Test equivalence

[vnigam]

I have a durability test profile at a set of pressure levels for mechanical components and the cycles to which the test needs to be conducted at each of the stress (pressure levels). If I conduct this test it takes a very long time therefore we are trying to compare another profile which tests the parts at different set of stress levels and cycles. We want to study the co-relation/equvalence of the two profiles.

My thought is to test a sample of parts using the two profiles and the set of stress levels and compare the distributions (time to failure data) and establish an equivalence by comparaing the data sets using contuor plots for a given confidence level. If the data sets are not from the same population I can try and look deeper to see if there is a co-relation or study the accelaration factor etc. to determine the relationship of the two tests.

Do my thoughts make sense? Please advise if you have some suggestions.

Thanks

[vnigam]

Can someone from Reliasoft respond to this thread I started? or could youh provide a phone number I could call?

Thanks and hoping for a response.

VN:confused:

[David]

Hi vnigam,

I apologize for the delay in answering your question. Thank you for your patience. :)

In regards to your situation, what you describe make sense, but what specifically are you trying to compare between the two profiles? I am not aware of a method that allows you to compare contour plots for two accelerated data sets, each with multiple stress levels. One thing that you could do is to compare the value of Beta to see if the values are similar. Another suggestion is to conduct a failure analysis to determine the failure modes that are generated with the new profile. This would allow you to determine if the failure modes generated with the new profile are different than the ones with the original profile. In addition, you could conduct an analysis using ALTA (http://alta.reliasoft.com) to determine the acceleration factor between the old profile and the new one. This involves using indicator variables with the General Log-Linear model. An example of how to conduct this can be found at http://www.weibull.com/hotwire/issue54/hottopics54.htm. This article is based on determining the connection between in-house and field data, but the same principles can be applied to your situation with the original profile and the new one. Another option would be to conduct a tests of comparison. This can also be done in ALTA. Information on the tests of comparison can be found at http://www.weibull.com/AccelTestWeb/tests_of_comparison.htm.

Sorry to be a bit long-winded but I was trying to cover the different options that are available depending on what you are trying to compare and what your concerns are.

I hope this helps.

[vnigam]

Thanks David. You do have valuable suggestions. I will go through the material you have suggested and see what can be done.

regards,

VN