Democratising AI through comprehensible, easily accessible machine learning operations (MLOps)
Developed in collaboration with senswork GmbH and eresult GmbH and under the auspices of the Fraunhofer Institute for Integrated Circuits IIS, inovex’s DeKIOps project aims to make AI tools usable for all companies by eliminating the need for in-house AI expertise.
Project aims – guidelines and demonstration projects
The collaboration team plans to have general guidelines in place and two demonstration projects in use in industrial applications by the end of 2025. The guidelines for developing readily comprehensible, easy-to-maintain machine learning systems are intended to enable users without AI expertise to operate and maintain such systems. The overarching aim of the research project is to democratise machine learning systems and to enable low-threshold access to machine learning solutions. In this way, the project aims to combat the shortage of machine learning experts.
The development guidelines are based on technical and user requirements, with generalisable aspects being prioritised. The resulting generic MLOPs framework will be evaluated based on demonstration projects from two different industrial manufacturing use cases. Following their successful evaluation, the development guidelines and the system concept will be published and made available via open source platforms.
For the field trial, senswork, an expert in image processing and AI, is developing a vision AI system which can be operated and maintained by users with no specialist AI knowledge. Here, the focus is on usability – both software ergonomics and UX design – in addition to the use of innovative AI technology. The project aims to significantly lower the entry threshold for AI in the field of vision technology. As a result, it will also help to combat the shortage of AI experts.
In order to ensure the universal applicability of the framework developed, inovex is also studying a second use case, this one in the field of time series analysis and predictive maintenance. eresult is lending its usability expertise to the project.
Demonstration projects at supply companies
The first demonstration project will test a wide range of products in the large-scale production department of a supply firm. This visual quality assurance process focuses on detecting product flaws and requires the development of a visual, automated, AI-based quality assurance (computer vision) solution.
The second case also involves quality assurance. Here, however, the focus is on the associated predictive maintenance requirements. The solution developed will be evaluated by the application partner’s end users on the basis of various metrics, including user-friendliness and comprehensibility.
Deriving general systems from real-world projects
In order to make the MLOps systems developed for the two different use cases more universally applicable, the target function, as well as a suitable training method which depends on the type and quantity of the data, is defined before modelling. The results of the machine learning models are then converted to a comprehensible format to enable even end users with no expert knowledge to achieve useful results and to interact with the system according to their requirements.