Artificial intelligence has triggered a lot of hype - industrial AI is definitely not one of them. It does not produce castles in the air, but delivers tangible benefits and has a direct positive impact on key performance indicators such as overall plant efficiency, reject rates, mean time between failures and energy consumption. As a result, it increasingly determines how competitive companies remain in a volatile, energy- and cost-sensitive world.

Industrial AI does not have to start on a greenfield site to achieve this. Thanks to open, IP-based IT architectures, AI systems can be integrated into existing operational technology (OT), which has grown over decades and is usually very heterogeneous. Standardized communication protocols and flexible architectures enable a reliable and interoperable connection to existing PLC, fieldbus and Scada landscapes.

Modern industrial IT systems can support typical OT requirements such as high availability, long life cycles and stable operating conditions. The experience of manufacturing companies from the DACH region shows that a consistent, scalable architecture concept is crucial - otherwise industrial AI will remain stuck in pilot status.

Translating data into decisions

Industrial AI can capture data along production lines, contextualize it for machines, orders and batches and translate it into decisions in real time. Data collection with sensors creates a closed control loop that actively intervenes in the production process. AI-based pattern recognition, for example, can detect anomalies long before they are noticed by machine operators and enables machines to be maintained according to actual need rather than rigid intervals. AI-based planning and simulations help to cope with product changes, smaller batch sizes and fluctuating demand without jeopardizing the robustness of OT. Employees in maintenance, quality and production have pragmatic assistance systems at their disposal instead of a black box that dictates non-transparent decisions.

The era of isolated AI projects is over

Because of these numerous advantages, more and more companies want to use industrial AI on a broad scale. This brings with it a number of challenges. The days of isolated AI projects are over; instead, industrial companies need reusable IT architectures that allow them to roll out individual applications, such as visual quality inspection after the pilot line, across entire plants, regions and OEM networks.

The industrial edge is becoming increasingly important. As part of the broad roll-out, more and more AI must be brought to where the data is generated: directly to the machine or production cell. The cloud will continue to play an important role, especially for the training and further training of AI models. However, the productive use of these models often has to take place directly on site because latency, bandwidth, availability and data sovereignty require it.

In addition to scalability, IT security and IT governance are also important requirements when setting up an appropriate IT architecture. With every additional networked system, the attack surface for cyber attacks increases. In order to secure the scaling of industrial AI, security concepts such as zero trust, network segmentation and centrally enforced security guidelines must be taken into account and integrated right from the design stage of the IT architectures. Subsequent security add-ons are not sufficient because they are usually only able to protect existing data flows and interfaces incompletely and therefore often lead to security gaps.

But new organizational requirements are also emerging. The integration of AI into operational technology requires close collaboration with data scientists. These usually have a completely different technical background and generally do not speak the same language: data scientists rarely understand cycle times and maintenance windows, while there are generally no AI experts in the OT teams. Industrial companies therefore need platforms that provide technical support for collaboration between the two parties instead of making it more difficult.

Develop, test and roll out applications

For the widespread use of industrial AI, companies are best advised to set up an AI factory: an IT infrastructure with which new use cases can be efficiently developed, tested and rolled out. This infrastructure brings together AI, data and edge workloads on a consistent platform, tailored to the requirements of industrial OT environments.