Non-destructive evaluation (NDE) is highly relevant to assessing micro- and macrostructural changes in ferritic and ferritic/martensitic steels subjected to high temperature loading. These materials are widely used in energy generation, where they undergo extreme thermal and mechanical loads. This study examines the feasibility of micromagnetic NDE techniques, i.e., micromagnetic measurements, supported by machine learning methods, to identify and characterize the micro- and macrostructural changes caused by the mechanical loading at high temperatures of power plant steels, i.e., ferritic/martensitic P91 and the high chromium ferritic steel HiperFer-17Cr2. While the P91 did not show any systematic changes in micromagnetic measurements, which generally correlate with the evolution of the microstructure and the mechanical properties, for the HiperFer-17Cr2, pronounced changes in the micromagnetic properties were observed. In correlation with the evolution of the hardness and cyclic deformation behavior, which are both mainly attributed to Laves phase precipitation, the micromagnetic measurements significantly changed depending on the temperature, number of load cycles and load amplitude applied. Thus, these NDE methods can be used for early diagnosis and preventive maintenance strategies for HiperFer-17Cr2, potentially extending the lifespan of the components and mitigating safety risks.

Early detection of material degradation is critical for ensuring structural reliability. The majority of conventional evaluation methods require both normal and defective reference samples. This increases the effort needed for calibration since producing defective samples is challenging, costly, and impractical. Therefore, in this work, an algorithm for learning the normal behavior directly from the data, without requiring predefined anomalous samples was developed. A self-learning anomaly detection algorithm is implemented. It is based on a dense autoencoder trained solely on micromagnetic data representing frequently occurring conditions. The autoencoder learns a data-driven representation of the undamaged state and detects deviations during the measurement. A dynamic thresholding strategy is applied to adapt to changes in the feature distribution, enabling the detection of early anomalies without manual limits. The results show that the method can detect subtle variations in micromagnetic responses before any visible failure or load reduction occurs. This capability enables continuous monitoring and allows the approach to adapt to evolving material behavior.

Nachhaltigkeit ist eines der wichtigsten Ziele der heutigen Gesellschaft, mit erheblichem Einfluss auf wirtschaftliche und politische Entscheidungen. Eine wichtige Initiative in dem europäischen Green Deal, der eine Strategie für Nachhaltigkeit darbietet, ist der Aktionsplan für die Kreislaufwirtschaft (CEAP). Ziel ist es, den ökologischen Fußabdruck der EU zu reduzieren und die Verwendung von Recyclingmaterialien in den nächsten zehn Jahren zu verdoppeln. Konkrete Maßnahmen wurden in verschiedenen Bereichen, wie Elektronik, Verpackungen, Kunststoffe und Textilien eingeleitet. Ein wichtiges Segment einer Kreislaufwirtschaft ist die Sortierung und das Recycling von Materialien. Um die Qualität des Recyclings zu steigern, sind Verbesserungen in der Abfallsammlung und –Sortierung erforderlich. Die derzeit auf dem Markt erhältlichen Sensorsysteme für die Sortierung von Kunststoffen in der Abfallwirtschaft basieren hauptsächlich auf Nah-Infrarot (NIR) und Kurzwellen-Infrarot (SWIR). Die Sortierung schwarzer Kunststoffe, einschließlich solcher aus dem Automobilsektor, bleibt für diese Systeme jedoch eine Herausforderung. Im Rahmen des Fraunhofer-Leitprojektes Waste4Future wurde ein aktives Thermografiesystem entwickelt, um verschiedene Kunststoffmaterialien zu unterscheiden. Dieses System wurde erfolgreich verwendet, um schwarze Kunststoffe auf einem Förderband zu klassifizieren. Die Analyse der Abkühlungskurven der Proben ermöglichte, aufgrund unterschiedlicher Erwärmungs- und Abkühleigenschaften, die Unterscheidung der Materialien. Mit der Verwendung maschineller Lernalgorithmen konnte ein pixelweiser f1-Score von über 97 % unter Verwendung von drei ganzen Proben je Material als Testdaten erzielt werden. Die Erkenntnisse aus der aktiven Thermografie eröffnen vielversprechende Möglichkeiten für die Integration in die aktuelle Sensorik zur Sortierung schwarzer Kunststoffe, ein wichtiger Schritt in Richtung einer nachhaltigeren Kreislaufwirtschaft.

Wind‑turbine decommissioning generates an inventory of at least 2,500 high‑strength bolted connections per unit that remain valuable for reuse, yet their reclaimability hinges on a detailed assessment for defects. Conventional NDE methods struggle to scale across such a large number of fasteners, and the presence of subsurface cracks, corrosion, or geometric deformations demands rapid, contact‑less inspection. This work presents an automated evaluation that fuses infrared thermography with high‑resolution optical imaging. Thermal data expose hidden cracks, while optical images capture surface anomalies and dimensional deviations such as bending or pinching. The two modalities are combined in a single workflow that generates defect masks and forwards the results to an AI classifier that determines bolt reuse eligibility

This work focuses on an AI-based data analysis for the robot-guided magnetic particle inspection (MPI) developed as part of the SeConRob project. The primary aim here is to create deep learning-based methods to detect defects in MPI images of real components using a robot-guided camera. The project in total strives to integrate defect detection algorithms into the robotized production environment. This contribution covers the practical application, the image acquisition system, the individual analysis regarding defects, and the evaluation results. The deep learning challenge comprises semantic image segmentation. The approach involves classifying each pixel of the image as either “defect” or “non-defect”. A ResNet based encoder with a U-Net style decoder is employed for the segmentation. ImageNet pretrained weights are leveraged as an initialization for transfer learning. The training datasets include both laboratory-generated images and images captured by robots in a real production environment.