Use ML/AI To Improve Decision Making

Benefits of increased use of ML/AI in reliability and maintenance include better insights into finding failure modes and recognizing pending asset failures earlier.

Q: Is RCM the final answer? What’s next: ML/AI-driven RCM?

A: By definition from John Moubray (Reliability-Centered Maintenance (2nd ed.), Industrial Press Inc., NY): “Reliability-Centered Maintenance is a process used to determine what must be done to ensure that any physical asset continues to do what its users want it to do in its present operating context.” (ISBN: 0-8311-3078-4)1

“Mantenimiento Centrado en Confiabilidad: un proceso utilizado para determiner que se debe hacer para asegurar que cualquier active fisico continue hacienda lo que sus usuarios quieren que haga en su context operacional actual.” (ISBN: 09539603-2-3)

I also have the definition in German and Chinese.

RCM is a global concept that has passed the test of time. It enables a logical methodology to implement a company-appropriate and cost-effective strategy for assets. It will not solve all reliability problems. All failures will not be eliminated. Remember that the goal is not just to maintain it better, but to see if the maintenance (reason for failure) can be eliminated. RCM is an ongoing process, not an event.

RCM is rarely, if ever, applied to all assets. It’s understood that most failures are random, so a time-based approach should not be a dominant strategy. Similar equipment may need different maintenance. You cannot maintain your way to better reliability. The asset is limited by the inherent reliability of the design/installation. More maintenance is probably not better. Allow select failures based on consequence and return on investment. RCM moves the thinking and decision making to a more comprehensive view and an economically feasible way of protecting asset functions at the system level. If any of these statements bother you, then you may not fully understand reliability-centered maintenance.

Many companies don’t have the patience to implement RCM or it doesn’t survive changes in management over the five-plus years needed to properly transition and instill a sustaining process and mindset.

The six curves (patterns) that resulted from the original RCM study by Nolan and Heap (published by the Office of Assistant Secretary of Defense, 1978) changed the paradigm of what’s important in applying maintenance. It’s been proven to work in military and the aerospace industry where there are robust adhered-to procedures. In industry, the RCM process is more tedious in application, often due to culture.

The ongoing question is what is the optimal inspection time in the asset-degradation spectrum?

RCM has been around for more than 40 years and albeit a few variations in implementation, the process still stands. I’ve not seen any new theories that are better or newer. The big question is, do the six curves explain everything that we need to know? Are we asking the correct questions? We don’t know what we don’t know. The learning increase that comes from condition monitoring and machine learning/artificial intelligence (ML/AI) will reveal some additional knowledge to take us to a still higher level of understanding of failure modes and RCM. As industry collects not just more data, but complete and accurate data, additional insights will be revealed.

Some of the potential benefits of increased use of ML/AI in reliability and maintenance include:

• Better insights on finding failure modes

• Recognizing pending asset failures earlier than condition monitoring alone

• More focused root-cause analysis and elimination

• Optimized planning and scheduling (coordination of people and parts)

• Increased confidence in reducing spare-parts inventory

• Greater access to historical data and cross-referencing collective knowledge (such as failure mode and effects analysis, maintenance-management system, production data)

• Better able to evaluate and make risk-based decisions

• Enable prescriptive maintenance (system evaluates data and generates work orders to minimize operation risk)

• Create overall understanding of equipment failures, downtime, repair time, and asset criticality

• Ability (over time) to transition more people to do root-cause analysis and elimination and other problem solving/long-term thinking practices

• Better coordinate (optimize) when maintenance should get done to minimize operational disruptions.

• ML/AI enables the analysis of big data to identify patterns that provide insights for improved decision making.

When you look at the generic P-F asset-degradation curve illustrated here, the struggle is always to find the optimal inspection time. The typical starting point is using an inspection interval about half the P-F interval, getting team-member and historical input, original-equipment-manufacturer recommendation, and other information. Too early/too often (point A) and you are wasting resources and reintroducing infant mortality to the component or asset. Too late (point C) and you risk a costly breakdown. Better data analysis with ML/AI can help optimize the inspection interval (point B). In addition, proper applications of ML/AI can detect potential issues earlier than predictive technologies and condition-based monitoring alone. Your collective knowledge from all data analysis and practical experience should be used to design-in better machinery and equipment to eliminate or minimize the consequence of failure.

Today, computing power and data storage are becoming more cost effective. What’s often challenging is getting good, meaningful, clean data. By that I mean data that is complete, accurate, and in the correct format. This is something that can be improved on now, so you have a database with which you are willing to make decisions as you transition to more digital-driven practices. ML/AI-driven RCM is the future and will improve the reliability of industrial operations. EP

Based in Knoxville, Dr. Klaus M. Blache is director of the Reliability & Maintainability Center at the Univ. of Tennessee, and a research professor in the College of Engineering. Contact him at kblache@utk.edu.