The white paper, subtitled “Requirements for Long-Term Operations of AI in Industry,” structures the requirements that are crucial for project success “beyond the prototype” into six categories: degree of autonomy, performance, monitoring and maintenance, integration and deployment, acceptance, and regulatory compliance. It describes a structured approach for implementation in real OT/IT environments and establishes a common language between AI and domain experts.

Robustness, drift handling, maintainability, and traceability

“AI only creates added value in industry when it is used over the long term. That's why teams need to anticipate operational requirements—robustness, drift handling, maintainability, and traceability—during the development phase,” says Dr. Constanze Hasterok from Fraunhofer IOSB. “Our white paper makes these requirements explicit and shows how OT/IT integration, monitoring, and regulatory aspects such as the EU AI Act can be set up properly from the outset.”

The white paper provides specific technical and organizational levers for the path to production. “For AI solutions to make the leap from prototype to everyday use, more than just good models are needed—in addition to MLOps, stable data infrastructures and a clear strategy for how edge and cloud services interact are particularly important,” says Lukas Rauh from Fraunhofer IPA. “Only then will systems for AI solutions become scalable and truly practical.”

"A functioning prototype is not enough to ensure the long-term success of AI in industry. Companies need robust processes, clear roles, and an organizational structure that supports continuous operation,“ says Dr. Holger Kett from Fraunhofer IAO. ”Our white paper shows how methodical AI Systems Engineering, MLOps, and human-centered design interact so that AI solutions can be used reliably, transparently, and scalably in everyday life."

Key contents of the white paper

• Operational scenarios from “proof of concept” to “external, critical deployment environment”
• Requirements for AI systems: degree of autonomy, performance, monitoring & maintenance, integration & deployment, acceptance, regulatory compliance
• Performance & robustness of AI systems by design
• Supervision & maintenance, e.g., detection of data drift, (automated) retraining, versioning of data/models/systems
• Integration & deployment in different environments: OT/IT convergence, edge/cloud strategies, scalability, latency/bandwidth, IT security, federated and distributed learning
• Acceptance & human-centered quality with user roles, explainability, usability, etc.
• Classification according to the EU AI Act and other regulatory requirements

Availability and partners

The white paper is freely available on the websites of the participating institutes and initiatives as well as via Fraunhofer Publica (DOI: https://dx.doi.org/10.24406/publica-4825, license: CC-BY-NC-ND). It is aimed at AI experts, project managers, and technical decision-makers in companies that are transferring AI systems beyond prototypes into real OT/IT environments.

The three Fraunhofer Institutes involved in the development and available to answer any questions are the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB), the Fraunhofer Institute for Industrial Engineering IAO, and the Fraunhofer Institute for Manufacturing Engineering and Automation IPA. In addition, the AI Progress Center “Learning Systems and Cognitive Robotics” is involved as an applied branch of Cyber Valley, as is CC-KING, the Competence Center for AI Systems Engineering Karlsruhe.

Contact for press inquiries:

• Constanze Hasterok, Fraunhofer IOSB, constanze.hasterok@iosb.fraunhofer.de
• Holger Kett, Fraunhofer IAO, holger.kett@iao.fraunhofer.de
• Lukas Rauh, Fraunhofer IPA, lukas.rauh@ipa.fraunhofer.de