Cognitive Power Electronics: AI for data-driven system optimization

The Data Analytics group develops innovative AI-based solutions to get the most out of data generated in the context of smart power electronics and Industry 4.0. The basis for this is an application-oriented approach that includes system analysis, design, data acquisition, filtering, clustering, and finally the development and implementation of intelligent algorithms in embedded systems or in industrial processes. Important for viable solutions in the dynamic field of data analytics and AI is a close collaboration with relevant teams and institutions, e.g. the Modeling and Artificial Intelligence department of IISB or the ADA Lovelace Center, of which Fraunhofer IISB is a founding member.

From Fraunhofer IISB's core topic Power Electronic Systems, two main areas for the application of data analytics solutions were chosen:

• Cognitive Power Electronics: Intelligent Power Electronics
• Dr. Production: Intelligent Manufacturing Equipment

In the research area Cognitive Power Electronics (CPE), Fraunhofer IISB combines its core competence in the field of power electronic systems with data analytics and artificial intelligence. This integration enables innovative applications for intelligent power electronics: from data acquisition and analysis in the converter to predictive maintenance in interaction with the cloud.

Cognitive Power Electronics is made possible by linking data knowledge with system knowledge within Fraunhofer IISB: Expertise in the field of power electronics regarding the conversion, supply and storage of electrical energy is incorporated into the intelligent functionalities. Examples of current developments are inverter-based health monitoring functions for electric drives without additional sensors, converter-based impedance spectroscopy and stability optimization in DC networks.

Use Case of CPE: Intelligente Drive Technology

One use case of CPE is electrical drive technology, where the drive inverter is modified and the electrical parameters of the inverter are used for predictive analysis. The electric drive thereby becomes an integrated intelligent system that can provide information about its current and future operating state and can also (re-)act independently.

Use Case of CPE: Intelligent DC grids

Many renewable energy sources supply electrical energy in direct current. Energy storage systems and consumers are also mostly DC-based. By intelligently interconnecting them in DC grids, AC-to-DC conversions are reduced, saving energy. AI-based solutions for smart DC grids are based on learning with sparsely annotated data, sequence-based learning, and mathematical optimization.

Temperature sensors in electric car batteries may not measure reliably, for example, when the attachment at the measurement point of a battery has degraded. To ensure that such dangerous situations are detected quickly, we developed a highly effective, tried-and-tested AI-based solution within the European project AI4CSM. The method employs an intelligent algorithm that monitors the condition of conventional temperature sensors, which lack the capability of self-supervision, and identifies potentially defective sensors. The approach is based on dynamic mode decomposition and utilizes only the data that is already captured by the available sensors in the battery system. Temperature deviations that are far lower than the range of the temperatures occurring in the battery system are sufficient for the algorithm to detect defective sensors. This combination of available sensor data with machine learning and AI-techniques makes the intelligent temperature sensor monitoring an excellent example for the IISB's Cognitive Power Electronics (CPE) concept.

Dr. Production comprises a toolbox of methods and algorithms that has been developed over two decades of research on automated process control in semiconductor manufacturing. It consists of three modules for application:

1. Application-oriented consulting and conception
2. Analysis of manufacturing processes and data acquisition
3. Development and implementation of intelligent algorithms

Depending on the requirements, all or only selected modules can be run through to develop a customer-specific solution. The diverse optimization options include, for example, condition monitoring of equipment, fault detection and classification, or predictive methods such as predictive maintenance and virtual metrology.

Use Case of Dr. Production: Rapid prototyping of electron devices

As part of the iDev4.0 (Integrated Development 4.0) project, Fraunhofer IISB has developed components for an intelligent platform for prototype and small series production. Here, customer-specific electron devices are developed in the sense of flexible rapid prototyping. A central element is the optimization of the continuous Si-CMOS and SiC process line (π-Fab®) operated by Fraunhofer IISB by learning from existing manufacturing data.