CAD models using a boundary representation (BREP) are widely used across industrial applications. However, extracting and interpreting information from these models typically requires expert knowledge. In this paper, we present a GraphRAG-based system that lowers this barrier by enabling natural language access to detailed geometric and topological information. The approach builds upon an ontological representation of geometries, which is accessible via the semantic query language SPARQL. Large language models are used to translate user questions into formal queries and generate natural language replies, allowing for flexible, domain-aware interaction with the geometric data. The effectiveness of the system is demonstrated through a systematic evaluation using diverse CAD models and a range of natural language queries. Our system empowers domain experts to explore and analyze complex CAD models without requiring expertise in query languages or ontology design. By reducing the complexity of geometric information retrieval, the GraphRAG-based system enables more accessible, automated, and intelligent use of CAD data in industrial applications.

Manufacturing companies are exposed to increasingly complex products and shorter product engineering cycles. Unstructured data hinders the integration of knowledge over the different product engineering stages and complicates structured product development. However, combining an integrated view on relevant data sources following the Advanced Product Quality Planning (APQP) approach provides guidance for product engineers. In this paper, a semantic Knowledge Base (KB), a Process Execution System (PES), and a Computer Vision System (CVS) are introduced, which, in their interaction, compose a Socio-Technical Assistance System (STAS). We combine semantic models of production knowledge, APQP-guided product development, and ontology-based geometric representations of products and manufacturing resources. The PES coordinates the interaction with the user and other system components. The CVS tracks used tools and parts during the assembly and, therefore, enables traceability features and creates confidence in the quality of the assembly. As a result, the developed STAS prototype offers support from customer inquiry through product design and development to manufacturing and assembly, as well as after-sales support. The assistance system enables handling of complex products efficiently in order to reduce required times and costs.

Digital tools for handling the whole product engineering phase are getting more and more important in the context of Industry 4.0 and an increasing product variety. However, especially in small and medium-sized enterprises, a lot of information about product development and production is stored in different documents or isolated data silos. A promising way to arrive at a solution is to model data and knowledge with ontologies and enrich it with context information. This paper presents a concept and a showcase implementation of a company-internal and personalized assistance system for an end-to-end digital product engineering process. We combine a generic and cost-efficient human assistance solution focusing on social aspects and a company-wide knowledge graph to create a seamless and highly integrated data structure that assists many stakeholders in the product engineering process, from product designers to assembly workers. As a result, more complex products can be handled and the product engineering process can be accelerated.

Data driven analysis and optimization of production processes has become a pivotal instrument to use enterprise resources more efficiently and to improve product quality. However, availability and quality requirements still limit the prevalence of big data and learning techniques in industrial applications. Therefore, retrofitting sensors to brownfield systems has been suggested as a solution to acquire relevant real-time process data. In this paper, a low-cost retrofit approach to analyze the operating state of manually operated punching machines based on sound analysis is presented. The machine operating states provide additional information about the metal forming process, required for the enterprise resource planning (ERP) system to optimally schedule orders in prefabrication and plan available resources. As an analysis tool, a transfer learning approach with a convolutional neural network was used to assess data accuracy and prediction results. The input data consists of Mel Spectrogram images acquired by sound sensors retrofitted to the punching machines. The experiments show that the adapted EfficentNet-B0 achieves an accuracy, sensitivity, and precision of approximately 98 % on unseen data in real environment thus demonstrating the applicability of the implemented system.

Crack detection is an important quality control task for metal working processes. In small and medium-size businesses quality inspection is often performed manually. This approach is time consuming and sensitive to errors. In this work an automated visual inspection system for defect detection is presented which classifies cracks in quality control images of a metal forming process taken under varying ambient conditions. The classification is based on transfer learning with AlexNet and GoogLeNet. Supervised learning based on a stochastic gradient descent (sgd) solver was used to train the convolutional neural network. Oversampling and data augmentation with rotated, scaled and shifted images were applied to overcome limitations of the available dataset like huge imbalances of training data sizes for different classes and to prevent overfitting, respectively. Different training parameters were compared and tested to obtain the best classification results for the specific task. GoogLeNet outperformed AlexNet and reached an accuracy of 0.998 and an F1-score of 0.836 for crack detection. Data pre-processing and labeling as well as tuning of the training parameters had a significant influence on classification accuracy and require human decision making.