the Failure Rate over time.