A total of 18 AI experts and researchers from the Institute for Modeling and High Performance Computing (IMH) of the Hochschule Niederrhein, the Bergische Innovationsplattform für Künstliche Intelligenz of the University of Wuppertal and Siemens Gas and Power GmbH & Co. KG met at the Technopark of Siemens Gas and Power GmbH & Co. KG in Mülheim an der Ruhr for the kick-off event on Jan. 17, 2020 of the three-year BMBF program in the FHprofUnt funding line with a funding amount of 1.2 million euros. The global working group "Probabilistic Design" of Siemens Gas and Power GmbH & Co KG, as a new independent power plant and energy division of Siemens, which will be listed on the stock exchange in less than a year, had invited to this workshop. In the new rooms of the DataLab of the Siemens Technopark, an interdisciplinary team consisting of mathematicians, physicists, computer scientists and engineers will from now on work on the project topic

"Multivariate machine learning algorithms for optimal design and real-time capable lifetime prediction of power plant components".

will be researched.

The strategic goals of the High-Tech Strategy 2025, which are, as it were, the guiding themes of the national sustainability strategy, include research into key technologies for economical, efficient and secure energy supplies. Against the backdrop of climate protection, this gives rise to new technical challenges. Increased requirements on emissions with regard to greenhouse-relevant climate gases must be taken into account, as must the guarantee of unrestricted security of supply even in the case of highly volatile electricity feed-in. These ambitious challenges are the direct focus of the research content of the joint research project, which envisages not only the optimal operation of individual plants using new, real-time AI algorithms, but also the sustainable integration of renewable energy sources into the grid infrastructure.

In order to reconcile economic efficiency and environmental compatibility, targeted, continuous and application-oriented research and development is necessary, particularly in the field of power plant technology. In this context, both gas and steam (CCGT) power plants in their wide output range and small-scale plants in the form of, for example, combined heat and power plants and their integration into so-called "smart grids" are predestined partners for renewable energies. Due to the increasing share of electricity from renewable energies, the demands on the flexibilization of the operation modes of conventional power plants are constantly rising. In view of this, the classic, deterministic design methods with conservatively accumulated worst-case assumptions are increasingly proving to be no longer economical. In order to be able to represent the changing prevailing operating conditions more accurately in the simulation, for example by means of a digital twin, extremely computationally intensive models are usually required. This numerical complexity and the associated computing time can be significantly reduced by using efficient machine learning algorithms of the IMH.

The novel approach developed in the three predecessor projects, which were funded by Siemens AG, the BMWi and the BMBF from 2013 onwards, for the explicit consideration of e.g. temporal and/or spatial correlations of multivariate simulation results such as stress or temperature fields, or also lifetime consumption rates, offers a considerable potential for improving the prediction quality in the design and optimization of components. Thus, e.g., a more realistic lifetime prognosis is made possible, which in turn is an essential basis for an optimization of the operation. Application areas of this kind are not limited to conventional power plant construction, but rather similar problems can also be found in all corresponding areas of energy management.

The scientific research of new mathematical methods is to be carried out at the HN with subsequent transfer and transformation of the research results to the energy sector of Siemens Gas and Power GmbH & Co KG, as the world's leading supplier of the complete spectrum of products and services for power generation.

The machine learning approaches to be developed enable ad-hoc optimizations, which thus offers the great opportunity to be used as measures in the large number of already existing plants. Thus, there is a great potential to increase the Energy Efficiency of existing plants and to reduce emissions in the short term.

Within the scope of the research project, selected combined cycle power plants and their components will be used to demonstrate that optimized design and operation can be realized. The new optimization methods will ultimately lead to an increase in the efficiency of turbomachinery with low emissions even at partial load, lowest life cycle costs and reduced investment costs per kW while ensuring high reliability and availability. Furthermore, the introduction to these design methods not only leads to a sustainable energy supply, but also to an increased competitiveness of the German turbomachinery industry with indirect and also direct effects.

On the basis of so-called cyber-physical systems (CPS), dynamic, real-time capable and self-organizing value chains of power plants are designed, which can be optimized according to various target variables, such as costs, availability, energy and resource consumption, flexibility, lead time, etc. CPS can use information and knowledge from real state and process data via software tools and simulate a digital, virtual image of the physical power plant systems via application-specific engineering applications. Every change to a component can be fed back into the virtual image.

Thus, physical products get an assigned, numerical model, which is also referred to as a digital twin in the Industrie 4.0 field of action. CPS can be integrated into components, machines and plants that can adapt to changing operating conditions through self-optimization and reconfiguration. Typical applications for such systems are condition monitoring and predictive maintenance. Machine learning is considered a key technology for the development of CPS. Machine learning deals with the automated development of metamodels based on empirical data or training data. An artificial system, such as the CPS learns from examples and can generalize them after the learning phase is complete. Instead of expensive prototypes and lengthy test chains, Digital Twins can be used to run through data-based optimizations or stochastic analyses within a very short time using machine learning algorithms, to develop and reject solution strategies, and to explore and implement possible improvements.

This is achieved in a scientific cooperation with Prof. Dr. Hanno Gottschalk, Executive Board of the Bergische Innovationsplattform für Künstliche Intelligenz. In addition, the cooperative doctorates envisaged in the project are to be carried out at the Bergische Universität Wuppertal. Together with other IMH researchers, Prof. Dr. Georg Vossen in the area of optimal control and Prof. Dr. Peer Ueberholz in the area of neural networks, the competencies will be expanded to form a joint core research area. Its goal is the development of safe, reliable and optimal designs, products and processes while taking into account the unavoidable, scattering effects and system properties.

The sustainable economic usability of these research results for Siemens Gas and Power GmbH & Co. KG is considered to be very high, since the methods to be researched can be used not only for the development of new, more efficient, more cost-effective and more durable power plant components, but furthermore for an improvement of any technical systems at all locations of Siemens Gas and Power GmbH & Co. KG.

The collaborative research project ML-Real-Time is financially supported by the German Federal Ministry of Education and Research (BMBF) through the VDI Technologiezentrum as project sponsor of the BMBF for the funding program FHprofUnt 2018 (sponsor code. 13FH174PX8) based on a resolution of the German Bundestag and by Siemens Gas and Power GmbH & Co. KG.