There are a lot of data-driven maintenance strategies, how do you know which one is right for you and your assets? How do you know if any are even better than simple preventative maintenance or no proactive strategy at all?!
In this article, I will provide an overview of the following 5 types of maintenance using lots of graphics:
As you can see, there's a lot to cover in this post!:
To define and understand the difference between these types of maintenance, let's first look at a P-F curve which starts with a potential failure (P) and ends with a functional failure (F).
Reactive maintenance waits for the system to experience a functional failure before maintenance occurs, so this doesn't even appear on the P-F curve.
Preventative maintenance performs maintenance on some fixed schedule ideally aligned to be slightly shorter than the typical P-F interval for a particular machine. This will ideally let it catch a system while it is in the middle of the curve. The trouble is that the equipment may deteriorate too fast for the preventative maintenance schedule to correct it in time. Or, the preventative maintenance may be wasteful if it is "fixing" a perfectly fine piece of equipment.
Condition-based maintenance uses sensors and data with preset conditions or thresholds that when met will signal maintenance is needed. The trouble is that this requires a fairly noticeable amount of degradation to have taken place in order to hit these thresholds.
Predictive maintenance uses sensors and data to detect trends in the health of a system and predict when failure will occur. This allows it to detect the deterioration of a machine earlier than CBM and allows maintenance teams more time to schedule maintenance at a convenient time, knowing when the PdM predicted failure to occur.
Prescriptive Maintenance uses sensors, data, and advanced analytics to determine the route cause of a potential failure so specific corrective action can be prescribed. The advanced data and analytics needed for successful prescriptive maintenance also ensures the potential failure is identified even earlier which makes fixing the problem easier and less expensive.
Another way to look at the difference between maintenance types is to understand their dependence on data. This is the most defining distinction between these maintenance strategies.
Reactive maintenance has absolutely no dependence on data, it waits for failure.
Preventative maintenance uses either time or usage to inform its maintenance schedule, it's not a lot of data - but it's a start!
Condition-based maintenance will typically look at a single metric and check against preset thresholds.
Predictive maintenance typically looks at a handful of data sets and relies on slightly better analytics to pick up on various trends in the data and the health of the asset.
Prescriptive maintenance uses many data sets and metrics and likely some proprietary analysis techniques to determine the root cause of a potential failure.
Now that we've given a good overview of the different maintenance strategies, let's dive a bit deeper into each one.
Reactive maintenance is just as it sounds: wait until the machine fails and then fix it. In general, this is often not really a maintenance strategy -- it is often the result of having no strategy at all. Once the machine goes down, the team scrambles trying to find a fix and they feel like they are constantly putting out fires!
There is an exception, though, when you can make reactive maintenance an actual strategy that has some value. If your machines are:
Then a reactive maintenance strategy can be a good idea assuming you are planning for this and it is not costly in both time and money. This is called a "Run to Failure" strategy.
Preventive maintenance is using some routine, either time or usage, to define when you replace or rework some components of your system. An example of this is how you replace the oil in your car every 5,000 miles.
This is a good idea to prevent any deterioration in the machine from occurring in the first place. It is easier to plan and work around and it's also generally less expensive to replace a simple component a few times than repairing a damaged machine after failure. But this has a few notable issues:
The more complex your machine is, the more unlikely preventative maintenance will be a feasible maintenance strategy. This is because it will require too many different things to maintain (all at different rates) making it probable that one of those things were excessively fixed (wasteful) or that a failure was missed altogether.
Condition-based maintenance uses sensors mounted to a machine to periodically check the health of that asset. Then it compares the most recently measured value against preset "conditions" to be reached which signify an alarm or warning. Typically this is done by:
In the example below, about 70 hours into the profile the velocity RMS (a common vibration metric to use) on this machine reaches the warning level which a condition-based maintenance system could detect and report.
In the example above about condition-based maintenance, there was clearly a trend in the data before it reached the warning level. This trend is exactly what a more advanced algorithm would be able to detect and is more accurately considered "predictive." Predictive would then go that added step further and predict how much usable time remains in the asset to schedule maintenance -- it predicts when failure will occur.
Predictive maintenance (PdM) is not magic; it is about measuring one or many parameters and using this data to make an informed prediction about how the health of the machine will look in the near future.
In order to have this data-driven crystal ball, predictive maintenance solutions typically require more & better data as well as advanced analytics compared to CBM solutions. And this will often require some machine learning algorithms to be effective.
In order to have a prescriptive maintenance program, you need to be able to correlate specific signatures in the raw data and trends to specific root causes. And this tends to require you to have been able to previously find these signatures after the fact.
Let's look at an example from the US Navy using enDAQ sensors detailed in this case study. The US Navy begins seeing a rise in physiological episodes that put crewman and aircraft at risk -- this was both a safety and financial risk [see news article]. Physiological episodes were manifesting as pilot hypoxia symptoms thought to be a result of cockpit air contamination, lack of oxygen, or issues with the Environmental Control System (ECS).
They responded to this issue by:
To quote the Rear Admiral Fredrick Luchtman:
“...we are almost to the point with our data analytics where we receive information from the fleet, we analyze that information, and then we can tell the fleet, hey this particular aircraft is exhibiting signs that this particular part may be needing to be replaced pretty soon.”
This quote and example is especially import to highlight that prescriptive maintenance (and predictive for that manner) requires an evolution: You start by being in a state where you have a problem and are reacting to failures. Then you begin doing some preventative maintenance to and in that may start periodically gathering data. This data can then start being used as a condition-monitoring program. With more time and more data, you may have enough to start being able to use that to predict failure. And with even more data you begin being able to find root causes and prescribe the specific corrective action ahead of time. It's an evolution, so let's revisit that earlier image!
When looking at the cost of a maintenance strategy, there will be a few considerations:
These maintenance investments will (hopefully) have a return on said investment in the form of:
The plot below compares our maintenance strategies and illustrates how they stack up on that return on investment.
There isn't a one-size-fits-all maintenance strategy. The best strategy depends on the machine asset's complexity and criticality. The chart below breaks these two variables into 4 quadrants and plots where each maintenance strategy aligns best for that type of machine.
Hopefully the benefits of prescriptive maintenance are intriguing, but how do you get started!? Well, you can't start off by doing prescriptive maintenance -- you can't even start off by doing predictive maintenance. The way you get started is by getting some data on your machine at a few different times.
Ideally, you are already doing some preventative maintenance and so you are utilizing a schedule to go out to the asset, inspect it, and potentially repair parts. Use these opportunities to gather some vibration, temperature, and other sensor data on your asset.
You don't have to use a fancy wireless monitoring system for this, either. You can use a simple vibration meter or data logger (See our blog post 6 Ways to Measure Vibration for an explanation of the full range of vibration measurement systems readily available to a test engineer).
As you go through the process of collecting data on your asset, you will be able to determine:
Once you're able to answer these questions, you'll be able to understand if a data-driven maintenance strategy can work for you and your machines. Then you can begin expanding your "exploration" period. This data-driven maintenance exploration will eventually evolve into a prescriptive maintenance strategy over the course of time with continuous data collection and data analysis.
Like I demonstrated earlier with the maintenance evolution chart, prescriptive maintenance requires an evolution. All you have to do to start the process is start taking some data!
I hope that after reading this post that you now have a good understanding of the differences between reactive, preventative, condition-based, predictive, and prescriptive maintenance and that you're ready to start collecting some data and exploring how you can perform prescriptive maintenance on your own assets.
If you have any questions, please don't hesitate to leave a comment or contact us directly. We are here to help you with all your testing and analysis needs! And, if you liked this post, I hope you'll subscribe to our blog for monthly updates on monitoring, vibration analysis, data acquisition and more!
Related Posts:
For more on this topic, visit our dedicated Wireless Vibration Monitoring Systems resource page. There you’ll find more blog posts, case studies, webinars, software, and products focused on your condition monitoring and maintenance needs.