Students are introduced to Weibull analysis from a high (graphical) perspective, where the concept of being able to easily ‘pull’ information from reliability data is introduced.

The first step in doing data analysis (or any other activity) well is to understand ‘why’ you are doing it. This lesson shows students many of the business and individual benefits that can be achieved from Weibull analysis.

Failure is a random process, and reliability is one way to characterize this process. Reliability is also one of the measures we use to understand the way failure affects business and mission outcomes. This means reliability is subject to requirements and specifications. This lesson outlines the relationship between reliability and the (random) failure process.

Students will build on the previous lesson and learn basic statistical concepts like the ‘probability density function (PDF)’ and ‘cumulative distribution function (CDF)’ that help us describe (and then interrogate) random processes.

The exponential distribution is one of the most (over) used probability distributions when it comes to reliability engineering. It has some useful statistical underpinnings, but it models products and systems that never ‘wear-in or wear-out.’ Students will learn about the ‘constant hazard rate’ and what that means for reliability engineering.

The ‘bell curve’ is one of the most well-known phrases in probability and statistics, but not many people know where this very special shape comes from. Students will learn about where the ‘bell curve’ or ‘normal distribution’ comes from, and when to apply it to certain reliability scenarios that involve wear-out.

The ‘lognormal distribution’ is related to the ‘normal distribution’ and is great for modeling things that accumulate damage in a multiplicative way. This models a whole new range of failure processes, especially those that involve ‘fatigue.’

The ‘Weibull distribution’ can very accurately mimic the probability distributions taught in the previous lessons and more. This means that one distribution can model a vast range of failure mechanisms that include different types of wear-out and wear-in.

‘Probability plotting’ is an intuitive, entirely graphical way of trying to ‘fit’ one of the distributions discussed above to data. ‘Weibull plotting’ is a particularly simple way of finding which Weibull distribution best describes the random process you are examining. Students will follow Weibull plotting for a wireless modem router’s failure data.

Students will practice Weibull plotting by hand. This is the best way to truly understand how software (that they will later rely on) interrogates the data, meaning they are more comfortable understanding what Weibull plotting software ‘spits out’ at them.

Students will repeat the previous manual plotting exercise with the Acuitas Weibull Plotting Tool and start to become familiar with it.

Not all data creates the ‘straight’ line on a Weibull plot that is needed to conclude that a particular Weibull distribution could describe the observed failure data. Different curves and shapes are signatures of different failure scenarios. Students will learn to look for and deduce key failure characteristics to help inform decision-making.

Students will use the Acuitas Weibull Plotting Tool to plot data that contains the behaviours outlined in the previous lesson. They will then learn how to spot them in ‘real data.’

This lesson examines system failure data (where failure is caused by any one of a number of components), and how Weibull analysis can help identify which components are wearing in, wearing out, and doing other things that can inform engineering management decisions.

Students will use the Acuitas Weibull Plotting Tool to learn how to ‘isolate’ component failure data from system-level failure data to understand (for example) which components need to be replaced, upgraded, better installed, and so on.

The ‘line of best fit’ is a ‘best guess’ at what we think is happening in the random process we are observing. There is always uncertainty that this is not the ‘true’ line.

Students will use the Acuitas Weibull Plotting Tool to create confidence bounds on lines of best fit for data, allowing them to gain (for example) upper and lower estimates of warranty period reliability or service life.

The ‘3-parameter Weibull distribution’ incorporates a ‘failure-free period’ that can be very dangerous to include in reliability life models. However, there are plenty of failure mechanisms that take time to create enough damage to be all but impossible to create early failure. Students will learn when and if the ‘3-parameter Weibull distribution’ can work for specific decision scenarios.

Students will use the Acuitas Weibull Plotting Tool to conduct a ‘3-parameter Weibull distribution analysis’ and determine things like installation time for incomplete data.

Many parts (and their failure mechanisms) have very specific ‘failure signatures.’ Weibull distributions can be ‘programmed’ to behave in a way that corresponds with these signatures. This allows (for example) the ‘failure signature’ of a ball bearing to be defined upfront, meaning that less data is needed for testing to obtain meaningful results.

The final worked exercise will allow students to conduct ‘WeiBayes analysis’ along with other revision exercises, along with any data students have from their place of work.

Students will be introduced to incorporating ‘accelerated life testing (ALT)’ into Weibull analysis (and vice versa). ALT involves the careful increase of stresses to quantifiably increase the amount of damage a part or component accumulates. This means (for example) 10 years of actual use can be compressed into 2 weeks of testing – as long as the underlying failure mechanisms are known.

Final questions will be answered, along with a discussion on the pros and cons of Weibull Analysis.