The best time to repair or replace something is just before it starts to break down. Doing so unerringly would require a crystal ball, but modern predictive maintenance technology offers the next best thing.
Hardware and software are available that can constantly monitor equipment and, by applying machine learning to historical data, warn when a breakdown or other problem is imminent. Bolstered by wireless technology and the Industrial Internet of Things (IIoT), these customizable systems have the potential to bring predictive maintenance to a new level.
Predictive maintenance is a step beyond preventive maintenance, the latter of which basically sets a schedule for equipment to be serviced or replaced, based on estimated time between failures. Predictive maintenance takes that further by applying data gathered through constant monitoring to determine how much actual wear and stress the equipment is going through.
The idea is to make maintenance more efficient by performing maintenance as close as possible to the moment when the machine would start to fail without it. According to a McKinsey estimate, 35% to 50% of all time-based preventive maintenance tasks could be eliminated with predictive maintenance. Conversely, predictive maintenance can flag a problem in real time, on a piece of equipment that, under a preventive maintenance program, may not be due for servicing until it’s too late.
Food processors who are thinking of using high-tech predictive maintenance have to make some basic decisions: which machines they’re going to monitor, what will be doing the monitoring, what kind of conditions and ranges will be monitored, and how the extracted data will be handled and used.
What’re you looking at?
The first decision, of course, is which equipment to monitor – meaning which is most important, in terms of wanting to avoid a breakdown at all costs.
“Most customers would identify critical pieces of these types of equipment where they would want to receive alert notices for inspection and to prevent any unplanned downtime,” says Chris Teague, key account manager for the food and beverage team at ABB Motors and Mechanical. “This could be on many types of equipment in different areas of the plant such as conveyors, mixers, compressors, fans and pumps.”
John Wright, vice president of sales for software provider Somax, would put conveyors at the top of the list. Customers often want to concentrate on motors that drive major equipment, like mixers and ovens. But those motors are not as prone to failure as the ones that drive conveyors, he says.
“Conveyor systems go down more than anything,” according to Wright. “Even though the customers never really want to start there, for an immediate return on investment, start monitoring motors and gearboxes on your conveyors. Because if you go back through history and look at where your real downtime comes from, it’s not always your big equipment. It’s the smaller equipment that’s going to cause your downtime.”
Once the critical assets have been identified, the question then becomes which of their components will be monitored. The most common ones are motors, the gearboxes and other drive chain components that transmit their power, and the bearings that allow everything to rotate.
“Anything that moves in any way, shape or form, it’s going to be a gearbox and motor primarily” that needs monitoring, says Frederic Baudart, lead consulting specialist manager for Fluke.
Getting a good sense
The next question is what will be used to monitor the equipment. Specialized sensors, small enough to attach almost anywhere, can monitor for relevant conditions like heat, vibration, air pressure, moisture, current draw and more. These kinds of sensors can be hooked up wirelessly to send a stream of data to an app of the customer’s choosing.
With an IIoT connection, this app can be run by an equipment vendor or other outside party. These systems allow constant monitoring of critical equipment, with alarms and other feedback sent back to the end user instantly.
An example of these kinds of sensors is the Ability series from ABB, which can be mounted on bearings and other equipment. They are set to monitor temperature and vibration and can send that data, usually through a gateway server, across the internet to ABB’s app for data gathering and analysis (although it can go to another app of the customer’s choosing).
“Most customers do utilize the ABB cloud platform that sends data back to their secured log-in portal, as it is easier to monitor and perform changes for multiple pieces of equipment being monitored,” Teague says. “But there are customers that might manage and analyze data from just a cell phone or tablet if they only have a few pieces of equipment being monitored.”
Maintenance monitoring doesn’t have to be done with extraneous sensors. If a machine has internal monitoring and computing capability, it can in effect monitor itself, with the data extracted through its PLC or other controls.
Harpak-ULMA, a supplier of flow wrappers and other packaging equipment, has machinery that can deliver real-time data on parameters like cycle counts, run time on critical components, seal time, temperature and vacuum. Most of this data is displayed on the machine’s operator interface, but “where and when possible, [end users] will also use this data for on machine predictive analytics,” says Peter Seward, Harpak’s vice president of engineering & technical service.
What to look for
The means of monitoring will, of course, depend heavily on what is being monitored for. Choosing the right parameters is the key to a predictive maintenance program. Probably the most common one is temperature. Running hot is a symptom of a variety of problems, like bad lubrication or electric current fluctuations, and temperature monitoring is relatively straightforward.
“Temperature is where we’re finding the most value, in checking on the temperatures of critical motors and gearboxes,” says Somax’s Wright. “Almost any imminent failure of a motor is going to be preceded by an increase in temperature.” A single temperature sensor, mounted at the junction of a motor and gearbox, can monitor them both, he says.
Another parameter is vibration, which can also indicate problems like misalignment, improper mounting, overload and bearing failure. However, reading vibration patterns is more complex than taking temperatures. To be useful, a vibration sensor should be set up, and its readings interpreted, by experts.
“A lot of times, when people have a high-end vibration sensor, they have no idea how to read it, what to do with the data, how to set the limits,” says Marc Williams, IoT project lead for Parker Hannifin. “So it becomes almost useless.”
Vibration is not a straightforward measurement, like temperature. It can change in an instant, it usually takes place along axes in three dimensions, and it’s highly dependent on external factors like vibration in neighboring equipment, the timing of production run cycles and even the location’s harmonics and acoustical properties. That’s why it takes specialized interpretation.
“The layman cannot read that data whatsoever,” Williams says. “You have to call in specialists, people with a Ph.D. or whatnot in acoustics or vibration technology.”
Parker Hannifin offers a way to simplify matters: software that can convert data from three dimensional axes into polar coordinates, which express values as vectors, with variations in direction and magnitude. The result is a graph that plant maintenance personnel can be easily trained to interpret.
Interpreting vibration data correctly is a key to properly setting the limits that trigger an alarm or other notifications. Without that, vibration sensors are useless.
If limits are set too low, “maintenance teams would be false-alarmed to death,” says Anthony Stanley, business development director for digital services at pump manufacturer Grundfos. “When the team checked on a vibration alert, they would often find no issues with the asset. Instead, the alarm was caused by a loose sensor or the sensors were not communicating properly. Because of so many false alarms, the maintenance team increased the threshold, which immediately made the predictive maintenance program irrelevant.”
Grundfos Machine Health is a service that supplies Grundfos experts with data processed by artificial intelligence and machine learning to, as Stanley puts it, “validate every conclusion the algorithm delivers and in real time contact the plant with actionable items.”
Artificial intelligence and machine learning have applications beyond vibration analysis; they’re the keys to a successful predictive maintenance program. To properly set limits and detect anomalies requires accurate data that’s specific to the equipment and its environment.
That data can come from multiple sources, including the equipment manufacturers or the end user’s own supervisory software. The key is collecting it over time and analyzing it properly. That takes an average of about one to six months of historical data, says Doug Cornwell, business development and controls engineering manager for Barnum Mechanical.
Historical data is required to determine an acceptable range for parameters like vibration. “Although it could be possible to run a system one time and use the data from that run as a baseline, you would see many more deviation alerts (nuisance alarms) than if you were to stack data from multiple runs where the asset was considered to be operating within its designed range,” Cornwell says.
The data can be cleared of anomalies that are traceable to outside events, like operator interference or a line starting or stopping, making it easier for predictive maintenance software to set ranges and start predicting asset failures.
Predictive technology, aided by remote data collection and transmission, has the potential to greatly improve a plant’s maintenance program. If applied correctly, it can enhance both efficiency and effectiveness.