Monitoring with AI "Because we come from mechanics"

Related Vendors

Festo SE & Co. KG

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Current AI applications allow pneumatic systems to be optimized and monitored with AI, even without deep specialist knowledge of artificial intelligence, explains Festo digitization expert Eberhard Klotz.

Under the term "Automation Experience," Festo offers various AI tools. What can users do with them?

Eberhard Klotz: There are three main focuses: the prediction of component or machine failure, quality approaches, meaning whether the machine or plant will still produce the required quality tomorrow, and energy topics: if, for example, an energy profile drifts away, maintenance is informed or error patterns are derived.

Our solution is special because, as Festo, we come from mechanics and mechatronics, and therefore naturally bring extensive experience with wear parts and motion profiles of machines and systems. Together with the software, we offer much better value creation than a software company or start-up with little production experience can.

Are there cases that can be predicted well or less well with AI?

These algorithms are always particularly effective in rapidly recurring and fast-running processes. Then I have typical behavior patterns of good states, as well as errors with a certain frequency, and that’s what these algorithms need. Because these applications work based on statistical probability. For applications that run very slowly and where an error pattern only occurs every seven years, these algorithms are not suitable.

To train and deploy AI for these use cases, a lot of data is needed. Where does Festo get this information from?

We have our own manufacturing facility where we can collect data. And: We have been conducting our own tests in trials for decades, where we could learn everything relevant from the data. This means that our standard AI apps come with a wealth of prior knowledge, and the algorithms are pre-trained. However, in customer projects, the basic rule is that the data remains the property of the customer: On-site, only the good state of the real application/control chain is briefly trained.

We started working on AI projects more than ten years ago. By now, we have so much experience and over 100 successful customer projects behind us that we can offer small algorithms encapsulated in so-called Docker containers. This means the customer no longer needs an AI specialist but can download this app from the app store and simply commission it themselves.

A machine builder can also realize new business models with it, with the advantage that they don't have to hire an AI specialist or develop the software, but simply use an app and offer, for example, a digital maintenance service. Their customer has the option to book the service or not—as they wish.

What about existing plants?

This solution is suitable for all customers who currently have no solution to monitor pneumatics.

But in which cases is it also sensible?

The classic PLC only checks whether the cylinder has arrived at the front or not. Pneumatics is inherently a very robust and reliable technology. Therefore, many customers have the mindset that you install pneumatics and forget about it because you don't need to worry about it. Nevertheless, there are applications that cause more wear and tear because the environmental conditions are so demanding, such as welding, cement dust in the air, or high heat. Such factors can lead to higher wear, and for all those customers, there is the possibility to monitor the system or machine with AI to avoid unplanned downtime.

Another application case relates to demographic change: when experienced employees retire, it is often a problem to fill the position. And know-how is lost. Anomaly detection by AI offers the possibility to replace some of this employee experience.

What has your experience been so far?

Our experience from the past ten years is that customers who have used these apps have all been successful. For example, maintenance improved by 20 to 25 percent in terms of less unplanned downtime and reduced repair times because I can read from the diagnostic messages what is broken. And as an operator, I don't have to search for the damage first; instead, I can see beforehand when a cylinder or an electric axis is drifting away.

Do additional sensors need to be installed?

Take a pneumatic cylinder as an example: To use the app, certain standards must be met, such as needing two end switches. From their signals, position and acceleration data are derived. If a customer wants more precision, additional data, such as compressed air flow profiles, can be used.

How is Festo planning to proceed in the AI field?

With the AI apps, we reached a turning point and are now moving from project business to mass business. We believe that the AI apps will be worthwhile for a very, very large number of customers.

The next steps are to add other AI areas, for example, to make robots more flexible. Assistance systems to guide machines are also a large field of work from our perspective. It would also be interesting to monitor larger manufacturing processes with AI. (jv)