Fraunhofer Institute of Optronics, System Technologies and Image Exploitation IOSB
Fraunhofer Institute of Optronics, System Technologies and Image Exploitation IOSB

AI engineering, process optimization

AI Systems Engineering addresses the systematic development and operation of AI-based solutions as part of systems that perform complex tasks.



The design and engineering of complex systems that contain AI and ML components differ from classical engineering, which only uses clearly described components whose behavior can be predicted relatively accurately in advance. Systems with machine learning and decision-making capabilities, on the other hand, may only reveal their final behavior or functionality at runtime, depending on the data. Nevertheless, such systems with intelligent components must be designed in such a way that reliable predictions can be made about their behavior during runtime and guarantees can be given.

AI Systems Engineering, as a discipline complementing basic research into AI methods, makes the use of artificial intelligence systematically accessible and available to engineering. Particularly with regard to the possible certification of AI systems (especially with regard to functional safety, IT security, and privacy), a reliable description of AI and ML components and systems is necessary in order to be able to plan AI and ML components accurately during system design, i.e., at the design stage.

Through the AI Systems Engineering methodology (PAISE® - Process Model for AI Systems Engineering) developed under the leadership of our institute and the use of appropriate tools, we address the systematic development and operation of AI-based solutions as an integral part of complex systems for performing demanding tasks.

Expand all Close all

  • The goals of AI Systems Engineering are:

    • To enable the use of AI within the systematic approach of (software) engineering disciplines.
    • To develop methods, tools, and processes to support the development of AI engineering solutions. This includes a formal characterization of the performance of AI solutions at the time of development (as opposed to purely statistical considerations of empirical performance).
    • The establishment of AI Systems Engineering as a new discipline that combines computer science, data-based modeling and optimization with systems engineering and classical engineering disciplines.

  • An essential aspect of process optimization is taking into account unobserved changes in environmental conditions. AI-supported systems must be able to recognize dynamic changes and respond adaptively to them in order to ensure continuous process stability and efficiency.

    AI is integrated into process control systems in several group steps:

    1. Quality prediction (software sensor): Here, AI serves as a prediction tool that provides precise assessments of process quality and detects potential deviations at an early stage.
    2. Human in the loop: In this step, human operators support AI in decision-making processes by providing feedback and making adjustments.
    3. Human on the loop: AI monitors the process independently, while humans only intervene when necessary to control unexpected situations or deviations.
    4. Human out of the loop: In highly automated systems, AI acts completely autonomously. Here, AI takes over both quality prediction and control response, steering the process fully automatically toward a defined goal.

  • Europe is striving to promote the use of AI in line with European values and laws and to set its own market priorities (GDPR, AI Act, Data Act, etc.). The reliable operation of AI systems is becoming an increasingly crucial issue. Securing AI systems is therefore becoming a key competence for the successful use of AI in critical technical applications. AI Systems Engineering offers the appropriate methodology to meet these challenges.

    In addition, AI Systems Engineering helps to ensure that AI components are actually ready for productive use and can be operated successfully on an ongoing basis. This is because AI projects often reach the status of an impressive prototype but never progress beyond this stage. There is a lack of reliable performance forecasting, especially for longer-term operational use. Concepts for dealing with data drift, changing environmental conditions, and other changes are required. All of this is addressed in AI Systems Engineering.

Projects

CC-KING: Competence Center AI Systems Engineering

The Karlsruhe Competence Center for AI Systems Engineering bridges the gap between cutting-edge AI research and established engineering disciplines. It conducts fundamental research and develops tools to facilitate the use of artificial intelligence (AI) and machine learning (ML) methods in business practice. Its scope focuses on industrial production and mobility.  

Find out more

AI Alliance Baden-Württemberg Data Platform

The “Data Platforms” subproject of the AI Alliance Baden-Württemberg lays the technical and organizational foundation for companies and start-ups to gain low-threshold access to data and AI models from others. The aim is to establish a market for data that promotes the development and application of innovative AI solutions. The project is funded by the Baden-Württemberg Ministry of Economic Affairs.

AI Alliance Baden-Württemberg AI Challenge

The aim of the AI Alliance Baden-Württemberg's “AI Challenge” sub-project is to methodically identify areas of application for AI and tap into potential for new business models. To this end, regional workshops bring together expertise and perspectives from different fields. Practical knowledge, scientific approaches, and the experience of users and manufacturers of AI systems are combined to work together on specific challenges. The project is funded by the Baden-Württemberg Ministry of Economic Affairs.

AutoLern

The AutoLern project focuses on drift management for production process data and uses machine learning (ML) to increase the efficiency and longevity of ML models in industrial environments. Data and concept drifts caused by factors such as tool wear or seasonal fluctuations can significantly impair model performance. AutoLern implements both performance-based and distribution-based drift methods for early detection and adaptation to such changes. Through continuous monitoring and automatic model adaptation, the project ensures that ML models remain accurate and adaptable, even in dynamic production environments.

AI Systems Engineering Solutions

 

PAISE®

AI projects in domains such as mobility or industrial production are usually complex, require heterogeneous teams, and carry a high risk of failure. The “Process Model for AI Systems Engineering” describes how they can nevertheless be made a success.

PAISE® Whitepaper (pdf)
 

Explanatory video about PAISE®

The Process Model for AI Systems Engineering, or PAISE for short, is our process model for the systematic and standardized development and operation of AI-based system solutions. This video explains what it does.

Video and further information
 

CHAISE

From data integration, model development and training to continuous monitoring during operation: CHAISE, the tool chain for AI Systems Engineering, offers a software toolbox for implementing AI applications that is perfectly matched to PAISE.

Chaise Project Page