Reliability Prediction

[gupta]

While predicting the reliability, how come we know that component follow exponential or weibull or lognormal failure rate.

[RS Support]

If you have data it is fairly straightforward to fit candidate distributions (models), and then ascertain the most appropriate model. One accomplishes that by looking at different models and appropriate goodness-of-fit tests/metrics. This whole area is covered under the subject of “Life Data Analysis”. If you are new to this area, the on-line eTextbook on life data analysis (which can be found on this site) describes such methodologies (href: http://www.weibull.com/lifedatawebcontents.htm).

Additionally software tools like ReliaSoft’s Weibull++ (href: http://www.reliasoft.com/Weibull/features.htm) are designed to do such analyses.

[Larry Strouse]

How can I forecast the number of future failures in a subset of data from a population with a few failures if I know the usage (milse) on the subset and the beta and eta for the population?

[Pantelis]

I am assuming you have an underlying life (weibull) distribution by saying that you know beta and eta. You also say that you know the usage rate (miles). To explain this let me set up the following scenarios.

New Products
Assume that 10 new product have been introduced in the field.
Their usage rate is 1000 miles a month.
Then the expected number of items that will fail from this population after a month is (1-R(1000))*10, after 2 months the cumulative number that will fail is (1-R(2000))*10 and so forth. (Here R(1000) is the reliability for 1000 miles computed based on the underlying life distribution. If eta is in miles then use it as is. If it is in months you will need to scale that based on your usage rate to make sure that your units are all the same.)

Fielded Products
Assume that 5 products are in the field and have accumulated an age of 10,000 miles. How many of these will fail in the next month.
For this conditional probability will need to be utilized. In this case the expected failures are (1- R(t=1000,T=10000))*5. (Here R(t=1000,T=10000) is the conditional reliability for a mission of 1000 miles for units that have already accumulated 10,000 and is again miles computed based on the underlying life distribution.)

One could repeat this for every age group and sum up the failures.

For R definition for the Weibull distribution (including conditional) see: http://www.weibull.com/LifeDataWeb/weibull_statistical_properties.htm

An example of doing this in Weibull++ using the warranty utility is given at: http://www.weibull.com/LifeDataWeb/warranty_analysis.htm

Now what I did not address is the fact that you have few failures. If these are used to compute your distribution, you should look at the confidence intervals, and perhaps utilize these in the predictions to convey the uncertainty on your beta and eta due to the small size utilized.

Lastly, if you are predicting this directly from a single data set, where the failures are included as failures and survivors are included as suspensions in Weibull++, then you can use a built-in function for doing this in the function Wizard. The function is ‘Table of Predicted Failures based on Suspensions’. It basically automates the steps I discussed above for a data set. It is also similar to the warranty analysis tool I referred to earlier.

Hope this helps.

[Larry Strouse]

Then what I need to know is does W++ have a tool where I can run the conditional probabilities for 7600 (approx.) pieces in a batch mode, each of which has a unique known operating time (no grouping), for an additional time of 12,000 miles?

[Pantelis]

Yes,

Once you have created the dataset, from the function wizard choose ‘Table of Predicted Failures based on Suspensions’.

[Dan Wilhelm]

In the Warranty Analysis Tool example, why are the failures at 2 months summed together (Fjun,2 + Fjul,2) but the suspensions only include 1 month (Sjul,2)? It seems to me that the suspensions should also include Sjun,2.

[RS Support]

The nomenclature on this is S (suspended) July (from the July shipment) 2 (after an age of two months). Then this number is the total July shipment minus the failures from the July shipment --- or after two months, there are 134 (140 – 2 – 4 = 134) units from the July shipment still operating, or Sjul,2=134.

[LSTROUS]

Can I use the Predicted Failures based on Suspensions in the Function Wizard with mixed (failed and suspended) data?

[Pantelis]

Yes. The resulting predicted failures are based on the number of suspension and their ages.

[Larry Strouse]

If I want to know which units in the data set to replace first does Weibull++ tell me which units have the highest conditional probability of failure? Is this the measurement I should use to prioritize replacement?

[Pantelis]

Larry,

If the units are wearing out (beta>1) then the oldest units should be replaced first.

[durgarao]

While predicting reliability of electronic equipment using MIL-HDBK-217F(2), how to account for different failure modes of various components. Because no where in the hand book, they have not mentioned how to take into failure modes into account. can u clarify please..

[Pradeep]

I represent Larsen and Toubro Ltd, India. I am carrying out reliability prediction of an electronic component of a low voltage switchgear device. I want to know as to where I can get the failure rate data for various SMD components like resistors, capacitors, zener diodes etc and IC's(CMOS) for MTBF calculations. I am not able to find the failure rates either in the datasheets or on net or in the reliability handbook.

[Pradeep]

I represent Larsen and Toubro Ltd, India. I am carrying out reliability prediction of an electronic component of a low voltage switchgear device. I want to know as to where I can get the failure rate data for various SMD components like resistors, capacitors, zener diodes etc and IC's(CMOS) for MTBF calculations. I am not able to find the failure rates either in the datasheets or on net or in the reliability handbook.

Please help me out by informing as to where these failure rates are available.