An additional gross value creation of 32 billion euros in the German manufacturing sector by 2023—this was the prediction published in 2020 by the German Federal Ministry for Economic Affairs and Climate Action regarding the use of artificial intelligence in the industrial sector. The year before, a Bitkom industry survey found that 10 percent of industrial companies are already using artificial intelligence. In the meantime, automation specialists, machine manufacturers and plant operators have become more and more interested in this technology.
AI’s ubiquitous advance
As a matter of fact, people are not always aware of how many AI applications are already shaping their environment today. The voice assistant on their smartphones, digital translation services, recommendations for online shopping, or browsing YouTube videos are just a few better-known examples of the AI support used daily in the consumer sector. Image recognition in Google Lens, a health cloud that supports medical diagnoses, the automated approval of loans on online platforms, or even insurance rates based on the real-time analysis of a person’s driving can also be named here.
And most consumers are probably not aware of the fact that unlocking their smartphones via face recognition or the automatic text completion feature in chat applications can only work by using AI. IT giants such as Amazon, Google, Microsoft and Apple have invested large amounts of money and resources in the development of AI so that today, it can be used easily for many different applications. Ultimately, this also benefits the control technology sector.
Distinction between automation and AI
Even though industrial applications entail different tasks and challenges, the basic function of artificial intelligence is the same here as in the consumer sector: AI must simplify and improve certain processes.
This leads us to the question: Where does “classic” automation end and artificial intelligence start? Or in even more concrete terms: What exactly is artificial intelligence in the field of industrial control and automation technology?
This question can be answered from different perspectives. One aspect is the effect of AI: Artificial intelligence in the context of automation means that machines are enabled to communicate, see, interpret, feel, think, and decide.
The other aspect is the implementation. In automation, there is a program for every problem. The problem solution is hard-coded, meaning that it is – unfortunately – not scalable. By contrast, AI does not consist of fixed code; rather, it depends on data. Algorithms and neuronal networks are modeled and trained using historical data.
Starting from this foundation, results keep improving as more and more data is being processed. This ability to take current data and adapt to new environmental conditions is why artificial intelligence offers such fantastic potential for the scaling of application scenarios. Because unlike programming code, where the continuing development is linear, the progress in AI development is exponential.
Three AI application categories
If we examine the potential use cases of AI in automation, we can identify three basic categories:
- Assistant systems that help the operator.
- Local AI systems for the autonomous control of processes in real time.
- Analytic applications in the cloud.
All three categories have a different kind of potential for using AI in industrial automation and control technology. Assistant systems can simplify complex processes by providing suggestions or help, for example for operation, commissioning or programming. Cloud applications, for example, are suited to predictive maintenance or the detection of process anomalies.
Local AI solutions, for example, can provide the machine with new communication abilities so that it can use its sensor system to think, learn, and decide. Ultimately, the goal is to increase productivity and efficiency, for example through shorter cycle times.
Local applications alone are able to reliably control running processes, such as the optimal movements of a robot or an automated guided vehicle (AGV). In cloud applications, the delay times are too long; in addition, there is a risk of
Anyone who does not fully embrace the use of AI in industrial applications is at risk of missing the boat.
Things AI cannot yet do
Today, more and more machine manufacturers and automation specialists are looking into using AI. But so far, they are not taking full advantage of the possibilities. Applications are frequently limited to putting the data analysis into the cloud and to create suitable dashboards that create data transparency for operators. This means that the collection and smart analysis of machine data and production data and the subsequent derivation of productivity enhancements are working quite well. However, this does not provide the complete picture of the entire process. There are several reasons.
In many areas of automation and control technology today, machine manufacturers and plant operators can rely on uniform standards and interoperable technologies. This means that even components and machines made by different manufacturers can work together relatively smoothly. Development can use function modules that are easy to assemble and replace.
By contrast, a uniform AI platform does not yet exist; instead, there is a multitude of ecosystems that are mostly not compatible with each other and that come from different automation specialists as well as from machine manufacturers.
In some parts of the process manufacturing industry, such as the pharmaceutical and food industries where the process chain has great overall importance, great efforts are already underway to overcome this fragmentation. But even here, it becomes apparent how difficult it is to integrate all process participants into one single platform and to create a common data base that could harness productivity potentials.
Other industries, such as packaging, plastics manufacturing or sheet metal processing, are even further behind on this road. The logistics and the robotics sectors have emerged as pioneers in environment detection. However, these are only localized usage scenarios that find a highly efficient solution for a narrowly defined partial process. The users’ goals—highly flexible production as profitable as possible – are only partially fulfilled.