The digital factory provides undoubtedly great potential for future production systems in terms of efficiency and effectivity. A key aspect on the way to realize the digital copy of a real factory is the understanding of complex indoor environments on the basis of three-dimensional (3D) data. In order to generate an accurate factory model including the major components, i.e., building parts, product assets, and process details, the 3D data that are collected during digitalization can be processed with advanced methods of deep learning. For instance, the semantic segmentation of a point cloud enables the identification of relevant objects within the environment. In this work, we propose a fully Bayesian and an approximate Bayesian neural network for point cloud segmentation. Both of the networks are used within a workflow in order to generate an environment model on the basis of raw point clouds. The Bayesian and approximate Bayesian networks allow us to analyse how different ways of estimating uncertainty in these networks improve segmentation results on raw point clouds. We achieve superior model performance for both, the Bayesian and the approximate Bayesian model compared to the frequentist one. This performance difference becomes even more striking when incorporating the networks’ uncertainty in their predictions. For evaluation, we use the scientific data set S3DIS as well as a data set, which was collected by the authors at a German automotive production plant. The methods proposed in this work lead to more accurate segmentation results and the incorporation of uncertainty information also makes this approach especially applicable to safety critical applications aside from our factory planning use case.

The 3D modelling of indoor environments and the generation of process simulations play an important role in factory and assembly planning. In brownfield planning cases, existing data are often outdated and incomplete especially for older plants, which were mostly planned in 2D. Thus, current environment models cannot be generated directly on the basis of existing data and a holistic approach on how to build such a factory model in a highly automated fashion is mostly non-existent. Major steps in generating an environment model of a production plant include data collection, data pre-processing and object identification as well as pose estimation. In this work, we elaborate on a methodical modelling approach, which starts with the digitalization of large-scale indoor environments and ends with the generation of a static environment or simulation model. The object identification step is realized using a Bayesian neural network capable of point cloud segmentation. We elaborate on the impact of the uncertainty information estimated by a Bayesian segmentation framework on the accuracy of the generated environment model. The steps of data collection and point cloud segmentation as well as the resulting model accuracy are evaluated on a real-world data set collected at the assembly line of a large-scale automotive production plant. The Bayesian segmentation network clearly surpasses the performance of the frequentist baseline and allows us to considerably increase the accuracy of the model placement in a simulation scene.

One major challenge towards a fully digital factory is the understanding of complex and dynamic indoor scenes. This knowledge is needed to determine the as-is state in production plants and to set up factory simulations. While in recent years laser scanning and photogrammetry techniques have facilitated the digitalization of factory environments, the actual simulation model generation is to the greatest possible extent still manual work. In this paper we propose a cross-industry simulation model generation framework that takes in raw point clouds from laser scanners and photogrammetry and outputs a simulation model. We present a deep learning architecture based on PointNet [1], which is capable of semantic scene segmentation, and integrate a CAD model placement routine based on unsupervised learning and point cloud registration. We prove the feasibility of our framework by collecting a scan and photogrammetry dataset at a German automotive assembly plant and applying our framework.

From cutting shapes out of metal or wood to welding, screwing or lifting heavy objects, the application and versatility of robots in industry is vast. Their programs however are typically very static and only used in pre-defined, static settings, thus they cannot deal with changing production tasks. Whenever a special problem has to be solved several times with different parametrizations, transferring simulation results to real-world applications is desirable in order to reduce the time investment necessary for program adjustments. The current paper discusses the major challenges in including powerful simulations as a tool to improve robot programs at runtime. Further, an evaluation of the Virtual Robot Experimentation Platform (V-REP) with respect to its applicability to direct robot control via simulation is presented.

The road transport sector is currently facing significant challenges, due in part to CO₂ emissions, high fuel prices, and a shortage of staff. These issues are partially caused by more than 40% of truck journeys being “empty runs” in some member states of the European Union and heavy under-utilization of deck space for non-empty runs. In order to overcome said inefficiency, this paper proposes a decentralized platform to facilitate collaborative transport networks (CTNs), i.e., to enable horizontal collaboration to increase load factors and reduce costs and CO₂ emissions. Our solution leverages secure multi-party computation (MPC) to guarantee that no sensitive business information is leaked to competing hauliers. The system optimizes truck assignments by modeling logistics as a weighted graph that considers orders and truck capacities while maintaining strict confidentiality. Our approach addresses key barriers to CTN adoption, such as lack of trust and data privacy. Implemented using MPyC without extensive optimizations, we demonstrate the efficiency and effectiveness in increasing the average load factor, while achieving acceptable running times (in the order of hours) for arguably meaningful instance sizes (up to 1000 orders). After leveraging a rather simplistic modeling inspired by previous work, we finally give an outlook of possible extensions toward more realistic models and estimate their impact on efficiency.