Moving Maintenance From Preventive to Predictive

Analysis and simulation

An ability to detect patterns in this potentially huge amount of sensor data is critical. For example, data trends can show jagged peaks and valleys simply because of changing loads on the machine. A batch mixer, for instance, might see vibration spikes during the beginning or end of a run. Because of this, an engineer might set relatively large out-of-spec margins—which in turn could conflict with the ability to spot signs of trouble as early as possible.

Traditionally, the type of knowledge necessary to detect such signs resided with a maintenance technician—someone who might have been on the job for decades, comments Jonathan Darling, consumer packaged goods industrial automation market segment leader at Schneider Electric. That experience makes it possible for a tech to walk through a plant, listening to the sounds of the machinery or otherwise collecting information on operating conditions. Like someone driving a familiar car, the tech would be able to tell from the sound and other clues when something is off and then take corrective action.

One issue with this method is that it takes time to acquire that experience. Another problem is that it can lead to a conservative approach—a tendency to swap parts out early and continue to do things the way they’ve always been done.

Machine learning offers an alternative. With this technology, software ingests large amounts of sensor data from both good and bad operation, which can be defined as making in- or out-of-spec product or by some other criteria. From this information, software builds models used during day-to-day manufacturing to find problems. In manufacturing and elsewhere, machine learning has proven very successful in finding these hidden clues.

Rather than fixing machines as they fail, new capabilities will enable a more proactive approach.Photo courtesy of Schneider Electric

Machine learning, though, requires data—the more the better. The data must be from times when the machinery is working as desired and also at times when it is not. In a manufacturing plant that is performing well, it might be hard to come up with out-of-spec examples, making it take longer to collect the needed data.

The money must be spent upfront to put in all the sensors necessary to gather the data. “There’s a whole conversation around cost, with investing into adding those sensors or monitors to those pieces of equipment in the facility,” Darling notes. “But what is the cost of a piece of equipment going down?” The stoppage figure for a large CPG operation can be in the tens of thousands of dollars per hour, he points out.

Sometimes there are follow-on indirect costs as well. A chicken processing line, for example, might have only a limited amount of time that product can be left hanging on hooks because of health concerns or regulatory requirements, Van der Aa comments. A machinery breakdown that lasts long enough can lead to waste with raw product being scrapped and discarded.

Besides machine learning, another technology that can make use of IoT sensor data is a digital twin. This is a software replica of a physical system that responds to input changes in ways that mimic what happens in the real world. A digital twin makes it possible to get an idea of what will happen without it actually having to happen. Maintenance staff can simulate what variations in input feed rate, for example, would do to a machine’s output.

This capability makes it easier to troubleshoot problems, potential or otherwise, and investigate solutions without taking systems out of production. It also allows experimentation, such as what will occur if technicians adjust a system’s operational parameters. Being able to see such effects could lead to a benefit beyond maintenance.

Someone who has decades of experience might be skilled at keeping equipment up and running, Darling notes, but that might not be the optimal operating point. A digital twin offers a way to find the operating conditions that lead to the best possible outcomes.

Overcoming a skills shortage

A final area where technology advances promise to change maintenance for the better is AR and VR. Both present images to a user’s eyes, with these being either completely digital in origin in the case of VR or digital information overlaid on a real-world image for AR. Oculus, Microsoft, Meta, Apple, Sony, and others either have or will likely soon have products in this space. Most of these are aimed at consumers, but some are already being used in industrial settings.

These developments are attracting the attention of machine makers. “At Gericke, we’ve been paying attention to innovation in augmented and virtual reality where all the equipment documentation and maintenance steps are in front of the operator visually at any time, for example,” says Rene Medina, executive vice president at Gericke USA.

One reason to roll out this technology is the growing shortage of skilled maintenance personnel as well as the retirement of experienced workers. Training replacements and getting them in-the-field experience takes time. AR could be used to fill in the voids in the current landscape by providing distilled information and guidance, Smucker’s Zembas says. “It’s an ability to address both the skills gap as well as the standard maintenance activities.”

AR technology need not involve glasses or goggles, which today can struggle in brightly lit settings and other environments. The display could be something as simple as using a tablet or phone to scan a QR code on equipment, thereby pulling up information on a machine.

VR can be particularly useful in training scenarios. It can allow trainees to practice a repair task on a simulation, for example, giving them valuable experience without having to take an actual machine out of service.

“We need to bring down the learning curve by making it easier to understand systems and what is going on,” Darling says.

Amway’s Energy Drink Meets Skyrocketing Demand Through Digital Thread

Challenges

Though IoT, analytics, digital twinning, AR/VR, and other technologies—even using 3D printing to overcome spare part supply chain issues—are promising when it comes to maintenance, there are still challenges to overcome. One is a perceived risk of change, Darling notes.

In greenfield plants, there is plenty of willingness to embrace the latest technologies. But legacy operations are a different story. Some food and beverage manufacturers have 20-year or even older control systems. And if the machines are running and producing goods at a profit, it could be difficult to overcome a reluctance to deploy new technology. It might even be the case that implementation requires ripping out the old, yet working, system entirely.

Paralysis in rolling out these innovative technologies can also arise when users are confronted with many different places to insert them. One approach to overcoming this hurdle is to identify which assets are critical and begin by implementing maintenance changes there, advises Quantum Solutions’ Davis.

In general, hardware is not an issue, Van der Aa contends. Sensors and other technologies are at the point that generating and collecting data is not a problem. Analysis is more difficult because it involves turning that data into actionable information. A typical first step, he says, might be to produce a dashboard with gauges to indicate plant health. Dashboards, though, do not say what to do once a problem happens or is predicted to happen.

Counterintuitively, the many technology options and choices can make achieving this goal of actionable feedback difficult, and so simplification can be helpful. For a maintenance service company like SupportPro, there is also the fact that customers are conditioned by office software, phones, and computers to expect continuous evolution and improvement. An answer for needs involves picking the right products and developing easier solutions, an approach SupportPro is exploring.

“At our company, we’re working on a standardized platform so we can make it a lot easier on our customers,” Van der Aa says.