Predictive equipment maintenance is the ultimate goal of many digital initiatives in heavy industries, but the reality is much more difficult than operators may anticipate. In this article, I discuss the path to predictive maintenance, including common challenges and pitfalls.
The promise of predictive equipment maintenance is compelling to any asset operator, plant manager, maintenance or reliability engineer: a program will tell you that your equipment is going to fail – how and when. Based on this, you can automate tickets, or maybe even automatically initiate spare parts requests and work orders. In this vision, the equipment itself drives its maintenance and turnaround needs – faster, better, and more cost-effectively than any current scheduled maintenance or traditional reliability program.
Unplanned downtime becomes a thing of the past. This future state of fully automated, universal predictive equipment maintenance is a big idea.
But true predictive maintenance isn’t here yet
Here’s a dirty secret about predictive maintenance: in order for a machine to recognize all potential failures, it must see a pattern of each possible failure many times. Furthermore, it must see such failures in specific pieces of equipment installed for specific applications, in specific systems. This is because the exact same type of equipment will behave differently based on different field conditions. For instance, a pump installed in Alberta, Canada will experience different “normal” operating temperatures than the same make and model of a pump installed in Texas, even if they are working in similar applications with similar fluids. It’s impossible for a program to know what “normal” temperatures are for the Texas pump just based on historical data from the Canada pump.
In other words, for any specific pump, compressor, heat exchanger, generator, or motor installed in any specific system, a predictive maintenance program must see that equipment fail in every possible way, many times, to truly understand the total universe of potential failure modes.
The challenge is that equipment doesn’t fail that often
Expensive, highly-engineered assets rarely fail. They almost never fail the same way multiple times. This is the uncomfortable fact of predictive maintenance. Without sufficient data, it’s difficult to get useful results. Even with detailed sensor data, maintenance logs, and well-labeled events, most industrial equipment doesn’t fail frequently enough to enable accurate predictions.
Where does that leave the future of fully automated equipment operations? Are we stuck in the present?
The path to predictive maintenance is a journey
Reality is that you probably won’t be able to do predictive maintenance out-of-the-box in the short-term. However, there are many no-risk steps to cumulatively lay the foundation for true predictive maintenance in the future:
- Start by determining what’s valuable. What actions would you like to change or take based on new equipment insights? For instance, if you already have detailed instrumentation on critical assets, will the new data enable any meaningful new changes?
- Determine what’s feasible. Where and how is data collected? How unified is the format? How accessible is it? What processes will be changed, and how easy will it be to implement these changes?
- Start with the easiest path that creates value.
- Is it valuable to visualize real-time data?
- Is it valuable to have threshold alerts on specific sensor values?
- Would you do something differently if you could compare actual equipment performance against simulated or tested performance expectations?
- Could you take different actions if you classify failure modes – or non-optimal behaviors – as they happen?
- Could anomaly detection programs create value in your equipment fleet? Is this most valuable for critical assets or less-instrumented assets, perhaps that are currently run-to-failure?
There are many potential goals in the predictive maintenance journey. As with any major technology shift, starting with the end-user and end-process in mind could clarify the steps to make the journey valuable, immediately and in the long-term.
Want to learn how you can start the journey to predictive equipment maintenance? Click on the button below and download our whitepaper “The Reality of Predictive Equipment Maintenance - Cluster based Anomaly Detection for Industrial Operations”.