Reliability discussions

[R.A.Madgulkar]

I have a failure data on a part which failed at following hours on field

1482,3300,4562,4392,3292,6245,1150 hours. I discontinued use of this part and had a part from other supplier which I put on test at these locations and have not failed on any of the locations. The hours completed are 1533,1332,1269,908,722,679,611. How can I statistically prove that the new part has a better reliability than old part as no new part has failed as yet in field.

[Pantelis RS]

There are methods for doing this. As a starting point see the discussion at http://www.weibull.com/LifeDataWeb/tests_of_comparison.htm


Now with regards to your data, and with no failures on the second set, some assumptions would have to be made as to the behavior of the failure rate of the new component (as an example the value of the Weibull Beta) before proceeding.

Now as a secondary item and just by looking at your data – it is not prudent to start making any comparisons using the above data. The reason for this is that not enough hours have been accumulated on the second design to compare.

[jocdelarche]

I have a problem with a Weibull parameter that I need to calculate for a MTBF. How can I find this parameter from data sheet, or how can I calculate it for each components?

[Pantelis]

I presume the Weibull parameter you are referring to is the beta parameter (shape parameter/slope). If the data set has no failures (as with the first item in this discussion) then one cannot compute this parameter from the data. This should be fairly obvious since the beta parameter dictates the behavior of the failure rate function – thus with no observed failures one cannot make any statements about the behavior of this rate (is it increasing, decreasing or does it remain constant).

In such cases the analysts can make some assumptions as the value of this parameter (a so called 1-parameter Weibull formulation –see http://www.weibull.com/LifeDataWeb/weibull_probability_density_function.htm --) e.g. assume the same beta from the other set, etc. Of course - and as with any assumptions - the validity of the results is directly correlated to the validity of the assumption.

Hope this helps

[Larry Strouse]

What does the Z value signify when I do multiplot analysis using contour plots? It is featured on the edit plot.

[RS Support]

Larry -- not sure what you z you are asking about?

[Larry Strouse]

I can send you the file and you can view. Direct me to where to send it.

[Larry Strouse]

I have a Weibull beta of 2.00. Can a MTBF be assigned to this distribution and, if so, is it the average time of all the failures from 0% to 100% unreliability?

[RS Support]

Q1: send the file with the question to Support@ReliaSoft.com

[RS Support]

Q2: The mean of the Weibull distribution (2-parameter) is
Mean=Eta*GAMMAF(1/Beta+1)
See also http://www.weibull.com/LifeDataWeb/weibull_statistical_properties.htm

Now, since the mean is a function of both of the parameters its determination requires the knowledge of both.

The mean is the average -- and for symmetrical distribution (i.e. normal) this is at the 50% point. However for non-symmetrical distribution, such as the Weibull the mean can be at a value other then the 50% point.

[Larry Strouse]

Can I read off MTBF/MTFF in directly from W++ or BS6?

[RS Support]

Larry,

In Weibull++, MTBF can be accessed via the Quick Calculation Pad (QCP). The QCP can be loaded by selecting Quick Calculation Pad from the Tools menu (data must be calculated). Under Options for Calculations select the Mean Life option and click on Calculate. In BlockSim, the MTBF (for analytical solution) and MTFF (for simulation) can also be estimated using the QCP. When conducting a simulation, MTFF can also be accessed from the simulation results.

Hope this helps.

[Larry Strouse]

If I specify 90% reliability at one-sided 95% CI, how can I determine the shape of the distribution from 95-100%? I am not comfortable simply stating it is normally distributed.

[Larry Strouse]

Do you know of a learned treatise that discusses the legal implications of reliability analysis, life prediction, etc. in the context of product liability, "failure", etc.?

[Larry Strouse]

Does BlockSim 6 support generating a graph of cost vs, replacement time as shown on page 217 of the BlockSim 6 analysis handbook?

[Pantelis]

Larry, not directly. You can generate the values in Weibull++, from a general spreadsheet, using the Function Wizard and the "Table of Block Replacement Times" function.

[Larry Strouse]

How can I construct a RBD diagram made up of components that have been in service for differing times. I know the distribution parameters for each of the components.

[RS Support]

Yes. You will need to use conditional probability for each block - given the current age of each item. In BlockSim you can specify the current age af each block.

[Larry Strouse]

Can you tell me how to do this?

[Larry Strouse]

It does not seem that I can enter a different start time and run time for each component in the RBD diagram. I need to do this since some of the components were repaired at different times.

[RS Support]

Larry, you can enter a different current age for each component from the Block Properties window for each block. The run time however is assumed to be the same for all.

[Larry Strouse]

Where is the current age field in Block Properties??? I have mine open and don't see it.

[RS Support]

Larry,

BlockSim can run in two modes, simple and advanced. You are running in simple mode (a choice you made when you first installed BlockSim) and thus do no see the option.

You need to change to the advanced mode. To do that from the File|User Setup

From the "Other" Tab click on "Modify Default Presets" and switch to the advanced setting.

This and other options will now be available to you.

[Larry Strouse]

THX-got it! Will be back with more questions.

[Larry Strouse]

The FAA defines "improbable" as .00001 events per flight hour. How can I translate this into reliability values (Weibull, exponential, etc.)

[Pantelis]

For the exponential case
lambda=0.00001 events/flight hours
then
R=exp(-lambda*t) and t is flight hours.

Where now the Reliability is then the probability that an event (instead of a failure) will not occur from 0 to time t.

Since Weibull has a time varying failure rate (or event rate in this case) direct translation is not possible unless one assumes some value for lets say the beta parameter.

[Larry Strouse]

If I run the function wizard for conditional probability and conduct an inspection program to remove the highest probability units from the population, how should I recalculate the conditional probability - with all of the inspected and uninspected units or just the uninspected units?

[Pantelis]

Larry, I am not sure I follow.

[Larry Strouse]

I have a group of units and calculate the group reliability in W++. I now apply a modification to the surviving units which is designed to extend their life. How should I calculate the new reliability i.e. by starting the time at zero from the date I applied the modification or from the original built date which includes the unmodified usage time?

[Pantelis]

What did the modification do? By restarting at time at zero you are assuming 100% renewal, or as-good-as-new on the survivors. Furthermore did the modification change their life characteristics? If so you shouldn’t be mixing these with the old units.

[Larry Strouse]

The modification to the existing units was a structural modification that strengthened the failure area and reduced stress but did not make the part "good as new". What do you mean "shouldn't be mixing these with the old units." Both the old units and the modifed units are in service. So I am back to my original question-how should I calculate the reliabilty of the modified units?
Thank you in advance.

[Pantelis]

Well lets say you have 100 units and 90 survived up to time t. At time t when you fit a life model based on this homogeneous population. Now we’ll take the 90 survivors and modify them, keeping in mind that the survivors already accumulated an age of t.
The question is what did the modification do to the units reliability wise? If one assumes that the “failure rate behavior” did not change (Beta is the same if using Weibull) but the life has doubled then eta_new for the new units would be 2*eta . The modified units will now fail based on these parameters (using conditional probability). The problem is with the assumptions I made, i.e. how much better are the new units. If you have no data on the new units then it’s a SWAG.

[Larry Strouse]

I have life data for the 90 survivors. Then I apply the modification and I have life data since the mod was applied. The would not expect the betas to be the same since I have reduced the stress and may have introduced some welding defects during the retrofit which could result in a different beta right? My main question is should I refigure the reliabiity paremeters for the 90 modified units with T=0 at the time of the mod application???