What is the difference between classic automation and artificial intelligence?
Linde: One of the most obvious differences is the programming. Automation used hard-coded programs, in other words, fixed code. Artificial intelligence, by contrast, does not get programmed—it gets trained. The foundation is provided by algorithms and neuronal networks that are populated with historical data that optimizes them for a task.
What advantage does this provide for AI-based solutions?
Linde: In hard-coded systems, any change in the environment must be reflected by a change in the software. In view of the ever shorter cycles for product and production updates, this adds up to significant manual effort. AI solutions, on the other, can adapt to change by learning from the accumulated data. Less update effort, fewer disruptions, and continuous process improvements with their ensuing positive effect on the quality and efficiency of the system—these are among the most salient benefits of AI-based solutions.
But does the “continuous improvement” achieved by AI not hit a ceiling at some point?
Linde: Only if nothing were to change. But for many industries, the days are over when manufacturing remained unchanged for 20 or 30 years. Production framework conditions keep changing—and AI is able to adapt; a pure automation solution, by contrast, needs to be replaced.
Machine engineering has always followed the mantra “defined input, defined output” in order to ensure safety and security. How is that going to work in a self-learning system?
Linde: There will always be classic logical processes on a machine that are relevant for safety and security. These must continue to work with utmost reliability and availability. But aside from that, there is a shift regarding technology and features, because the machines need additional capabilities. Data-based AI for the detection of anomalies, for example, is more able to ensure consistent product quality than hard-wired solutions. Machine manufacturers and automation specialists now face the challenge of recognizing these two different paths and finding approaches that create a smart combination of the two.
What is the most important change on the way to more AI in machine engineering—a shift in awareness?
Linde: Yes, that too, but what is crucial is for companies to build their in-house expertise. They don't necessarily have the required skills. This is a paradigm shift, after all. There is a lack of qualified specialists and a lack of experience with this innovative technology. While we do observe that these two different worlds—automation and AI—are slowly growing together, they are far from harmonizing at this point in time.
“After a certain point in time, companies who do not use any AI solutions in the future will no longer be competitive.”
Why do AI experts not jump in to fill this gap?
Linde: Oh, but they do. And so do providers of IoT solutions. Their experience with the digitization of processes is frequently giving them a head start. But on the flip side, they are lacking the in-depth domain knowledge about customer processes—this is where machine manufacturers and automation specialists have the advantage. This is why the latter have to accept this challenge and expand their AI competencies as quickly as possible in order to develop better AI solutions than the competition who are not part of the industry.
Do you see differences in different sectors of industry?
Linde: The trailblazers in the area of control systems are the robot manufacturers who use AI for motion optimization. Machine manufacturers, by contrast, focus more on predictive systems for predictive maintenance and on visualization, for example through dashboards. Other sectors are still lagging behind. But this might change—there is a lot of movement in the AI market. Anyone who doesn’t joint the movement now is at risk of missing the boat.
Does the development of AI compare to historical innovation leaps?
Linde: Not really. Primarily because now, a number of trends are converging: greater digitization and interconnectedness, an extreme growth of computing power, and on top of that AI. These are all coming together now, and the result is a completely new way of thinking about things and making things. This has laid the foundation for disruptive development leaps.
Can you predict when these disruptions will happen?
Linde: We already see the signs. For example, there are new players on the market who develop sensor-based control solutions that combine sensors such as cameras with AI. These don’t use “classic” programs; instead, they use models or task sequences that are described and trained. When this technology hits the market, OEMs will have a problem because their solutions that are based on classic automation technology are nowhere near as scalable and thus no longer competitive. You could call that a disruption.
Could standards and common platforms, perhaps with the backing of industry associations, be a suitable defense strategy?
Linde: There are some initiatives that promote the use of AI in machine manufacturing and in the framework of Industry 4.0, but there are no standards yet. From our perspective, it is actually still too early for standardization. There is a lot movement in the market, and the actors currently still have all the options for positioning themselves on the AI market by driving their own developments and ideas.