cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Practice JMP using these webinar videos and resources. We hold live Mastering JMP Zoom webinars with Q&A most Fridays at 2 pm US Eastern Time.See the list and register. Local-language live Zoom webinars occur in the UK, Western Europe and Asia. See your country jmp.com/mastering site.

Choose Language Hide Translation Bar
Analyzing Reliability for Repairable Systems

See how to:

  • Understand the kinds of problems addressed by reliability tools for repairable systems
    • Problems where a product (or process) breaks and will be fixed
    • Problems where we are Interested in how often something breaks or how much will we spend (over period of time or product life) to fix it
  •  Understand types of analysies people might try to use for repairable systems (like comuputer board failures) 
    • Life Distribution ilooks at time between repairs, but his could  overestimate the failure rate of the computer late in life by measuring time to fail of the boards, not of the computer.
  • Use Recurrence Analysis to understand systems that are  repaired and placed back into service.
    • Examine mean time between failures for serial failures, to examine mean time between failures or whether one system is more reliable than another (eg., car engine repairs)
    • Uses  mean Cumulative Failure curve to give info on the System repair rate.
    • Understand that time between failures can be misleading if failures are not stable around the man
    • Analyze recurring problems, like a disease (e.g. tumors) reoccurring after patient treatments for the disease
  • Use Reliability Growth to understand how reliability changes over time on a single repairable system, perhaps as improvements are incorporated into its design, and whether changes made improve reliability
    • Determine if changes made actually improve reliability
    • Use model options Crow AMSAA, Crow AMSAA with Modified MLE, Fixed Parameter Crow AMSAA, Piecewise Weibull NHPP, Reinitialized Weibull NHPP and Piecewise Weibull NHPP Change Point Detection
  • Understand the types of problems addressed by Reliability Forecast
    • Major, month-to-month changes in a product's reliability
    • Portion of my production expected to fail within the first x months of its life
    • Time monthly warranty claims will peak
    • Cumulative cost of warranty claims for a product
    • Can a repair team keep up with returns or if more staff are needed
  • Build reliability flow using Reliability Block Diagrams (JMP Pro) using Analyze/Reliability&Survival/Reliability Block Diagrams
    • See an example of the final product 
  • Model and interpret Repairable Systems Simulation (JMP Pro)
    • Build your model  using Analyze/Reliability&Survival/Repairable Systems Simulation
    • See an example of the final product 

Recurrence Analysis.jpg

 

Reliability GrowthReliability Growth

 

Reliability ForecastReliability Forecast

 

 

 

 

 

 

 

 

 

 

 

 

 

Questions answered by Don McCormack @DonMcCormack and Jerry fish @JerryFish at the live webinar or previous live webinars on this topic.

 

Q:  Can you use $ amounts in the Cost column?

A: Yes,  A cost for the event (the cost of the repair, replacement, or adjustment). When costs are specified, the MCF is a mean cumulative cost per unit as a function of age and the markers in the Event Plot are sized by the cost values.  Note: Cost indicators for Recurrence Analysis are the reverse of censor indicators seen in Life Distribution or Survival Analysis. For the cost variable, the value of 1 indicates an event, such as repair; the value of 0 indicates that the unit is no longer in service. For the censor variable, the value of 1 indicates censored values, and the value of 0 indicates the event or failure of the unit (non-censored value).

 

Q:  Why is the term cumulative used?

A: Because where looking at failure counts.  See how JMP calculates cumulative

https://www.jmp.com/support/help/en/17.1/#page/jmp/fit-model.shtml describes how we calculate cumulative, and I will ask Don to explain the why verbally, also.

 

Q: Are reliability measures typically done at the product or individual part level?

A: It could be either. Often, you need the component measures to get to the system level.

 

Q:  Are reliabliity measures typically done at the product or individual part level?

A:  It could be either. Often, you need the component measures to get to the system level.

 

Q: Would these techniques be applicable to repairable systems that are in a standby condition?

A:   It depends on what the most reasonable measure of usage is. In some cases, an idle state would not be considered part of usage. Some situations, it may be considered in use.

 

Q: How do you reconsile the difference between noon-repairable analysis component based and ‘system’ repair analysis?

A:  The component life times would be modeled using the techniques we talked about for Non-Repairable Systems  (e.g., Life Distribution). To get at the system reliability from the components, you could use a tool like the Reliability Block Diagram (JMP Pro).

 

Q: Can recurrance analysis be expressed as Discrete Time Survival analysis? The "censor" variable would be treated as dependent variable in DTS. Does JMP have DTS capability/feature?

A:  Not sure what you mean by DTS. In the sense that censored data is grouped for the purpose of modeing the cumulative distribution (e.g., interval censoring) it may be considered discrete.

 

Q: Can we calculate/plot the rate of occurrence of failures within the recurrence platform?

A:  We have been showing the non-parametric approach. There is also the option of using a parametric model in Recurrance.

 

Q: Would these techniques be applicable to repairable systems that are in a standby condition (not functioning) but are expected to function when needed?

A: Ultimately it gets down to what the most reasonable measure of usage is. In some cases, an idle state would not be considered part of usage. In other situations, it may be considered in use.

 

Q: How do you reconcile the difference between non-repairable analysis ‘component-based’ and ‘system’ repair analysis?

A: The component lifetimes would be modeled using the techniques we talked about last week (e.g., Life Distribution). To get at the system reliability from the components, you could use a tool like the Reliability Block Diagram (JMP Pro).

 

Q: Re: Using Life Distribution, if we assume that this is the only failure mode of the system, should Life Distribution and Recurrence Analysis methods be the same?

A: Non repairable systems only fail once and have a time span starting at 0. In repairable system reliability, you more than likely need to consider the previous failure time, since the probably of failing (the next time) will be a function of how long the system has been operating.

 

Q: Will you please review the details on the Reliability Growth example?

A: See below.

 

 

 

Q: Can we calculate/plot the rate of occurrence of failures within the recurrence platform?

A: Yes. Jerry has been showing the non-parametric approach. There is also the option of using a parametric model in Recurrence Analysis.

 

Q: Can I use Recurrence Analysis for a group of components that are repairable, but not in a system, e.g., a traction motor on multiple trains or a coolant pump in a power plant?

A: Yes. Assuming that they are from the same population.

 

Q: Are the failures the same type for the Reliability Growth example or just any failure?

A: Any failure.

 

Q: If there are delayed fixes or fixes which are planned in next version (Improved or final) how can I accommodate those in Reliability Growth?

A: The phases in Reliability Growth assume that some change was made making the entity in the previous phase different than the one in the current phase.

 

Q: Can Reliability Forecast be done for repairable devices?

A: It is more often used for non-repairable devices.  However, it can be used when you assume the failure is either the first failure of the system or that after the system is repaired,  it is repaired as good as new. Typically, the distributions and assumptions are different for non-repairable and repairable devices. In his example, Jerry did include cost of repair of first failure.

 

Resources

Comments

this question came up during the session. If you don't have the 3 fails for Weibull on the repairable systems platform, how would you proceed? Also, are other fail distributions available for the repairable system distributions (i.e., lognormal/degradation)? Russ, @russ_r_mcdonald  Don McCormak @DonMcCormack said he would look into this with our developer team @peng_liu to see if Wei-Bayes is available for repairable systems.  

peng_liu

In JMP, the following platforms are related to repairable systems: (1) Recurrence Analysis, (2) Reliability Growth, (3) Repairable Systems Simulation.

I guess that we are talking about (1) and (2).

In this context, we use the word "process" not distribution. So both platforms can model a so-called Weibull Process, a.k.a. Power Law NHPP (Recurrence Analysis calls it Power NHPP).

The number of failures in the data may pose challenges to analysis. In Reliability Growth, if the platform complains about "cannot start", lack of data might be a reason.

There are no other "distributions", but other "processes". In Recurrence Analyses, there are 4 processes: Fit Model in Recurrence Analysis. In Reliability Growth, there is Crow-AMSAA (aka Weibull Process, aka Power Law NHPP) and its variations.

 

Recommended Articles