Abstract

Abstract
The growing demand for high-quality components in various industries, particularly in the automotive sector, requires advanced and reliable inspection methods to maintain competitive standards and support innovation. Manual quality inspection tasks are often inefficient and prone to errors due to their repetitive nature and subjectivity, which can lead to attention lapses and operator fatigue. The inspection of reflective aluminum parts presents additional challenges, as uncontrolled reflections and glare can obscure defects and reduce the reliability of conventional vision-based methods. Addressing these challenges requires optimized illumination strategies and robust image processing techniques to enhance defect visibility. This work presents the development of an automated optical inspection system for reflective parts, focusing on components made of high-pressure diecast aluminum used in the automotive industry. The reflective nature of these parts introduces challenges for defect detection, requiring optimized illumination and imaging methods. The system applies deep learning algorithms and uses dome light to achieve uniform illumination, enabling the detection of small defects on reflective surfaces. A collaborative robotic manipulator equipped with a gripper handles the parts during inspection, ensuring precise positioning and repeatability, which improves both the efficiency and effectiveness of the inspection process. A flow execution-based software platform integrates all system components, enabling seamless operation. The system was evaluated with Schmidt Light Metal Group using three custom datasets to detect surface porosities and inner wall defects post-machining. For surface porosity detection, YOLOv8-Mosaic, trained with cropped images to reduce background noise, achieved a recall value of 84.71% and was selected for implementation. Additionally, an endoscopic camera was used in a preliminary study to detect defects within the inner walls of holes. The industrial trials produced promising results, demonstrating the feasibility of implementing a vision-based automated inspection system in various industries. The system improves inspection accuracy and efficiency while reducing material waste and operator fatigue.

Abstract
Breakthroughs in sequencing technologies led to an exponential growth of genomic data, providing novel biological insights and therapeutic applications. However, analyzing large amounts of sensitive data raises key data privacy concerns, specifically when the information is outsourced to untrusted third-party infrastructures for data storage and processing (e.g., cloud computing). We introduce Gyosa, a secure and privacy-preserving distributed genomic analysis solution. By leveraging trusted execution environments (TEEs), Gyosa allows users to confidentially delegate their GWAS analysis to untrusted infrastructures. Gyosa implements a computation partitioning scheme that reduces the computation done inside the TEEs while safeguarding the users' genomic data privacy. By integrating this security scheme in Glow, Gyosa provides a secure and distributed environment that facilitates diverse GWAS studies. The experimental evaluation validates the applicability and scalability of Gyosa, reinforcing its ability to provide enhanced security guarantees.

Abstract

Abstract
Data-based approach for diagnosis of thyroid disorders is still at its early stage. Most of the research outcomes deal with binary classification of the disorders, i.e. presence or not of some pathology (cancer, hyperthyroidism, hypothyroidism, etc.). In this paper we explore deep learning (DL) models to improve the multi-class diagnosis of thyroid disorders, namely hypothyroid, hyperthyroid and no pathology thyroid. The proposed DL models, including DNN, CNN, LSTM, and a hybrid CNN-LSTM architecture, are inspired by state-of-the-art work and demonstrate superior performance, largely due to careful feature selection and the application of SMOTE for class balancing prior to model training. Our experiments show that the CNN-LSTM model achieved the highest overall accuracy of 99%, with precision, recall, and F1-scores all exceeding 92% across the three classes. The use of SMOTE for class balancing improved most of the model’s performance. These results indicate that the proposed DL models not only effectively distinguish between different thyroid conditions but also hold promise for practical implementation in clinical settings, potentially supporting healthcare professionals in more accurate and efficient diagnosis. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract