With machine learning and advanced analytics, new opportunities are arising to complement traditional reliability programs, for both your critical and less-critical assets.
In traditional reliability programs, the focus is generally to prioritize those assets that are most critical for the operation of your business. With machine learning and advanced analytics, new opportunities are arising to complement traditional reliability programs, not only for critical assets but also for a number of less-critical assets.
WHAT IS A CRITICAL ASSET?
In general, a critical asset is one that would have catastrophic health, safety, environmental, operations, or customer-related consequences if it failed. For example, a power turbine on an oil platform can be a critical asset as if it fails, the production stops. Likewise, this is also true for a mill in a mining processing plant. Sometimes critical assets are paired to create redundancy. However, we’ve observed downtime in redundant assets too (e.g. pairs of diesel generators on a platform or pair of pumps in a chemical manufacturing plant).
Most likely, you’ve already rated your assets based on their criticality. Simply put, the criticality of your asset is the relative importance of an asset or system to the mission of your organization. What impact would each asset’s failure have on your business and which assets would bring the greatest consequences? If the asset goes down, you’ll get bottlenecks or downtime in your production. Measuring your assets’ criticality can give you greater control over which preventative actions are needed to reduce or prevent downtime.
MAINTAINING YOUR CRITICAL ASSETS
Maintenance is at the very core of an industrial business, but where’s the balance of how much maintenance is needed? Your maintenance strategy should be different based on the varying levels of criticality. Even without changing the current maintenance regime, a proactive action like, for example, anomaly detection, will often deliver a lot of value without requiring complex maintenance regime changes.
However, one problem our customers often point at is that they often already have control systems with all sorts of threshold alerts, trend alerts, and user interfaces. How do you deal with that? Especially when such systems are producing thousands of alerts? Enter machine learning.
MONITOR YOUR ASSETS USING MACHINE LEARNING
Recent advancements in digital solutions enable you to listen to what your critical assets are trying to tell you. You’ll be able to build richer analytics and create digital solutions enabling you to monitor and estimate future states of your assets and thus improve your uptime.
Having an intelligent layer between your control systems and assets picking up the most important sensors and combinations of signals from your critical assets is very compelling, as it enables advance notice of potentially critical events.
This is achieved by creating an intelligence layer between the equipment and your operator. A machine learning approach allows you to ingest the past behavior of equipment (by processing, for example, the past several years of time-series from the sensors), learn about failure events and what a normal operating mode looks like. Based on this, the algorithm learns what’s normal and what’s unexpected – or may lead to failure. The intelligent layer is translating what the equipment appears to be babbling about (the flood of alerts) into selected few insights making it possible to be processed by a human. That’s distinguishing the signal from the noise.
BUT WHAT ABOUT YOUR LESS-CRITICAL ASSETS?
You might say, well I have dozens and dozens of less-critical assets too. There are costs associated with these too and downtime in these will also affect my business. Using machine learning, you can have a layer of continuous monitoring on your assets enabling you to optimize your fleet.
THE PATH TO PREDICTIVE MAINTENANCE IS A JOURNEY
However, as we’ve mentioned in our earlier blogs, true predictive maintenance doesn’t exist “out of the box.” For machine learning to recognize all potential failures, it must see that assets fail in every possible way, many times, to truly understand all potential failure modes. As you might already know, expensive, highly-engineered assets rarely fail.
Rather than attempting to build predictive maintenance capabilities immediately, you should start a roadmap approach of equipment monitoring. Start by streaming your data capture and threshold-based alerts and integrate anomaly detection techniques across multiple sensors as your groundwork.
You can read more about the journey to predictive equipment maintenance in this blog article.