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

Generative AI

Unlike traditional machine learning approaches, generative artificial intelligence (AI) not only enables data to be classified, but also generates new data independently according to specific specifications. Large language models (LLMs) in particular demonstrate the potential of generative AI in everyday life by summarizing, structuring, and formulating texts.

Multimodal models or vision-language models (VLM) are also capable of capturing and generating image data or videos. At Fraunhofer IOSB, the potential for using generative AI in a wide range of industries and throughout the entire product development cycle is being researched and put into practice.

Examples include automated quotation generation, PLC development, machine maintenance support through an LLM-based assistance system, anonymization and augmentation of data sets, and human-robot interaction using natural language. At the Fraunhofer High-Performance Center Engineering Automation, we focus on the use of generative AI to increase efficiency and quality in engineering processes.

Projects

ChatPLC: Generation of control code

The growing shortage of skilled workers and the need for profitable, scalable solutions are bringing generative AI into focus in the field of automation technology. Together with Phoenix Contact, Fraunhofer IOSB-INA has trained a local language model for generating structured text (ST) in accordance with IEC 61131-3. In another use case, an AI assistant was developed specifically for operating engineering software. These examples demonstrate the potential that is already tangible through the use of language models in the field of PLC programming.

SyDaPro: Synthetic data in production

In industrial production facilities, AI enables predictive maintenance or control and optimization of production processes. A major challenge lies in providing the large amounts of data required for training process models. SyDaPro has therefore dedicated itself to generating synthetic production data based on stochastic models and deep learning methods. This has increased the availability of training data, enabling data-based optimization of production processes and the prevention of anomalies.

Find out more

FA³ST CreAItor

We are developing FA³ST CreAItor, a user-friendly system for creating, validating, and improving Asset Administration Shells with support from large language models for standardized digital twins in all EU languages. Our goal is to enable non-experts to create standards-compliant Asset Administration Shells that are correct, complete, and consistent. This will save time and money for SMEs that want to develop a digital twin or an AAS-compliant digital product passport for their product, for example.

More about FA³ST