Stefan Bergstein: This industry is a prime example of the interplay between AI and automation. A wide range of trend-setting use cases can be observed here, particularly in the area of predictive maintenance. AI models analyze sensor data and can detect impending failures or material fatigue early on, avoiding unplanned downtime and optimizing service appointment scheduling. AI also plays a crucial role in quality assurance through computer vision, where image recognition systems perform visual inspections of components. This is a significant benefit because these systems can automatically detect faulty components, eliminating the need for time-consuming manual inspections.

Some manufacturers use AI for safety monitoring, analyzing human-machine interactions to prevent accidents. Additionally, visionary approaches, such as agent-based AI, could independently control entire production processes in the future. Lastly, containerized MLOps solutions are making it possible to use AI directly on edge devices, enabling rapid updates and flexible use in hybrid IT/OT environments. These use cases are no longer just promises for the future; they are already common practice in many cases.

Which AI methods, such as predictive maintenance, computer vision, or reinforcement learning, do you currently consider particularly promising for industrial automation?

Bergstein: In addition to the aforementioned methods, such as predictive maintenance and computer vision, new approaches that transcend mere monitoring and fault detection are gaining importance. One particularly exciting approach is reinforcement learning, in which learning agents optimize control strategies for complex production systems. This ensures adaptive, partially autonomous manufacturing in the long term. New, AI-based time series models for anomaly detection are also interesting: These models identify unusual patterns in sensor data or process parameters, helping to reveal deviations or safety risks early on. Another trend is the use of generative AI in the form of assistance systems, such as industrial co-pilots that support employees in programming and troubleshooting, thereby reducing downtime.

In what ways can AI-based automation be integrated into existing systems, such as PLCs, SCADAs, MESs, and edge devices?

Bergstein: This question will be crucial for many companies. A key term here is containerization. AI models can be packaged into container images and deployed flexibly on edge devices, gateways, or production servers. This approach ensures scalability and facilitates updates. Edge computing is important because it enables manufacturers to run their AI models on site with low latency without sending sensitive data to the cloud. Another closely related topic is the virtualization of control systems, such as software-defined PLCs. These decouple hardware and software, thus simplifying the use of AI services in traditional control environments. 

In practice, AI is often integrated with existing MES systems to incorporate quality forecasts and optimization suggestions directly into production logic. Open interfaces, such as OPC UA (Open Platform Communications Unified Architecture), are also becoming more important because they allow for flexible data integration, preventing companies from having to completely renew their systems. Thus, AI is gradually evolving from a separate tool into an integral part of the automation environment.

In addition to the DORA regulation, what other relevant standards and guidelines should companies in industrial automation be aware of? Examples include IEC 62443, ISO 9001, the Machinery Directive, and the EU AI Act.

Bergstein: The classic requirements of the Machinery Directive and Regulation, as well as safety standards such as IEC 61508 and IEC 62061, remain central. In the realm of IT and cybersecurity, standards such as IEC 62443, ISO 27001, and NIS 2 are particularly relevant, alongside requirements like the Cyber Resilience Act and the well-known GDPR. Quality management is another important consideration, with ISO 9001 and ISO/IEC 42001, the first standard for AI management systems, playing a role. However, sustainability and energy efficiency – especially ISO 50001 and ISO 14001 – as well as ethical governance, primarily the EU AI Act and IEEE standards, have also been coming into focus for some time now. In short, a multi-layered compliance portfolio is required that covers security, resilience, quality, sustainability, and ethics equally.

Could you provide examples or key figures demonstrating how AI-supported automation has improved efficiency, product quality, and plant safety?

Bergstein: Concrete ROI figures are not yet available because many projects are still in the rollout phase or have only recently gone live. However, the qualitative effects are clearly visible. In addition to predictive maintenance and computer vision, software-defined manufacturing environments are particularly efficient – updates are performed centrally and automatically instead of manually via individual controllers. Generative AI assistants and co-pilots significantly reduce troubleshooting time by processing machine data and log files in context. Additionally, greater IT/OT convergence, such as with containerized platforms, streamlines operations and improves overall security.

Which specific measures can manufacturing companies implement to increase their operational resilience with the help of automation, especially in the event of supply chain problems or energy shortages?

Bergstein: The strength of modern automation is especially evident when supply chains halt or energy becomes scarce. This is true not only in terms of efficiency, but also in terms of adaptability. AI-supported simulations are a key tool: they allow companies to run alternative production scenarios, such as using different supplier parts or modified energy profiles. This allows companies to make robust decisions quickly. At the same time, integrating edge AI ensures a high degree of resilience because processes can be analyzed and controlled locally, even if the cloud or central systems are unavailable.

However, we are also seeing that lifecycle management is becoming an increasingly important topic. Standardized platforms for updates, patching, and security minimize operational risks and prevent crises caused by outdated systems from worsening. Overall, this creates a stable production environment, even under difficult external conditions, and ensures flexibility.

In what ways can AI-supported automation relieve engineers and technicians in the production environment? For example, how can it help during commissioning, maintenance, or fault diagnosis?

Bergstein: Standardization is particularly helpful here. It applies not only to commissioning but also to updates that would otherwise have to be installed manually. Automation frameworks combined with AI take over configuration management and patching, reducing manual work and minimizing errors. Physical interventions on site are also less necessary. With virtual controllers or containerized systems, adjustments can be made remotely, and new functions can be implemented without downtime.

Even monotonous tasks, such as visual inspections and standard maintenance, can be increasingly outsourced to AI systems. This allows engineers to apply their expertise where it is needed most: in complex decisions and process development.

In industrial environments, how can automation ensure compliance with security guidelines and cybersecurity requirements, as well as proactively improve them?

Bergstein: Security in industry can no longer be understood as merely reactive, and automation is precisely where it shines. Rather than laboriously applying patches or manually coordinating configurations, companies can use frameworks such as GitOps to centrally define and enforce security guidelines. This reduces human error and ensures that systems remain up to date, which is a crucial step in preventing vulnerabilities.

Additionally, AI-supported anomaly detection surpasses traditional monitoring. It can detect suspicious patterns in network or process data in real time and trigger automatic countermeasures. This makes security dynamic rather than static, as it is a learning system that adapts to new threats.

Where do you foresee the greatest technological advances in the intersection of AI and industrial automation over the next three to five years?

Bergstein: Looking ahead, I foresee a paradigm shift in the role of AI in production above all else. Currently, we are testing many individual pilot projects, but in the future, generative AI will be an integral part of engineers' and machine operators' everyday work. Instead of a separate tool, we will have an industrial co-pilot that analyzes log files, supports fault diagnosis, and helps with programming, all via a natural language interface.

The second major leap will be agent-based AI. This will lead to the development of systems that can independently monitor production processes, detect anomalies, and implement initial countermeasures. This may sound futuristic, but it's clear that production lines will become increasingly self-optimizing and that the role of humans will shift from operation to higher-level control.

Finally, MLOps will become the new standard in manufacturing. This means AI models will be managed, versioned, and rolled out in the same robust manner as traditional software, rather than running in isolation. This will enable innovation on the shop floor to be implemented quickly and safely without compromising production stability.