Screw fastening is one of the most important manufacturing processes in the aerospace industry. For example, millions of fasteners are utilized in the Boeing 777 to connect around 3 million different structural components. However, screw fasteners mounted by articulated arm robots in the assembly of aircraft parts are subject to positional uncertainties in both dynamic and static applications. The resulting production errors not only have an impact on cost and time but also lead to safety risks. Despite this strong theoretical and practical relevance, only few research approaches focus on the recognition of defective fasteners in aircraft manufacturing, and the detection of incorrectly assembled aeronautical threaded fasteners remains open. To address these challenges, we identified important time series channels and developed a hard voting ensemble of Categorical Boosting (CatBoost) and Extreme Gradient Boosting (XGBoost) to predict the outcomes of screw fasteners mounted by articulated arm robots in the assembly of aircraft parts. Our model achieved a balanced accuracy of 81.08% on a time series dataset containing both kinematic and dynamic data obtained from a robotic assembly process of aeronautical screwed fasteners. While both CatBoost and XGBoost reached high performance metrics individually, the ensemble approach outperformed each model by utilizing their complementary strengths. Overall, these findings underscore the need for robust machine learning methods to enhance safety standards and reliability in critical assembly processes through advanced techniques.

For highly sensitive products such as pharmaceuticals, quality is a decisive factor in ensuring the therapeutic benefit that consumers expect and not jeopardizing consumers' health. So far, the quality control of pharmaceuticals is largely performed manually by qualified individuals. However, this is a time-consuming, repetitive, and error-prone process subject to natural performance fluctuations. To contribute to addressing this issue, we present an automated quality control approach for pharmaceutical capsules using a transfer learning-based convolutional neural network with a balanced accuracy of 97.27%, outperforming all current benchmarks. To increase trust in the model predictions, we incorporated two explainable artificial intelligence (XAI) methods into our approach.

In this paper, a literature review of recent advances in machine learning approaches for cyber defense is presented. Relevant articles in the databases ACM DL, IEEE Xplore DL, and ScienceDirect were identified and supplemented by forward and backward searches. In total, 70 articles were identified to meet the scope of the literature review. The following article gives an overview of classifications, datasets, and algorithms of machine learning in cyber defense. Limitations and future research areas are identified.

The consequences of alcohol use disorders affect around 17 million people in the United States. To prevent healthy people from developing an alcohol use disorder, early detection is achieved using different screening methods. However, these methods are mostly based on self-tests, which can be easily influenced by the subject. In order to prevent healthy people from developing an alcohol use disorder through their alcohol consumption, drinking behavior, and alcohol-related problems, we propose a novel machine learning approach. With this approach it is possible to classify healthy people with an accuracy of 69 percent based on EEG recordings in assessing the danger of developing an alcohol use disorder or not. To obtain this result, the frequency range of the EEG data used was divided into 99 fine bands. Using a machine learning algorithm, the five most important bands were identified.

One of the biggest challenges in pain therapy is the evaluation of pain intensity of patients who cannot express their pain verbally. The measurement of pain is a time-consuming process. Therefore, we investigate the main factors that influence pain in infants. We use facial expressions and random forest to predict pain in the three pain classes aligned with the Premature Infant Pain Scale (PIPP). Using our method, we can classify between low, middle, and high pain with a balanced accuracy of 95.51 percent, setting the new benchmark for three-level pain assessment. Our classifier does not misclassify the highest pain class, which is desirable for the classification of pain. Our algorithm has the potential for creating more time for workers that are constantly under time-pressure.

In recent years, the demand for gesture recognition has increased enormously due to many applications such as computer games, human-robot interaction, assistance systems, sports, sign language interpreters, and e-commerce. The recognition of hand gestures is one of the most important gesture recognition methods. With simple hand gestures, devices in the smart home area (TV, radio, vacuum cleaner robots, etc.) should be easier to operate. Our method is based on a convolutional neural network, or more precisely on MobileNetV2. With this lean and fast network, we have been able to achieve an accuracy of 99.96 percent in recognition of hand gestures, so that in the future, we will be able to offer an application in the field of Human-Computer Interaction to interact more easily with the ever-increasing number of technologies in everyday life.

In the production of screws, manual methods are often still used to detect defects. This paper aims to use a convolutional neural network-based technique to detect whether defects in screws are caused during production. Our experimental results show that a detection accuracy of 96.67% can be achieved with the proposed technique. Among the defects considered are defects on the objects' surface (e.g., scratches, dents), structural defects like distorted object parts, or defects that manifest themselves by the absence of certain object parts. Our more efficient method can be used in the future for quality control in the manufacture of screws.

Since the human inspection of small metal parts is complex, time-consuming and prone to human error, a convolutional neural network for the detection of defects on metal nuts was developed in order to grant fast and robust quality controls. For this approach, we built an image classification algorithm based on the Xception architecture. The evaluation of the trained model is robust and achieves reliable results after applying a hold-out 5-fold cross-validation. Implementing this algorithm on the MVTec Anomaly Detection dataset outperforms the existing benchmark on defect detection for metal nuts with a balanced accuracy of 90.00% and a value of 0.99 for the area under the curve.

The dependence of convolutional neural networks on large-scale datasets for training is no secret. This is even more problematic when using 3D-CNNs since sufficient 3D datasets for training are scarce and expensive. One possible solution is transfer learning. In this comparison, the state-of-the-art techniques for 3D-CNNs are analyzed and compared. Therefore, a literature search in the databases IEEEXplore DL, ScienceDirect, SpringerLink, and ACM is conducted. The results are compared using the criteria field of application, datasets, 3D-CNN architecture, transfer learning technique, hyperparameters, and final performance. This comparison provides a basis for future work to promote understanding and usage of transfer learning for 3D-CNNs.

Intellectual and developmental disorder results in lifetime impairments in cognitive capabilities and adaptability is highly prone to developing several forms of further diseases and is associated both with high care costs. In this study, we applied a machine learning approach for differentiating healthy individuals and individuals with intellectual and developmental disorder using resting-state electroencephalography recordings. Our approach sets a new benchmark with a balanced accuracy of 91.67%. In addition, by adopting novel model interpretability methods, we highlighted the low beta sub-band in the range of 19.5–21 Hz as the most important distinctive feature. Individuals with an intellectual and developmental disorder show significantly lower beta activity.

Based on a very large dataset of over 100 million Steam platform users we present the first comprehensive analysis of hardcore gamer profiles by over 700,000 hardcore players (users playing more than 20 hours per week) covering more than 3,300 games. Using an unsupervised machine learning approach we reveal the specific behavioral categories of hardcore players, i.e. First Person Shooter, Team Fortress 2 player, Action game player, Dota 2 player, Strategy and action combiner, Genre-switching player. Subsequently we derive individual patterns of hardcore gamers in the categories found, such as strategy-action games combiner or game switching players. Our results are useful for computer science and information systems scholars interested in individual differences in user behavior as well as practitioners interested in game-designing.