Bad Actors in 2020

As we enter a new year, we can all take stock of our success (or failures) in the last year and begin to think about how to make 2020 even better. In the book "Failure Modes to Failure Codes" (John Reeve & Derek Burley), they mention that:

"40-60% of all maintenance costs are due to chronic failures"

In my own experience, I saw that

5% of pumps accounted for 50% of all pump failures!

Additionally, I noted that

75% of pumps did not see any failures in a 2-year period

The take-away here is that high maintenance costs and production losses are caused by a remarkably small number of assets. Solving this situation is not necessarily a matter of applying best practices (as 75% of assets are not your biggest concern).

The secret is to look at the special causes of your asset failures... and you can do this by understanding your failure patterns.

Many people familiar with reliability have likely heard of the 6 failure patterns identified in Reliability-Centered Maintenance. You might also know that the failure patterns actually represent a type of probability plot. Some other names for these failure patterns include conditional probability plot and hazard distribution function.  

If you recall the formal definition of reliability (below), you can see that these probability functions can define reliability quantitatively – and if you are only looking at the shape of the plots then it is also qualitative.  

As the nature of these failure patterns came to be understood, their importance in maintenance increased. Reliability-centered maintenance hinges on being able to answer the questions of whether specific maintenance actions are technically feasible (will they work to reduce risk?) and are the maintenance actions worth doing (will it lower the life cycle cost?). A qualitative and sometimes quantitative understanding of these patterns is generally required to answer these questions (though experienced practitioners may be able to use their own substantial experience to infer them). 

There are many ways of estimating these probability plots, and in the original Nowlan & Heap report (1978), you can see that it was a very manual process. Setting up these types of actuarial analyses was labor intensive, and subject to considerable subjectivity by the analyst (such as how to set up the age intervals for calculation). 

It was precisely for this reason that Nowlan & Heap included several decision diagrams to help minimize the need of performing such calculations (despite their report relying heavily on the insights from a vast number of these calculations).

It is easy to see why Nowlan & Heap wanted to reduce the number of times actuarial (life data) analysis needed to be performed in their time. However, with today’s data availability in typical process industry environments, most plants and refineries easily have 5-20 years of electronic actuarial data; many having converted to computerized maintenance management systems or CMMS’s in the last 10-20 years.

In addition to the greater availability of maintenance data, the actuarial analysis has also been greatly simplified. Gone are the days of using specialized plotting paper to visually estimate the approximate function (one of the main reasons Nowlan & Heap chose not use formal probability distributions in their analysis).

Adding up the better availability and accessibility of today’s plant maintenance data, along with advances in statistical curve fitting (parameter estimation) and versatility of the Weibull distribution function (all bundled inside 1 software package), it is orders of magnitudes easier today to do the type of analyses that Nowlan & Heap tried to minimize.

The world is very different today than it was 40 years… Our approach to failure patterns should be as well. Gone are the days of expecting a group of people in a room to brainstorm how they think an asset approaches failure. Richard Feynman may have put it best in regard to using “group-think” in place of real-world observations and data:

In conclusion, whether you are developing a reliability strategy or a bad actor list for 2020, remember to keep failure patterns at the forefront. 

1.    Failure patterns (conditional probability plots) are the key to understanding an asset’s reliability.

2.    Failure patterns are required to understand if maintenance actions are feasible and worth doing (key questions in Reliability-Centered Maintenance).

3.    Your CMMS data is the logical starting point to figure out failure patterns in most facilities.

Alejandro Erives

Owner & Reliability Liaison - Blackstart Reliability LLC

Blackstart Reliability's Unique Services:

1. Bad Actor Asset Management & Analysis: "DataStart for Bad Actors"
2. Condition Monitoring Program & Maintenance Optimization: "DataStart for Predictive Maintenance"
3. OEM Services: "DataStart for OEM's"

In Blackstart Reliability's DataStart model, success is dependent on your input whether you are the end user, an OEM, or a product or application SME.