Win with Weibull

By Drew D. Troyer CRE, CMRP, Contributing Editor

The subject of Weibull analysis often evokes a surprisingly negative response from members of the plant-engineering community. Some complain Weibull is only for design engineering, i.e., not usable in the plant. Others insist that since nearly all failures are random, condition monitoring is the only tool required. Those perceptions have some merit, but overlook several important aspects of Weibull’s value in plants. Here, I’m beginning a conversation aimed at dispelling the myths. I’ll provide more detail in an upcoming enewsletter article. Visit the subscription section at efficientplantmag.com to subscribe to our newsletters.

Let’s start with a very simple definition of Weibull analysis: It’s a tool that helps evaluate the risk profile of operating a system over time. In the plant, we often talk in terms of Mean Time Between/To Failure (MTBF/MTTF). Alas, “average” can be a very dangerous number to know. Weibull analysis enables us to determine whether the risk of failure is:

• decreasing over time, which indicates early-life failures

• constant over time, which indicates random failures

• increasing over time, which indicates wearout failures.

The beta (β) shape is the key parameter that provides the risk-profile information. If β is approximately 1.0, the failure pattern is random. If β is less than 1.0, the failure pattern is early life (the closer to zero, the stronger the effect). If β is greater than 1.0, the risk of failure increases over time, i.e., the higher the number the greater the effect. Wouldn’t it be helpful for a plant reliability engineer to know that risk pattern? 

But back to the criticisms: Let’s first address the theory of Weibull being useless, given the randomness of most plant failures. To be clear: No statistical technique can predict when an individual machine or component will fail. Statistics provide information about populations. Thus, we don’t use Weibull to evaluate the health of individual machines.

Next, the reason plant failures look random is because we normally collect data at the system level. Plant machines are complex dynamic or static systems, with multiple components and failure modes.

Parsing the individual failure modes in a complex system helps solve the ‘random trap.’

When all failure modes are thrown together, there’s a randomizing effect following the run-in period (see figure). The system-failure rate, however, is the cumulative sum of all the various failure modes combined. When we untangle those different failure modes with effective reliability-data collection (the subject of a future article), the picture—and the utility—of Weibull analysis in a plant becomes clearer.

My e-newsletter article will illustrate these concepts quantitatively using four risk-profile scenarios, each with the same MTBF/MTTF. For now, go ahead and give Weibull analysis another look—and another chance. EP

This column explores, from a management perspective, the Five Pillars of Knowledge, as defined by the Society for Maintenance and Reliability Professionals (SMRP), Atlanta (smrp.org). Expanded, more technical, takes on topics covered here will appear in Efficient Plant’s monthly “Reliability Solutions” e-newsletters.

Based in Tulsa, OK, industry veteran Drew Troyer is principal with Sigma Reliability Solutions. Email
Drew.Troyer@sigma-reliability.com.