Machine Learning To Play Key RCM Role

Machine learning and artificial intelligence will play a significant role in taking your reliability centered maintenance program to the next level.

By Dr. Klaus M. Blache, Univ. of Tennessee Reliability and Maintainability Center (RMC) and Rajiv Anand, CEO, Quartic.ai

With so much of what we need to fix attributed to random failure, more sophisticated tools and technologies are needed to enable the best opportunity for early detection and intervention. Understanding this, the quest for predictive maintenance began decades ago but overall progress has been limited. Rajiv Anand, CEO of Quartic.ai, comments on the current state and benefits of machine learning.

The first leg of the journey was to move away from time-based maintenance to condition-based monitoring (CBM). This journey started with route-based condition data collection and then moved to using wired online sensors. Today, the availability of inexpensive wireless sensors is rapidly increasing access to more condition data. 

The next step is implementing reliability centered maintenance (RCM), an approach that Nowlan and Heap started for the aircraft industry, and was later enriched, and “industrialized” by John Moubray. It’s in this step that machine learning will play a significant role.

Let’s first look at it from an RCM perspective. Two key tenets of RCM are risk and FMECA (failure mode effects and criticality analysis). The fundamental value proposition of RCM is that successful implementation will lead to a greater understanding of the level of risk that the organization is managing.     

Risk is defined as severity x occurrence x detectability. The purpose of collecting route-based or sensor-based condition data is to increase detectability. We want to be able to detect as many critical failure mechanisms as possible, as early as possible, and under all operating contexts of the equipment.

Analysis and interpretation of the condition data is what provides the detectability. This analysis is still largely performed by machinery health analysts. We owe a great debt of gratitude for the hard work they perform to keep our businesses profitable, safe, and sustainable. Unfortunately, there weren’t enough of them to begin with and not enough are entering the workforce.

Any machinery health analyst will speak to the nightmarish challenges of analyzing condition data with constantly changing baselines. Their life is becoming more complex and stressful as the amount of condition data increases and, with the demand for agility in supply chains, changing operating contexts become the norm.

We need to add more advanced tools to their toolkit and eliminate some of the mundane analytical tasks from their task list. That’s where machine learning (ML) and artificial intelligence (AI) come in.

Machine learning will improve detectability, making it easier to move further up the I-P-F curve toward initial failure onset.

ML and AI

In industry 3.0 our focus was on automating manual tasks. Among other things, the goals of Industry 4.0 are achieved by automating cognitive tasks. Machine learning, a key element of Industry 4.0, can be loosely defined as “automation of cognitive tasks.” As mentioned previously, increasing detectability with machinery health analysis is largely a cognitive task. Using machine learning to increase detectability is, therefore, a logical and natural step in the journey to predictive maintenance.

When we increase detectability with condition-based monitoring, whether it’s route- or sensor-based, it is always with the point “P” on the I-P-F curve. This is because condition-based measurements only allow us to go that far up the curve. Can we go farther up? Of course we can and, if we are to achieve the true objectives of RCM, we must.

The basic definition of reliability is a measure of the ability of an asset to perform its intended function. Any time that ability is degraded, the asset becomes unreliable. We can and must start measuring function degradation as early as possible. Machine learning increases detectability. Condition sensors (traditional CBM) can deliver operational context to CBM and establish baselines with an infinite set of rules and increased accuracy using multiple techniques (vibration + tribology, for example). ML can move the “P” farther up the curve using operational data. 

The fixation with “predicting” failures and the bold claims today of predicting failure x days/weeks/months in advance with machine learning are just frivolous—they do not accomplish anything for RCM or risk management. Measuring degradation of reliable function, the risk that is being accumulated, and the risk trajectory allow us to achieve RCM objectives, resulting in efficient, safe, reliable, and sustainable operations. 

Machine learning can be used by machinery-health analysts, reliability, and RCM practitioners, with and without condition sensors, to increase detectability. Machine learning is not magical. Just like any other tool, it must be used properly by trained practitioners. Instead of performing machinery-health analytics, the practitioners can train an army of health-monitoring agents. The practitioner determines, based on their knowledge of RCM principles, the equipment, and the process, which agents should be built, how they should be deployed, when they should be retrained, what detectability they should increase, how frequently they should alert whom and, most important, what action they should inform.

Combining machine learning with rules, complex event processing, and risk calculation based on severity in a coherent platform can lead to RCM success. At the risk of coining another term, let’s call this integrated system a digital RCM system. EP

Interested in learning more? Join us July 13 and 14 for a two-day course that will prepare you to implement ML and AI for digital asset performance monitoring in your Industry 4.0 initiatives. Visit rmc.utk.edu/events/. 

Based in Knoxville, 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.

Rajiv Anand is the co-founder and CEO of Quartic.ai, San Jose, CA (quartic.ai), provider of a smart-industry platform
for process-manufacturing industries that implements digital transformation with AI and IIoT.