The growing share of renewable power generation leads to increasingly fluctuating and generally rising electricity prices. This is a challenge for industrial companies. However, electricity expenses can be reduced by adapting the energy demand of production processes to the volatile prices on the markets. This approach depicts the new paradigm of energy flexibility to reduce electricity costs. At the same time, using electricity self-generation further offers possibilities for decreasing energy costs. In addition, energy flexibility can be gradually increased by on-site power storage, e.g., stationary batteries. As a consequence, both the electricity demand of the manufacturing system and the supply side, including battery storage, self-generation, and the energy market, need to be controlled in a holistic manner, thus resulting in a smart grid solution for industrial sites. This coordination represents a complex optimization problem, which additionally is highly stochastic due to unforeseen events like machine breakdowns, changing prices, or changing energy availability. This paper presents an approach to controlling a complex system of production resources, battery storage, electricity self-supply, and short-term market trading using multi-agent reinforcement learning (MARL). The results of a case study demonstrate that the developed system can outperform the rule-based reactive control strategy (RCS) frequently used. Although the metaheuristic benchmark based on simulated annealing performs better, MARL enables faster reactions because of the significantly lower computation costs for its own execution.

Industrial load management plays an important role in the balance of energy consumption and electricity generation, which is increasingly fluctuating due to the growing share of renewable energies. Manufacturing companies are able to adapt their energy consumption by considering energy aspects in their production schedule. It may even be beneficial to temporarily force production resources into idle states and to thus reduce energy demand for a limited period. However, the resulting scheduling problem is very complex and at the same time should be executed in real-time. This paper presents an approach for industrial load management using multi-agent reinforcement learning for energy-oriented rescheduling. A simulation study serves to validate the approach. The results show good solutions and at the same time low computational expense compared to a metaheuristic approach using simulated annealing.

Multi-sine excitation signals give spectroscopic insight into fast chemical processes over bandwidths from 10¹ Hz to 10⁷ Hz. The crest factor (CF) determines the information density of a multi-sine signal. Minimizing the CF yields higher information density and is the goal of the presented work. Four algorithms and a combination of two of them are presented. The first two algorithms implement different iterative optimizations of the amplitude and phase angle values of the signal. The combined algorithm alternates between the first and second optimization algorithms. Additionally, a simulated annealing approach and a genetic algorithm optimizing the CF were implemented.

Machine learning has been finding its way into industrial systems with use cases like predictive maintenance, process optimization, anomaly detection, and many more. Low data coverage limits the usability of these machine learning models to only a small fraction of the overall parameter space. To make predictive machine learning models feasible for industrial processes, it is important to collect relevant data to reduce epistemic uncertainty in all areas of the parameter space. Approaches from experimental design aim to patch areas with low data coverage and provide useful information for previously unknown process settings. These methods assume that the factor limits in the parameter space are known. However, the high-dimensional parameter spaces in industrial processes with complex factor interdependencies make it difficult to find the dynamic physical limits of the system. Therefore, it is hard to define a set of experiments that are physically possible to reach. We provide a novel method to explore these high-dimensional parameter spaces. Through in-distribution data projection and out-of-distribution predictions with autoencoders, we estimate the physical limits of the parameter space of industrial processes.

The increasing popularity of wind energy has led to an extensive growth in wind turbine installations. This poses challenges when these wind turbine blades reach the end of their operational lifespan. Thus, suitable recycling techniques are required to extract the different materials from the turbine blades. Therefore, we explored the use of deep learning-based object detection for sorting shredded End-of-Life wind turbine blades into individual material classes. To reach this goal, a custom dataset combining images from shredded wind turbine blades, with a synthetic data generation approach, was employed. Three popular object detection architectures (SSD, YOLO, and Faster R-CNN) were implemented and tested. The impact of different popular backbone networks and Feature Pyramid Networks on accuracy and speed was analyzed. Our results showed that the SSD model with a ResNet18 backbone and a version of Feature Pyramid Network performed well in terms of accuracy and speed. Moreover, synthetic data generation proved useful but showed a performance decline when transitioning to real-world pictures, suggesting the need for combining synthetic and real-world data. This study highlights the potential of deep learning in recycling and sorting wind turbine blades, encouraging further research in this field.