FACT BOX
The advantages of the CALT process:
• We have decoupled sample size and time from the reliability requirement.
• Test time is flexible and is always managed by our selection of the third stress level.
• A Weibull plot is the result from which we make our decision of acceptability.
• The missing accelerated test parameter is acquired as a by-product of the validation process. This represents “Profound Knowledge” as Deming would say.
About the columnist
Larry Edson retired from General Motors as a technical Fellow for Reliability and Test in 2008. He now uses his 40+ years of experience to assist alternative energy companies in the field of reliability and accelerated testing.
Dr. W. Edwards Deming pointed out that the use of slogans and motivational speeches tend to frustrate and infuriate engineers rather than produce real change. His famous words were: “By What Method” will you make your wants come true? If you want your product to be reliable then you must have a reliability program plan to make that happen. Wishing or demanding will have little effect.
By columnist Larry Edson
Part one of this trilogy began with a discussion of the culture in a company, a culture that must foster the values of quality and reliability for “all the people all the time” (OceanWise, Nov. 2011). Part two described the process for developing a good reliability program plan with emphasis on marrying the process into the culture (OceanWise, April 2012). In part three we now look at one testing method that has proven very effective across many different industries, but first, a quick review.
The Reliability Program Strategy and Plan - We have five opportunities to affect reliability and it is essential that we use all of them:
The Reliability Program Plan shows how we assemble all of our opportunities into a process; our version of “By What Method”:
Testing To Demonstrate Reliability - Testing must be accelerated. We use the Arrhenius relationship to provide correlation for chemical reactions or diffusion phenomena, and we use the Inverse Power relationship for most other cumulative damage phenomena such as mechanical stress and thermal cycling. These relationships require either an activation energy value (Ea) or a fatigue exponent (m) in order to provide valid correlation of the accelerated test to real life. Many of these values (Ea and m) are well established for common materials and failure mechanisms:
The underlying principle in forming the accelerated test is to have the damage on the accelerated test equal to the damage in real life. We then apply either “Success-Run” or “Test To Failure” statistics to this accelerated test to make a decision of acceptability.
What if we do not have a value for the activation energy (Ea) or the fatigue exponent (m)? The CALT method of testing (Calibrated Accelerated Life Testing) generates the missing value and allows us to transpose all of our accelerated test data to real life values, resulting in a Weibull plot. A Weibull plot represents “the gold standard” for the analysis of reliability data. CALT is the process of testing at different stress levels, all higher than normal, to define the life-stress relationship and allows us to transpose the resulting accelerated test data to values that reflect when the product would have failed at the normal stress level. We test at these elevated levels because we do not have time to test for one-life at the normal stress level.
Example: We are testing a structural element in the yaw system of a wind turbine. We wish to demonstrate a reliability of 99.99% with one life requiring 336 hours of testing at the normal stress level.
Referring to the figure above, the foolish limit of stress for our stressed element is 300 Mpa, the transition from elastic to plastic behavior, while the normal stress level is 100 Mpa. We test our first two samples to failure at 10% less than this foolish limit and plot these values as a distribution on Weibull paper (step 1). Then we test two more samples to failure at 20% less than this foolish limit and plot these values as a distribution on Weibull paper (step 2). Transferring these two distributions to log-log paper, we can then connect the characteristic life values of these two distributions and establish a rough approximation of the life-stress line. We identify how much time we have left to test a third set of two samples (step 3), and based upon this available time, we can identify the third stress level for testing (step 4). We test two more samples at this third stress level and plot these values on Weibull paper (step 5). Transferring this third distribution to our log-log paper, we now have three characteristic life values from our three Weibull distribution through which we can fit a line. The slope of this final line provides our missing fatigue exponent. We use this fatigue exponent to translate the failure times of our six samples from their elevated stress levels to the normal stress level. These translated values are plotted on Weibull paper (step 6) and the decision of acceptability is determined by comparison of the line to the requirement of 99.99% at one life of 336 hours.
The final Weibull plot is shown in the final figure: