Artificial Intelligence and Data Management

Our work in the "AI and Data management" Research Topic is divided into the fields of data science and data engineering. In the field of data science, we develop artificial intelligence models for the quality inspection of workpieces and production processes along the PV value chain. We integrate our expertise in AI and solar cell physics in equal measure to develop fast, robust and interpretable AI models based on our extensive measurement technology portfolio.

We develop the latest AI methods for our partners in measurement technology and software development as well as solar cell production, such as (a) computer vision models for classifying and segmenting defects and objects in 2D and 3D, (b) methods of theory-based data analysis and semantic data compression for analyzing and efficiently storing high-dimensional data in the form of a digital twin and (c) generative models for developing accelerated and high-resolution measurement processes, e.g. for high-throughput production. 

In the field of data engineering, we develop metadata models, interfaces and software systems for monitoring complex production processes, for the automated acquisition and structuring of data and their provision in modern database networks and test the concepts and systems in the laboratory at Fraunhofer ISE. Current technologies and concepts of software development are used, such as micro-service architectures, container concepts, virtualization platforms and DevOps strategies.

The data science and data engineering competencies originate from PV cell production, but are technology-independent and can therefore be used in a variety of application fields.

... the following range of methods is used and continuously developed at Fraunhofer ISE:

• Toolbox with AI models for unsupervised and supervised learning methods
• Tools for rapid deployment of the developed AI methods
• Powerful NoSQL database technologies: graph databases for storing metadata and document databases for storing production data
• Tools for innovative management of context data in the form of dynamically adaptable, »growing« metadata models
• Software architectures based on micro-services to ensure (a) easy functional extensibility, (b) high scalability and (c) clear separation of responsibilities between the individual software modules
• Consistent application of the Docker container concept in programming in order to be able to operate the individual services in a virtualization environment
• Comprehensive quality assurance of the code through (a) a defined and automated Development & Operations (DevOps) strategy using Gitlab, (b) extensive unit tests to validate the functional requirements, (c) the Git Flow release management strategy for structured versioning, (d) automated deployment using the Kubernetes package manager »Helm«

... this infrastructure is available to us at Fraunhofer ISE:

• Extensive GPU server infrastructure for training the AI models
• Inline and reference measurement technology for efficient data collection and model training
• Kubernetes server cluster as a high-performance virtualization platform for (a) fully automated, hardware-independent rollout of software components, (b) efficient support during operation and (c) protection of the production software against hardware failures

All Research Projects on the Topic "Artificial Intelligence and Data Management"