What is Predictive Maintenance?

Blog     · 4 MIN READ

Posted on September 22

Back to blog

My father has an expression as he guesses when something will be completed: “Barring any major catastrophes.” This qualifier was applied to car trips, household projects, errand runs, etc. He believes he can make his best guess, but life has a way of creating impediments to efficiency—and that was out of his hands.

Old-School Maintenance
This was also a common problem with asset maintenance. Sure, a schedule based on the reliable advice of experienced engineers was useful, but it really didn’t cover those small problems that built into major incidents over time. The only option available was to shut down and inspect everything, replacing and maintaining to serve the schedule, not the business. Since you’d do so every so often, it was called “periodic” maintenance. 

Periodic—or preventative, as it’s called today—maintenance works pretty well for the major problems—your car needs oil every 3000 miles or three months, and you can’t drive it when they are working on it (they get very cranky when you try). Lacking the ability to see inside the asset without taking it apart, there’s really no other option. However, as technology advanced and allowed for non-destructive monitoring (infrared scans, acoustic measurements, vibrations analysis, etc.), a new style of maintenance was developed. Predictive maintenance (PdM) incorporated analytics and historicized data to predict when maintenance would be required. 

Predictive Maintenance
A version of PdM has always been around. There is the apocryphal-but-plausible tale of the first engineer to put a screwdriver to his ear, touch a machine, and pronounce it sounded like a bearing was going bad. Modern PdM is a bit more involved; to successfully implement it, you must first be able to monitor the asset. Without understanding what is happening mechanically, you cannot predict anything. Next, you must be able to diagnose areas of concern based on that information. Finally, you must be able to fix issues when they arise. Now, most of this is true of preventative maintenance as well. Where it differs is analytics.

The techniques to understand when maintenance will be required are both new and old. Statistical analysis is hardly a new science; however, easily applying it to massive amounts of data is new because it was impossible without modern computers. Doing so requires a baseline: how should the asset be performing when all things are working within tolerance. Once established, you can perform limit/range tests: is everything behaving within that norm? If not, you isolate and repair. With enough time, you can develop pattern recognition. With enough viewpoints, you can perform correlation analysis by comparing them. All of this is managed by computers with minimal oversight. 

The Benefits of Predictive Maintenance
It is deliberate in its methodology and execution, allowing operators to schedule maintenance when it is convenient and eliminate the dreaded “unexpected downtime.” If the entire process is automated, maintenance technicians spend their time maintaining and not dealing with emergent situations. The necessary work of maintenance is done when it hurts the business the least.

PdM also increases safety: non-destructive monitoring methods can be transmitted by wireless sensor networks. Visual inspections can often be done by a drone or other robot. This limits the risk of human injury by having the drone scale the heights of an asset, or the robot delve the depths of a vessel rather than the operator or technician. PdM in general also lowers the chance of a dangerous malfunction causing damage, which is much safer as well.

Predictive Maintenance Challenges
There are, of course, downsides. Good data, as in all things, is absolutely critical. Bad data leads to bad outcomes. If data is too sparse (say, a fault that happens once a year), it is very hard to create reliable predictions. If data is not accurate, false alarms can cause wasted time and effort—or worse, failures that go undetected until too late. But digital twins are working on that problem—it’s too much for this post, but human problems have human solutions, and the same is typically true of machine problems.

A recent survey conducted by Reliable Plant suggests 65% of readers already use PdM in one form or another. It’s here to stay and extremely beneficial to any plant that implements it correctly. While my father still qualifies predictions on when he will arrive home from the office, more and more this is not necessary for industrial operations.
Share this article

Back to blog