Test Planning Question

[Tim]

I am in the planning stages of a durability test of a system that has two modes.

Mode 1 = Front Cycle
Mode 2 = Rear Cycle

The total # of Cycles to run is 10,000.

What is not known is the end customer usage profile... ie How may cycles should be Mode 1 and How many should be Mode 2?

Mode 1 and Mode 2 use different drive systems for the action but are related through other drive system components...some components will see only mode 1 cycles, some will see only mode 2 and some will see mode 1 + mode 2...?


I need to obtain a system R/C or 90/90 and I am trying to determine the best way to combine Mode 1 and Mode 2 in testing...

I am thinking about doing the following

4 samples run Mode 1 / Mode 2 20 / 80
4 samples run Mode 1 / Mode 2 80 / 20
4 samples run Mode 1 / Mode 2 50 / 50

Would there be any advantage that you can see to biasing one mode over the other? (80/20)

All tests will be run to failure and a weibull analysis completed...is there a problem combining all samples for this analysis? or do I have to analyize them seperatly? Any suggestions on the best way to utilize parts and cycles/modes?

Thanks

[tarik]

I can’t recommend what test profile to use, without understanding how your customers use the product.

If you don’t know this piece of information, the only thing you can do is obtain a reliability model that can relate the reliability to the different modes and then use that model to study different scenarios, or use it to obtain the reliability information after you do surveys/collect data..etc to understand your customers usage behavior.

You can do this type of analysis using 2 approaches:

1-Test under the different conditions you mentioned (or other ones, ex : 70/30, 60/40, 90/10), collect the data, analyze it separately (can’t combine all the different tests data) using the Weibull++ software. This type of analysis will give you a separate reliability model for each scenario

Advantage:
- Simple, straightforward

Disadvantage:
- Doesn’t give you a good idea of the effect of each factor (what you call modes in your case) on the reliability
- Results are limited to the scenarios (profile) you tested under


2-Test under the different conditions you mentioned (or other ones), collect the data and analyze all the data together using the ALTA software. This type of analysis will give a general reliability model that relates reliability to the different factors (modes)

Advantage:
- Gives you a good idea of the effect of each factor (modes) on the reliability
- Provides you with a general model that you later use to plug in any combination of the factors (modes). This is especially useful after you get a better idea of your customer usage profile and don’t want to repeat the test (although I would recommend that you do repeat the test)

Disadvantage:
- More complex, needs some understanding of the failure behavior and you need to choose a life-stress relationship (in your case, most likely you will be using either the General Log-linear model or the proportional hazard model)

[tarik]

I have an additional comment to make, especially if you are going to follow the first approach. Note: This applies in case the testing under each mode doesn’t happen simultaneously.

When you are performing a test, make sure that you either switch between modes randomly or that you distribute the modes equally, that way you are not giving one mode even more chances to happen over the other. For example don’t complete all the cycles under mode 1, then switch to mode 2.

For example, if you are doing a 80/20 test , then either randomly test under mode1 and mode2 (mix them up) or test under mode 1 for 4 cycles then test under mode2 for 1 cycle, then test under mode 1 for 4 cycles then test under mode 2 for 1 cycle, that way you equally distribute the test.

On the other hand, if the ‘life’ accumulated under one mode doesn’t affect the reliability of the product when it’s tested under the other mode, then the way you do the test doesn’t matter ( you can do all the test cycles under mode 1 then switch to mode 2)

[GuruCharan]

I am new to reliability engineering. I would like to understand practical uses of NOT logic in fault trees. Can you please help me in finding some good/simple references or examples.

[Tim]

Thanks for the help!

I would rather use the ALTA method because I will never know the actual customer usage.

With this I could make the statement that at Profile "X" (80 / 20 ) the system meets R/C 90/90. and at Profile "Y" (95 / 5) you meet RC 90 / 50. This is much more powerful.

In order to do this analysis (ALTA) do all the failure modes need to be the same?

...Or is it only important that mode 1 testing creates failure A and mode 2 testing creates failure B.

This is a rather complex system and I can only imagine all the failure modes that will rear their ugly heads!!!

[tarik]

If you chose that approach, you can't do competing failure modes analysis. With that approach, a failure is considered a failure regardless of the reason. This is a disadvantage of this approach, the other approach (first one) would allow you to study different failure modes (meaning categorize failures by different causes) and do a competing failure modes analysis.

[Tim]

Thanks for the help...

One more possibility / Question

If we run a sample to failure during a 50 /50 test and the total cycles to reach a given R/C is 10K that means that Mode 1 will see 5K cycles and Mode 2 Will see 5K cycles....but because we are running to failure (assuming failure is later that 10K) we should be able to infer the same R/C value at a different level say 90/10...80/20...70/30...60/40
Depending on how many cycles are completed...Right?? you can assume you are good at 60/40 when you reach 12K?
70/30 when you reach 14K?
80/20 when you reach 16K?
90/10 when oyu reach 18K?

[tarik]

I don’t think I agree with you.
Here is how I understand it, let me know if I’m wrong.

Let’s say we decide to do the 50/50 profile and you test till failure.
Let’s say you tested a unit and you had a failure at 12000 cycles (which is equivalent to 6000 cycles on the front and 6000 cycles on the rear). And that you tested other units and you had failures at 12500, 13000, 14000, 9000

Now in ALTA in order to get meaningful results you need multiple stresses (or profiles as in your case). So let’s you take another 5 units and you test them under a 80/20 profile. Let’s say your failure times were: 4500, 6500, 7000, 7600, 8000

Also you take another 5 units and you test them under a 20/80 profile. Let’s say your failure times were: 14000, 15000, 16000, 16500, 17000

Note:
-It’s preferable to have more stress levels (profiles), in order to keep this short, I am going to use only the profiles I just presented
- You will need to choose an appropriate life-stress relationship

When you enter this data into ALTA, it will look as follows:

TTF---F----R
14000 20 80
15000 20 80
16000 20 80
16500 20 80
17000 20 80
9000 50 50
12000 50 50
12500 50 50
13000 50 50
14000 50 50
4500 80 20
6500 80 20
7000 80 20
7600 80 20
8000 80 20
-------------------

TTF: time to failure
F: Front
R: Rear

Do you see now how this works? You can add more stress levels(profiles) in the same fashion to get better accuracy .

[tarik]

I will email you an ALTA file that will help better see what I am talking about

[Tim]

I think we may have a slight misunderstanding... The last post I had was not using the ALTA method. I was simply looking at # of completed cycles. This situation is somewhat unique because the Strees Level is the % of cycles to the front or rear. This means that the longer you run the part (assuming running to failure) the higher stress level you reach.

A 50/50 test that fails at 16K cycles is >=
a 80/20 test to 10K cycles
a 20/80 test to 10K cycles

because in test 1 there are
8K Cycles Front and 8K cycle Rear

In test 2 there are
8K Cycles Front and 2 K cycles rear
Both tests have 8 K cycles on the front

In test 3 there are
8K Cycles Rear and 2 K cycles front

This statement can be made simply because you have completed the same # of Front and rear cycles in both tests.

The 80/20 test will be an advantage because you will see the failure sooner....which is the point of ALTA testing...

All I was saying was that there is a relationship between the # of cycles completed and the Stress Level because the stress level is this exact relationship.

I have a feeling that 8K front cycles during a 50/50 test are the same as the 8K front cycles during a 80/20 test the only difference being the numner of rear cycles during the test.

by the way I dont have the alta software so I couldn't read the file you sent me.

Hope this helps ...Thanks

[tarik]

In that case, I would recommend you do any test you want (50/50 would be better) and just obtain 2 separate reliability models, for each of the sides (front and rear), that relates reliability to number of cycles accumulated on the specific side.
Then once you find out what your profile looks like, you can plug the number of cycles expected to accumulate on the front side and the number of cycles expected to accumulate on the rear side within a certain period of interest (for example 1 year, life time or warranty period).

[Temanag]

Hello!

i will be very thankful if somebody have an idea and answer my question

i know that for a given failure destribution one can calculate the sample size which must be tested a defined time (test without failures-success run) to obtain for example 90% reliability with 95% convidence at defined time.

well... if one start to test the samples and a failure occur before the end of the required time...then he can use the binomial formel one more time and calculate how many samples are needed additionally to obtain the same reliability and confidence level, and then continue the test.

ok if a failure occur one more time what shoud i do now
--> i can use the formula one more time i hope that this time all my sample units survive
--> or i can say- no! i must improve my product

is there a possibility using weibull++ to decide what to do in such cases?

if not, what should i do

thanks

[David]

Hi Tim,

You could try and add even more units so that you can try to pass the test. But keep in mind that you originally designed the test for zero failures and that test failed. Then you redesigned the test for a single failure and now that test has failed. At this point, there is not much, if anything, to gain by throwing more units into the test in the hopes that it will pass. You would most likely be wasting more time and money.

Obviously you could make some design change(s) and redo the initial test. Another suggestion would be to continue the test with the units you have and test some or all of the units to failure. Then conduct a life data analysis on the data. For example you could fit a Weibull distribution to the data. Conducting a life data analysis would at least give you an idea as to where you are, even though it may not be where you want to be. Weibull++ (http://www.reliasoft.com/Weibull/features.htm) can be used to conduct the life data analysis. You can conduct a failure analysis to help determine what changes to make so that you can meet or get closer your goal. If you are dealing with a repairable system, you could conduct a reliability growth analysis using RGA (http://www.reliasoft.com/RG/features.htm).

I hope this helps.