Causal discovery—the systematic identification of cause-and-effect relationships among variables—forms the cornerstone of causal inference. Its application enables reliable predictions and targeted interventions across complex systems, from medical treatments to engineering processes. Traditional statistical causal discovery methods face significant limitations with high-dimensional data structures, while existing knowledge-based approaches rely on single large-scale models that raise fundamental concerns about computational efficiency and result reliability. The Agentic Stream of Thought (ASoT) addresses these limitations through a novel architecture that orchestrates multiple smaller open-source language models. The framework integrates hierarchical query decomposition with Model Compiler refinement, while dual-stream thought processing enables balanced analysis through parallel evaluation of competing hypotheses. Dedicated Direction and Transitive Processors enhance reasoning by resolving bidirectional relationships and refining transitive pathways. A two-tiered quality gate system and complementary consensus mechanisms—Delphi protocol and Ensemble Synthesis Method—iteratively refine outputs while mitigating hallucination risks. Empirical evaluations across causal discovery benchmarks and question-answering tasks demonstrate that this approach matches or exceeds state-of-the-art models while enabling local deployment, establishing that sophisticated orchestration of smaller models provides a more sustainable path than increasing model scale alone.

In order to optimize the timing of maintenance operations, this work aims at laying the foundation for applying predictive maintenance strategies for the In-Circuit-Test (ICT) by proposing a data analysis pipeline for capturing wear impact on process data. The ICT is an established inspection process within the electronic module production. Subsequently to the assembly of components on a printed circuit board (PCB) mainly using the surface mount technology (SMT) and through hole technology (THT), parts are tested for their conformity of electrical characteristics to predefined tolerances, as well as logical functionalities. Thereby, the ICT is conducted by physically connecting an adapter containing a bed of nails to the circuits of the assembled part and subsequently measuring component dependent characteristics such as the electrical resistance, impedance, inductivity and capacity. Due to the mechanical and electrical impact on the probes and the high-volume nature of the process, these temporarily connected components are subject to wear. Conventional maintenance approaches follow a scheduled preventive strategy by conducting maintenance operations following a predefined number of cycles or strokes. Due to the heuristic nature of the scheduling, the wear margin is not fully exploited and therefore additional costs for prematurely replacing the probes or the complete bed of nails occur. The findings for a real-world dataset from high-volume electronics production show temporal shifts in data distributions between maintenance operations, suggesting adapter-wide needle degeneration. Considering PCB and component batch effects is proposed for successive research methodologies in order to improve the accuracy of the approach.

The predicted drop in prices for automotive sensors and their increasing demand are putting pressure on sensor suppliers. One possible solution is to reduce production costs by expanding automation. Nevertheless, visual quality control in particular is a process step that is often performed by human inspectors, even in the age of Industry 4.0. Software solutions can currently not be used for all types of sensor assembly quality control. This is mainly due to the difficulty of detecting both structural and logical errors and evaluating their severity. We present a machine learning-based software framework that is able to mimic the methodical behavior of a human in error detection and assessment. The framework is based on three types of models, an object recognition model, an anomaly detection model, and a segmentation model. All models are based on convolutional neural networks. An initial proof of concept (PoC) has been performed to prove the usefulness of the models and shows promising results. The initial anomaly detection model is able to reduce the number of objects to be manually tested by 16%. The object detection and segmentation are still in progress and could not be evaluated yet. In addition, a dataset preparation method is presented to use data from industrial practice and relabel it with information from an inspector survey.

The usage of social robots in psychotherapy has gained interest in various applications. In the context of therapy for children with socio-emotional impairments, for example autism spectrum conditions, the first approaches have already been successfully evaluated in research. In this context, the robot can be seen as a tool for therapists to foster interaction with the children. To ensure a successful integration of social robots into therapy sessions, an intuitive and comprehensive interface for the therapist is needed to guarantee save and appropriate human-robot interaction. This publication addresses the development of a graphical user interface for robot-assisted therapy to train socio-emotional skills in children on the autism spectrum. The software follows a generic and modular approach. Furthermore, a robotic middleware is used to control the robot and the user interface is based on a local web application. During therapy sessions, the therapist interface is used to control the robot’s reactions and provides additional information from emotion and arousal recognition software. The approach is implemented with the humanoid robot Pepper (Softbank Robotics). A pilot study is carried out with four experts from a child and youth psychiatry to evaluate the feasibility and user experience of the therapist interface. In sum, the user experience and usefulness can be rated positively.

Due to ever-increasing data availability, computational power, and algorithmic advances, machine learning enables various novel industrial applications. A field with particularly strong potential for machine learning is computer vision. As part of that area, semantic segmentation refers to a process that links each pixel of an image to a corresponding class. In this task, deep learning has outperformed traditional image processing techniques as well as other classical machine learning techniques and therefore has become the new standard approach. Convolutional neural networks (CNN) are a specific form of deep neural networks, which exploit spatial information for the classification of pixels. This paper presents an approach in which CNNs are optimized and applied to a semantic segmentation problem by using multichannel microscopic images of tungsten flat emitters to determine their surface conditions. These are of strong interest since they may reveal information about potential defects and the life expectancy of the emitter, which is an indispensable component of X-ray tubes as they are used in medical applications.

Due to increasing complexity on the shop floor associated with the digital transformation and growing penetration of modern production environments with smart services, the operational phase of automated production facilities is gaining additional importance. The growing range of hybrid service bundles, i.e., the combination of digital services and cyber-physical production systems (CPPS), is expanding the scope of utilization and requires the application of platform-based solutions for service integration. Efficient, scalable IT infrastructures as well as technological flexibility and independence are premises for the successful implementation of these platforms. The availability of ubiquitous communication for seamless networking of all entities also plays a central role. These prerequisites enable the vertical integration of automation components on the shop floor up to the context of strategic business processes based on continuous data and information models. This paper presents an approach for integrating CPPS elements into scalable, highly available service platforms and shows how modern control approaches can be successfully linked with smart service concepts. The exemplary use case for this is a test bench for electric drives.

Connected factories offer more and more possibilities to bring business logic in the industrial related components like industrial control systems (ICS). These systems in the operational technology (OT) sector are usually harder to update and maintain compared to IT systems. In recent years, the number of cyberattacks that are specifically tailored to OT systems has increased. We analyzed BlackEnergy 3 (BE3), Crashoverride (CO), and Trisis (TS). After describing the occurrences of these attacks, we looked for similar strategies between these three approaches and propose promising methods to prevent such or similar attacks in the future.

The discussion about the term "Digital Twin" and its associated concepts has expanded remarkably in the scientific community in recent years. In the context of industrial production, the Digital Twin refers to a holistic, linked, virtual representation of a physical entity. The spectrum of these entities ranges from individual products to specific manufacturing processes up to complex automated production systems. Due to this diversity, many authors offer different interpretations of the topic, resulting in a wide range of use case-specific models. However, a common meta-model that integrates and classifies the different aspects is still to be presented. Starting with a brief overview of developments in recent years, this paper introduces a novel Digital Twin Structure Model in the context of manufacturing. After providing an overview and explanation of the dimensions, functionality and coupling of the Digital Twin in relation to its surroundings, specific Digital Twin realizations are characterized and compared using the proposed structure model.

The capacitor discharge welding (CDW) is a resistance welding process that excels through brief process times, low thermal stress, and good automation potential. Nevertheless, potential industrial users hesitate to use the CDW process, owing mainly to the unavailability of automated process control to ensure cost-efficient production and high product quality. For quality assurance, representative values or process curves of the most critical process parameters such as capacitor energy, contact force, electric current, and sink-in depth are monitored. The lack of full significance of this monitoring is indicated by the regular destructive testing of random samples through load tests and cross-section analysis. The essential aspect of the process quality is heat development in the welding zone. The current distribution, as one crucial parameter influencing heat development, is missing. Therefore, this paper presents an in-situ qualitative, indirect current distribution measurement system. The new measurement supplements the process monitoring and analysis, and thus, gain new process knowledge. With better process monitoring and thus understanding, new application areas can be developed for CDW.

In research, there are more and more successful approaches to operate production plants more resource-efficiently and more productively with the help of reinforcement learning. An important point is the reduction of the energy demand of production plants. Instead of manually implementing complex standby strategies rigidly in a PLC, an intelligent system can train to derive decisions about the optimal energetic state of each component autonomously. Since learning in a virtual environment has decisive advantages, simulation models with sufficient accuracy are necessary. Many of the previous implementations of reinforcement learning approaches in an industrial environment are usually tailored to a specific plant. In particular, the agent's scope of action and its connection to the environment must be adapted manually for a new plant. The presented solution shows an intelligent system which automatically optimizes the virtual learning environment with the support of plant knowledge and adapts the reinforcement learning agent to the respective production plant.

The scientific discussion on the Digital Twin gained substantial momentum within the last couple of years. In the context of Industry 4.0, the Digital Twin in the broadest sense refers to the concept of a coupled virtual representation of a physical asset. The asset types may range from products and processes to whole production systems. Different authors offer a variety of interpretations on the actual nature of the Digital Twin, often defining it via implicit functional aspects rather than giving explicit definitions. At the same time, some authors use the term merely as a catchphrase, thus applying blur to this emerging paradigm. At this point, a comprehensive understanding as well as a unifying model of the Digital Twin are absent. This paper therefore provides a structured review of the body of relevant literature with the specific focus on finding explicit definitions of the term "Digital Twin" as well as stated models characterizing the Digital Twin concept. As a result, a relevance assessment provides the most important literature regarding the research questions. The found explicit definitions, being partly conflicting, are presented. Similar notions like Product Avatar and Digital Shadow are assorted accordingly. Further, depictions of Digital Twin models stated in the most relevant literature as well as derived Digital Twin purposes are provided. Thus, this paper extends the theoretical foundation, thereby setting a basis for future augmented Digital Twin modeling.

In the context of Industry 4.0, the flexibility of production systems rises constantly. To fulfil this demand, wireless instead of classic wired communication networks can be used. This trend is supported by the advancement of new network standards such as 5G. Current design of wireless networks in production systems in the early planning phase relies on best practices and experience. In following planning phases worst case calculations are used to verify the network topology. Taking the dynamic behaviour of flexible production systems into account, networks are constantly changing. In this paper, existing discrete event simulation of production systems is augmented with planning components of wireless networks to allow a more precise planning of production systems using wireless communication.

Establishing adequate cybersecurity for their operational technology (OT) is an existential challenge for manufacturing enterprises. Domain-specific security standards should provide essential support in this challenge. However, they cannot be implemented equally for enterprises of all sizes.We investigate to what extent domain-specific security standards for operational technology are applicable by small and medium-sized as well as large manufacturing enterprises, and how their individual need for action can be identified and addressed. We support our investigation with the results of two independent surveys among manufacturers about their needs for cybersecurity support.In the course of this investigation, we learned that most domain-specific security standards are well applicable to large enterprises. In contrast, small and medium-sized enterprises (SME) seek the support of security experts, who, for their part, are often struggling with a lack of experience in operational technology. To facilitate this cooperation, we provide an introduction for OT-and cybersecurity-experts to the respective basic concepts of their collaborators.

We created an approach for a smart living platform called ForeSight which consists of different modules: a service engineering module, a Web of Things (WoT)-based Internet of Things (IoT) module and an artificial intelligence (AI) component. This paper describes how openHAB, a smart home middleware, is extended to fulfill platform requirements related to a successful interaction with the IoT module of ForeSight, more precisely, to add identity and access management (IAM) to openHAB and comply with European privacy laws.

Global trends such as mass customization and lot size one, demand flexibility, autonomy and adaptability in production. Interoperability of devices is a key challenge as production systems evolve into Cyber-Physical Production Systems (CPPS). As part of the German strategic project Industrie 4.0 (I4.0), concepts and solutions for continuous digitization of production are being developed, in order to meet these numerous challenges. The concept of the Asset Administration Shell (AAS) was introduced in order to provide data and information in a standardized and semantically described manner, thus enabling interoperability and easy interaction. In this paper, we show how users can translate a semantic description of plants, machines or individual components that are based on a standardized Open Platform Communication Unified Architecture (OPC UA) information model, into the AAS information model.

Tool wear is one of the main factors of manufacturing costs in subtractive manufacturing processes. To control manufacturing processes while taking the tool wear into account, a variety of tool condition monitoring systems have been investigated. In this paper, we present a new approach to support the manual analysis of tool wear images by the means of semantic image segmentation. We utilize deep learning for image evaluation through semantic classification of different defect regions. In this study, a small-sized dataset of 100 cutting tool inserts at different tool conditions, exhibiting various wear defects, is acquired and masked by a process expert. A sliding window approach is used to extract small size feature maps from the raw images, with the class of the center pixel as the label. The relationship between the features and the label is trained using a convolutional neural network. Our investigation shows that this network can predict the wear defect class of each pixel with an accuracy of over 91%. Compared to other approaches, the proposed solution can differentiate between various defect types, for instance, flank wear, groove formation and build-up-edge. From the resulting segmented image, different wear metrics are computed, such as the maximum flank wear width or the occurrence and size of other wear defects. This information is fed back to the machine operator to support the decision process of whether to continue machining, adapt the cutting conditions or exchange the insert.

In future, smart home and smart living applications will enrich daily life. These applications are aware of their context, use artificial intelligence (AI) and are therefore able to recognize common use cases reliably and adapt these use cases individually with the current user in mind. This paper describes a concept for such an AI-based platform. The presented platform approach considers different stakeholders, e.g. the housing industry, service providers and tenants.

Automated detection of cancer metastases in lymph nodes has the potential to improve the assessment of prognosis for patients. To enable fair comparison between the algorithms for this purpose, we set up the CAMELYON17 challenge in conjunction with the IEEE International Symposium on Biomedical Imaging 2017 Conference in Melbourne. Over 300 participants registered on the challenge website, of which 23 teams submitted a total of 37 algorithms before the initial deadline. Participants were provided with 899 whole-slide images (WSIs) for developing their algorithms. The developed algorithms were evaluated based on the test set encompassing 100 patients and 500 WSIs. The evaluation metric used was a quadratic weighted Cohen's kappa. We discuss the algorithmic details of the 10 best pre-conference and two post-conference submissions. All these participants used convolutional neural networks in combination with pre- and postprocessing steps. Algorithms differed mostly in neural network architecture, training strategy, and pre- and postprocessing methodology. Overall, the kappa metric ranged from 0.89 to -0.13 across all submissions. The best results were obtained with pre-trained architectures such as ResNet. Confusion matrix analysis revealed that all participants struggled with reliably identifying isolated tumor cells, the smallest type of metastasis, with detection rates below 40%. Qualitative inspection of the results of the top participants showed categories of false positives, such as nerves or contamination, which could be targeted for further optimization. Last, we show that simple combinations of the top algorithms result in higher kappa metric values than any algorithm individually, with 0.93 for the best combination.

Over the next few years, more and more autonomous mobile robot systems will find their way into modern shop floors. However, it will be necessary to provide human-machine interfaces for interventions in unexpected situations like system-deadlocks, algorithm failures or inabilities. Using virtual or mixed reality-technologies, multi-modal teleoperation offers potential for being a suitable human-machine interface. Essential challenges in this field are, among others, a real-time remote control, a time-efficient and holistic environment detection using multiple sensors, a noise-reduced visualization of sensor-data, and capabilities of object recognition. This paper summarizes research results regarding an architecture capable of a near realtime, interoperable, and operator-supporting teleoperation. The focus of this paper is on a method to efficiently process and visualize point-clouds to meet high frame rate demands of virtual reality applications. To provide near real-time feedback of the robot and its environment over large distances, the presented method is capable to segment known objects from unknown objects to reduce bandwidth requirements. The results of this paper were evaluated using a industrial articulated robotic arm for teleoperation via a long distance UDP/IP communication.

The cooperation of several industrial robots can improve processes or even makes new processes possible. In the scope of this work, the effect of the cooperation on the reachable workspace is investigated. Thus, the approach of reachability maps is extended to cooperative processing by simultaneously moving the tool and the workpiece. This allows a general statement about the improvement of the workpiece related reachability due to the simultaneous movement of tool and workpiece. A dual arm robot setup is used for implementation and the cooperative movement is compared to single arm movement by generating reachability maps for each case. In doing so, it is also presented how the integration of a collision detection improves the representativeness of the reachability maps.

Documenting unwritten languages is a challenging task, even for trained specialists. To help linguists in better and faster documenting new languages is the goal of the French-German ANR-DFG project BULB. To discover the phonetic inventory of a language the project follows three steps: estimating phoneme boundaries, classifying articulatory features (AFs) for each individual segment and clustering the segments into a phoneme inventory. In this work, we focus on estimating the phoneme boundaries and the extraction of AFs, but also perform a first simple clustering based on the recognized AFs. We demonstrate that our Deep Bidirectional LSTM-based approach for identifying phoneme boundaries achieves state-of-the-art performance and evaluate AF extraction based on feed forward neural networks.

The work of research institutes in the field o! engineering is characterized by a work-sharing fulfillment o! knowledge-intensive tasks. For the success of these institutes, it is of great importance that this knowledge is properly networked. In order to improve this and to develop the competences of institutes a competence management system is created that comprehends all competences available, makes them plausible, facilitates competence retrieval and supports a competence exchange. In this paper we propose an approach for capturing, structuring and visualizing competences. This approach can be used for both research facilities and companies in general. Moreover, we show how the competence data is integrated into an ontology-based competence management system. This also includes the connection between the ontology in the backend and the graphical user interface, being created in SharePoint.

This paper describes how Functional Mock-up Units (FMUs) can be used multifunctional in production engineering applications. There are presented three different examples: Example one shows the behavior simulation of mechatronic components in Virtual Engineering (VE). The second application example describes the energy simulation of an automated production system while Virtual Commissioning (VC). The last presented use case of FMU implementation is an application after engineering phase, while real production runtime (digital twin). Additional to the application examples the methodical and historical background of the integration of FMUs is presented.

Industrial production systems meet strict requirements regarding availability, process control and condition monitoring. As a key enabler of Industry 4.0, cyber-physical systems form the core of a modular, web-based framework, which delivers more efficient condition monitoring mechanisms for maintenance staff in production facilities. The present approach illustrates the potential of a decentralized and centralized, distributed system by using standardized communication protocols and semantic information models. The use of web-based platforms like node.js and protocols such as OPC UA offers the ability to automate transferring information about anomalies, root causes and nominal data directly between cloud-based services and condition monitoring systems on the shop floor. To conclude, a validation is accomplished within a case study on a flexible handling unit.

The increasing complexity of production plants in the context of Industry 4.0 poses new challenges with regards to technical maintenance. In industrial domains, Condition Monitoring Systems (CMS) enable fault-tolerant, predictable production systems. Unfortunately, CMS development is challenging due to the distribution of computational tasks among heterogeneous industrial Internet of Things and Services (IoTS)-architectures. It is usually started from scratch, which is time-consuming and error-prone. In times of IoTS the employee as a maintenance worker, who ensures the availability of production machinery, is still important. In order to address these challenges, we propose a modular framework for web-based condition monitoring and fault diagnostics. The implementation relies on technologies such as Hypertext Markup Language 5 (HTML5), JavaScript (JS) and node.j s as well as industrial internet communication standards. Finally, verification and validation are performed in a case study on a modular robot cell for the inspection of electronic assemblies.

This paper describes a method for extrinsic sensor calibration for a 2D laser profile sensor on a robot arm used as a robot based 3D scanning system. In order to establish a relationship between the sensor measurements and the robot positions, the transformation between the robot flange and the sensor reference frame are determined. The developed calibration method is implemented as an automated self-calibration. It is based on the use of a pin as a fixed reference. The robot is aligned to the tip of the pin automatically with certain orientations. The sensor transformation is calculated based on six recorded robot positions. Finally, the accuracy of the calibration is determined with an additional robot position.

This paper describes a color extension of a 2D laser profile sensor by extracting the corresponding color from a camera image. For these purpose, we developed a routine for an extrinsic calibration between the profile sensor and the camera. Based on the resulting translation and rotation vectors a belonging pixel can be calculated for each profile point. Consequently, the color for each profile point can be extracted from the image. This approach is used to extend the geometric data of a robotic based 3D scanning system by color data.

Regionalization of software and hardware engineering in international project business is a key driver for the success of multinational companies. These companies usually act as prime contractors and therefore have to execute large engineering projects across several countries. This requires a shift of the whole or at least a portion of their engineering value chain to a foreign country (regionalization). Since a shift of the value chain means a major change in the project execution process, a regionalization initiative is of high risk for the company. For that reason, we describe an approach to effectively check the extent to which a given company is ready for regionalization. The goal of this readiness check is to identify areas with need for action and preparatory work before starting a regionalization initiative. The check is structured alongside the fundamental questions as to why regionalization should be performed, what should be regionalized, who should do it, where regionalization should take place and which conditions apply. It provides management with a cockpit-like overview of the current state and addresses required action fields. By measuring the extent of regionalization readiness, important insight into the regionalization effort can be gained, such as the need for investment into the early phases and the required balance between all participating parties (stakeholders). Examples of practical use of the readiness check are presented and directions for further research are shown.

The efficiency of electric drives is determined by the mechanical and magnetic properties of the functional elements, such as stator and rotor. Due to the electromagnetic interaction with the magnetic field, generated by the stator, the rotor transforms the magnetic energy into a rotational motion. It is characterized by its magnetization, its magnetic field strength, the field distribution and its geometric dimensions. The current challenge is to produce electrical drives with minimal loss. This requires ensuring the magnetic characteristics of the rotors. Errors that can occur during manufacture are cracks, blowholes or inclusions in the permanent magnet. Those can lead to commutation problems or to increased noise. For this purpose, a non-destructive and non-contact measurement method, based on the measurement of the magnetic field with Hall-effect sensors to detect defects in plastic-bonded permanent magnet rotors was developed.

Due to increasing energy costs and the demand to lower CO2emissions in the production, companies try to integrate a continuous energy efficiency improvement process. By implementing the energy value stream analysis, an optimized improvement process can be established which leads to a more efficient production. This approach shows a practically optimized method. It enables a holistic analysis of the energy productivity and helps to identify and to realize sustainable solutions. Furthermore this structured improvement process leads to a higher transparency of energy consumers.

The use of modern production systems like the Toyota production system led to significant productivity gains in several industries. The methods and tools used in these production systems have mainly been applied to the mass production of goods. This raises the question if any of these methods and tools can also be applied to the engineering of complex systems. For that reason, this paper focuses on a set of criteria comparing a modern production system with the engineering of automation systems or other complex industrial systems. Based on these insights, this paper shows how the concept of an Engineering Community provides basic tool support to realize ideas from modern production systems in the field of engineering projects. This will be illustrated by several examples in order to show how these ideas can be implemented in practice.