In diesem Beitrag wird die Anwendung des parallelkinematischen Manipulators Hexaglide in der THT-Bestückung untersucht. Kostengünstige optische und taktile Sensoren ermöglichen präzises Referenzieren und Einstecken von Bauteilen. Eine kraft- und bildgeregelte heuristische Steuerungsstrategie sowie Reinforcement Learning wurden erprobt und die Tauglichkeit des Hexaglide für die Bestückung wurde experimentell gezeigt.

Sustainability is the current global challenge. This is reflected in the demand for healthy food and CO $$_2$$ neutrality. These challenges can be met with the industrial cultivation of algae: Algae can be used as food supplements, nutraceuticals, pharmaceuticals, fuel, CO $$_2$$ sinks, and obtain high relative yield density per area. Current limitations in their large-scale use exists, as scaling up from laboratory environments to pilot applications typically requires more than 5 years, because of highly complex interactions in the growth behavior: They are influenced by current and past environmental conditions. These interactions make current pilot applications inefficient due to insufficient control and monitoring techniques. This limitation can be countered: By using modern communication and evaluation technologies, a "smart" bioreactor can be developed, which evaluates algae growth in real-time, performs process adaptations and thus significantly accelerates algae growth and scale-up. Therefore, an algae bioreactor was established at the University of Technology Sydney. The subject of this paper is the study of algae growth using Long Short-Term Memory Neural Networks (LSTMs). In order to learn the behavior of algae in the shortest possible series of experiments, repetitive change intervals were run by systematically varying the environmental parameters. LSTMs were trained to model algae growth. Attention mechanism is used on variable and temporal direction for importance. The LSTM is compared to a Transformer and an ARIMA. Based on the trained models, the behavior of algae growth is interpreted.

Although industrial robots are predetermined to overtake repetitive and physically demanding assembly tasks of human workforce while delivering constant product quality, their precision comes at high hardware costs. Efficient control strategies make lightweight robotic solutions more attractive to companies with limited resources. This is especially important for assembly tasks requiring high adaptability, sensitivity and precision such as through-hole electronics assembly (THT). In this paper, a cost-efficient system for lightweight robot THT assembly and methods for approach, search and insertion via visual and force control are presented and evaluated. Using this system we can deal with large initial deviations of electronic components in the robot gripper and successfully insert THT components with two pins into circuit board holes under tight sub-millimeter tolerances. The proposed control methods do not require training and the system can be easily reconfigured for different hardware or assembly requirements. We evaluate our system on six different electronic components with two pins such as capacitors and inductors.

Support Vector Machines have been successfully used for one-class classification (OCSVM, SVDD) when trained on clean data, but they work much worse on dirty data: outliers present in the training data tend to become support vectors, and are hence considered "normal". In this article, we improve the effectiveness to detect outliers in dirty training data with a leave-out strategy: by temporarily omitting one candidate at a time, this point can be judged using the remaining data only. We show that this is more effective at scoring the outlierness of points than using the slack term of existing SVM-based approaches. Identified outliers can then be removed from the data, such that outliers hidden by other outliers can be identified, to reduce the problem of masking. Naively, this approach would require training N individual SVMs (and training $O(N^2)$ SVMs when iteratively removing the worst outliers one at a time), which is prohibitively expensive. We will discuss that only support vectors need to be considered in each step and that by reusing SVM parameters and weights, this incremental retraining can be accelerated substantially. By removing candidates in batches, we can further improve the processing time, although it obviously remains more costly than training a single SVM.

Due to IoT and Industry 4.0, more and more data is collected by sensor nodes, which send their data to a central data lake. This approach results in high data traffic and privacy risk, which we want to address in this paper. Therefore we use an existing Learning from Label Proportions (LLP) algorithm, to use the decentralized properties and extend this approach by applying Differential Privacy to the transferred data. This yields to reduced data transfer and increased privacy.