What role does AI play in functional safety?

AI is increasingly influencing developments in automation technology in general. However, many AI functions are not yet directly transferable to safety technology. To use AI reliably in functional safety, more time and further research are needed. Nevertheless, we are closely monitoring this innovative potential and are already actively researching in this area.

Can safety functions become more flexible with AI?

In the future, AI combined with radar technology and sensors could provide great flexibility in safety and thereby increase the productivity of plants. For example, by reliably distinguishing between people and mobile robots: in order to increase plant productivity, it should be possible for a machine to continue operating when robots or automated guided vehicles enter a protected area. If, however, a human enters the protected area, the system could reduce production speed or bring the plant to a safe stop—depending on the person’s direction and speed of movement.
A key advantage is that no personal information needs to be generated. Instead of using camera images, objects are detected as point clouds. AI then distinguishes between a human and an object such as an automated guided vehicle.

What challenges do you see?

To use AI in machine safety, not only safety and reliability but also transparency are essential components. In this regard, AI is unfortunately still a “black box.” It is neither reliably predictable nor fully traceable how such a complex system reaches decisions. However, every reaction of a safety system must be justified.
Another challenge is the interaction between humans and AI: a so-called “strong AI,” equivalent to or surpassing human intelligence, does not yet exist. Consequently, AI systems today can only function within previously defined human parameters.
Furthermore, one must consider whether self-learning software could potentially result in a new state of safety functions in a machine.

Are there application examples at Pilz? Which partners are you working with?

A concrete application example is our Operational Safety Intelligence project in collaboration with the SmartFactory-KL research platform. TÜV Süd and B&R Automation are also involved.

Within this research project, we integrated a dynamic safety system into a production ecosystem. The production environment adapts to human behavior. With AI, we gain new possibilities: to continue prioritizing human safety while also maintaining machine productivity as effectively as possible.
In this application, our research radar sensor dynamically detects how a human moves: it records the position, direction, and speed. AI then determines the best possible response. For example, the plant can resume normal speed as soon as a person starts moving out of a warning or protection zone.

Another productivity optimization is distinguishing whether a human or an object is approaching the plant. Since our research radar sensor also provides this information, AI can decide that the plant does not need to stop or slow down if, for example, an automated guided vehicle—rather than a human—enters the warning or protection zone.

What role do standardization play?

Both play a major role here. In addition to the general EU regulation on artificial intelligence - the 'AI Act' - the EU Machinery Regulation must also be observed.

The EU's AI Act regulates what an AI system is generally allowed to do. Above all, it also regulates safety and the associated fundamental human rights. The manipulation of people by AI is therefore prohibited. People may not be induced by AI to make a decision that causes significant harm to themselves or others. As part of this AI Act, certain AI applications in the areas of education, law enforcement and critical infrastructure have been classified as high-risk AI systems. These must meet special requirements and be certified with the CE mark.

In addition, the EU Machinery Regulation already stipulates that a conformity assessment procedure must be carried out when AI systems are used in functional safety. Safety-relevant AI systems are therefore also classified as 'high-risk machinery'. For us at Pilz, this means that a conformity assessment procedure can only be carried out with the involvement of a 'notified body', even if the relevant harmonized standards are applied.

The current status of standardization is that the first standardization in the area of 'AI & Safety' is currently being developed with ISO/IEC CD TR 5469 'Artificial Intelligence - Functional Safety and AI Systems'. Pilz experts are also involved here.

Where is safety heading with AI support?

In the future, we also hope to increase efficiency in the area of safety through the use of AI. Where high engineering effort is currently still required, AI will be used profitably and in a resource-saving manner in the future. One example of this is the safe detection of people: in future, AI-based object detection, object tracking and sensor data fusion will be an elementary component here. If users know for sure where people or machine operators can safely be in production areas, they can ensure the safety of people and increase plant productivity.