Video was updated May 2024.
See how to:
- Understand the kinds of problems are addressed by reliability tools
- Problems where a product (or process) is repeatedly stressed, eventually leading to failures and stressor could be anything ( e.g., voltage cycles, mechanical loadings, Car washings, Start/stop cycles on a motor)
- Problems where a product (or process) is repeatedly stressed, leading to Degradation in performance (e.g., tire wear vs. mileage, HVAC efficiency over number of cycles, adhesion strength over time, drug shelf life)
- Consider issues addressed by Reliability Analysis
- When will it break?
- How long do certain employees stay and when will they leave?
- How long do patients survive based on medical treatments?
- What is the pattern for credit card churn?
- In Pharma, how do I set my expiration date?
- How long of a warranty should I offer customers?
- What is a typical repair time?
- How can I put pieces together to give best reliability of a system of parts?
- Understand censoring used in analysis
- Sometimes we know exactly when something fails
- Sometimes we end a test before all parts have failed and the "still alive" parts are called "Right Censored", or often just "Censored"
- Sometimes we check parts periodically during a test, find one that failed, and we don't know exactly when it failed (just that it failed before the inspection) and These are called "Left Censored"
- Sometimes we check parts periodically and know that at time T the part was functioning, but a T+1 the part had failed and this is called "Interval Censoring"
-
Planning the test - Explore the sample size, study length, or precision needed for a reliability life test
- Use Life Distribution to analyze reliability with or without multiple failure causes
- Determine Data Structure, which distribution to use, use other Profilers, other Failure Causes
- Interpret Event Plots and Parametric Estimates – Exponentia
- Profile Behaviors using Distribution, Quantile, Hazard and Density Profilers
- Use Sample Size Explorers to adjust study parameters and assumptions to explore different study designs to use for reliability life testing
- Perform Accelerated Life Testing for situations where it takes too long for a product (e.g., flourescent light bulbs) to fail so we can’t conduct a reasonable-length reliability test . Use accelerator (e.g., temperature), to run test units at accelerant conditions and extrapolate back to a "standard" condition
- Analyze and interpret results from one accelerator at a time using Fit Life by X.
- Consider using Fatigue Testing Models, new in JMP 18, for S/N data.
Questions and answers by Jerry Fish @JerryFish and Don McCormack @DonMcCormack that followed the live demo, plus Q&A from a previous session on this topic.
Q: Are we limited to two-parameter Weibull or is three-parameter an option?
A: The distributions beginning with TH are the three parameter distributions.
Q: When do we look at the profilers under Weibull?
A: When you are planning to use the Weibull distribution. The controls at the top of the report only apply to the graph. To use a profiler for a specific distribution, you need to use the fit under the appropriate outline box.
Q: What does the relative width indicate?
A: Relative to the point at which the estimate is made.
Q: If our test time is actually a time period, and not a number of cycles, will the units always be hours? Does JMP always interpret it that way?
A: Think of the data as a usage value. It doesn’t matter if it’s time, cycles, miles, etc. In the same respect, the usage units only matter when it comes to interpretation.
Q: For the degradation prediction is each device considered totally separately or is there a common model for all the devices
A: You have the flexibility of looking at either common or separate slopes and/or intercepts.
Q: Can you fix the slope in Life Distributions when you have several datasets (groups)?
A: I’m not sure what you mean by slope, you’re fitting a failure distribution. That said, you can fix any/all of the parameters of the distribution you choose. If you mean you are testing for differences between groups, you might want to use Fit Life by X where X is the group.
Q: The setup he showed appeared to work for only right censored data. How do we setup for left censoring and a mix of right and left?
A: If you have mixed censoring, you set the data up with two columns. The first column indicates the start of the period, the second the end. For right censored data, the right column is set to missing. Interval censoring has a value for both columns.
Q: Does this analysis apply to destructive samples, like solids that are stored in different temperatures and need to be reconstituted with a buffer to be tested?
A: In this case, we’re dealing with true failures, i.e., when something stops functioning. When we’re using a measurement as a stand-in for failure (e.g., when a certain %age of degradant forms) then we would use the Degradation, Destructive Degradation, or Repeated Measurements Degradation platforms, depending on when one (Degradation/Destructive Degradation) or multiple (Degradation/Repeated Measurements Degradation) are made.
Q: Can you briefly explain TH and DS?
A: TH stands for threshold. These are the three parameter models such as the three parameter Weibull. DS is defective subpopulation. In that case, there is an assumption that there is a mixture of early failures items and items with a considerably longer life. Using those distributions will give you a single distribution with a parameter that indicates the proportion of early failures.
Q: Can the Fit Life by X platform handler two accelerator conditions, like acceleration due to Temperature and acceleration due to Voltage, at the same time?
A: You can use Fit Parametric Survival. We will present an entire session covering Accelerated Life Tests on August 25, 2023. Consider attending that Mastering JMP session .
Q: Is it appropriate to use reliability in pharmaceutical for an accelerated stability study, using proteins?
A: First of all, there is a stability platform in JMP. If we're not talking about accelerated testing, you can do shelf life calculations using JMP Degradation capabilities. We have heard of people using the Arrhenius Equation for proteins, when you are in a sense baking them or cranking up the heat on them until they start to denature. I don't know whether or not there is FDA or UPC guidance in terms of which models, to use. Life by X Arrhenius equation might be one way of fitting those types of models also keep in mind that if you're measuring something like amount of unraveling of the protein, it will put you in the area of degradation analysis as opposed to Life by X, which is looking for a true failure.
Q: What does ‘Censor’ indicate in the Life Distribution Analysis?
A: Censor indicates the object is still functioning when last observed.
Q: What will change if we select a different censor code?
A: When you select a different censor code, you change the observations that correspond to censored observations.
Q: For Life Distribution, why is the hazard high at low kilocycles?
A: Hazard is the ratio of the cumulative failures to the probability of failure. I suspect it is large because the probability (i.e., the denominator) is very small.
Q: I see a tab that says "Acceleration Factor" as part of Accelerated Life Test. Does JMP calculate your acceleration factor for you based on experimental data!? If so, that's AMAZING!
A: Yes, it is based on the data and the assumptions you make about the acceleration factor (e.g., it is Arhhenius).
Q: Would the 'X' section be used to do multivariate degradation?
A: In the platforms that take multiple Xs, you can model a (single) response with multiple predictors. There are certain reliability platforms that allow you to do that. Fit Parametric is the most general of them.
Q: How do we get to the Stability Studies analysis?
A: It’s under Degradation (Analyze>Reliability and Stability>Degradation), the third tab in the initial dialog.
Q: Is there a way to determine if a new form of degradation is kicking in at higher temps?
A: Possibly. In these cases, you may be dealing with mixtures. Life Distribution lets you fit mixtures.
Q: If your equipment only has discrete settings, then that would be the correct way to do the analysis, right?? 14:49
A: Yes, that is one way to do the analysis.
Q: Is it possible to do Nevada Chart Analysis in JMP?
A: Yes. In Reliability Forecast, the Reliability Forecast red triangle menu contains the option Save Data in Time to Event Format, which saves Nevada or Dates data in a Time to Event formatted table.
Q: Regarding Confidence Interval, I know JMP 16 can change the CI in the life distribution and in the Fit Life by X. Is there a way to change the CI in the Parametric survival fit? The default is 95%, some cases, I might want to change it to other values, for example, 75%. I don't see this option in Fit Life by X.
A: In Fit Parametric Survival, you can change your Alpha level from the screen where you specify variables for analysis. See the main red triangle.
Resources:
- Applied Reliability by Paul A. Tobias and David Trindade
- Statistical Methods for Reliability Data by William Q. Meeker, Luis A. Escobar
-
More about Censoring
-
Documentation and examples of Life Distributions
-
JMP reliability capabilities overview
- Documentation and examples of Accelerated Life Testing
- Peng Liu's JMP 18 Developer Tutorial on Fatigue Test Modeling.
- Mastering JMP video, Selecting Proper Sample Size for Your Designed Experiment