All process equipment must undergo maintenance. And – just like the automobile – planned maintenance like a routine oil change is less disruptive than unplanned maintenance. While it is easy to push out an oil change or certain procedures with worthy excuses like, “the production numbers are lower than target” or “there’s not enough staff this shift”, failing to perform maintenance increases the likelihood of unplanned and even more expensive downtime.

One of the key metrics used to classify unplanned maintenance is the Mean Time Between Failure (MTBF), which is an average of the uptime between failures. As a plant engineer, the goal is to extend the MTBF as long as possible. This means shifting from the “if it ain’t broke, don’t fix it” mindset to the “what will fail next, and how should it be scheduled” mindset. To do this, it is worth thinking of maintenance in three categories: reactive, preventative and predictive.

Reactive maintenance

Reactive maintenance is the lowest form of maintenance planning; it means only performing maintenance when there is a unplanned failure. Most people have been guilty of the reactive maintenance mindset in some area of their lives or others, be it driving on bald tires or procrastinating changing the furnace filter.

In industry, reactive maintenance is marked by new management who may not understand the importance of planned outages over unplanned downtime, or short-sighted management who try to keep the uptime maximized, hoping to avoid failure.

Armed with this knowledge, plant engineers can start prioritizing which maintenance tasks can be shifted away from reactive maintenance towards other categories.

Preventative maintenance

Preventative maintenance tasks are ones that are performed on a regular schedule or replaced before failure based on convenience, such as a vehicle’s routine oil change.

Another method for scheduling preventative maintenance is to combine tasks to maximize uptime. Consider a piece of process equipment that needs a filter changed every 10,000 service hours. A vital security patch and a software upgrade (reactive maintenance) is required, and so the machine is taken out of service. While performing the software upgrade, a technician notices that the filter has served for 9800 hours. Rather than bringing the machine back online, just to take it down again in two weeks, it might be cost-effective to replace the filter, even though it still has some service life left, and maximize uptime on that machine.

Preventative maintenance was the gold standard for many years, as it made reactive maintenance less necessary. However, preventative maintenance, by its nature, removes serviceable components before they have reached the end of their service life. The balancing act of changing components often enough to prevent a reactive maintenance situation without wasting components that are still serviceable is difficult to manage. For example, perhaps instead of changing the oil every 5000 miles, the oil could be changed every 1000 miles, which would mean good oil and a good oil filter are disposed of at every change.

Furthermore, the act of performing maintenance can damage components. In the oil change example, every time a technician unscrews the oil filter, there is a chance they will damage the threads. Repeated, unnecessary maintenance, even preventative maintenance, can increase the chances of unplanned outages.

Some organizations will keep serviceable parts as a backup. A robotic gripper with a belt drive may require replacement of the belt every three months, but it was replaced during a maintenance task with only two months of service. That belt can be saved and brought back into service as a backup, should there be supplier issues or to compare belt wear at different stages of life. In either case, the belt should be clearly marked and documented with its remaining service life.

Predictive maintenance

Predictive maintenance looks at past data to forecast when future maintenance should be performed. It is a method of keeping all of the advantages of predictive maintenance, while avoiding the trouble of throwing away good components. The catch is that more data must be collected, and that requires additional sensors, hardware and system complication.

Some modern vehicles can evaluate the quality of motor oil while in service. Instead of simply discarding oil after a set number of miles or the passage of time, the oil properties are evaluated to see if the oil is still serviceable. This requires a new set of sensors and a comparison routine of sensor data to a library of oil conditions.

Industrial equipment can benefit from this process as well. Belt-driven systems can use displacement sensors or tension to indicate normal wear patterns. Ammeters can be used to measure current draw and see if motors are becoming overloaded, which can indicate worn bearings. The key is collecting as much data as is economically possible.

At first, predictive maintenance is tedious. Data collection is a must- if there is no past data to compare against, then there is no way to make an accurate prediction. To get results, one must have data and the statistical backing and tools to find the important trends.

Leveraging AI and ML for predictive maintenance

At first glance, it can be overwhelming to move from reactive maintenance towards predictive maintenance. There are numerous data points to collect, store and analyze, which requires sensors and other hardware. From there, technicians to collect additional notes and engineers to look for trends take human power from other tasks and can initially slow production.

However, once the data acquisition systems and hardware are in place, advances in artificial intelligence (AI) and machine learning (ML) can be used to find trends in the data much quicker than humans. They can also find subtle fluctuations and other variables that are hard to track.

All of this requires performing a cost benefit analysis to weigh the time and hardware investment against the expected Return on Investment (ROI). Some maintenance tasks may not be cost effective to automate; it might be better to leave it as a preventative maintenance task.

Over time, these decisions will become more intuitive. In larger operations with high throughput, the ROI will be realized as maintenance mindset shifts to utilize today’s technology and best support the business.