Efficient Modeling of Digital Shadows for Production Processes: A Case Study for Quality Prediction in High Pressure Die Casting Processes

Chakrabarti, Arnab; Sukumar, Ravi Prasanna; Jarke, Matthias; Rudack, Maximilian; Buske, Paul; Holly, Carlo

doi:10.1109/dsaa53316.2021.9564113

“…Yet, the extensive breadth (number of values) and depth (frequency and precision) require domain knowledge of the process to select required data to facilitate product quality and productivity improvements. In an experimental setup, Rudack et al (2022) and Chakrabarti et al (2021) accessed multiple data sources based on the Open Platform Communication Unified Architecture (OPC-UA) and transmitted them to a streaming pipeline defined in the low-code programming environment Node-RED and executed by Apache Kafka. The data are subsequently stored in our data lake, utilizing a MinIO object store as the underlying storage system.…”

Section: Toward An Empirically Grounded Iop Infrastructurementioning

confidence: 99%

“…The data are subsequently stored in our data lake, utilizing a MinIO object store as the underlying storage system. A systematic hierarchy of high-dimensional DSs for data analytics is combined with an AI-based recommender system for interactive visual analytics, e.g., in product quality assurance (Chakrabarti et al, 2021).…”

Section: Toward An Empirically Grounded Iop Infrastructurementioning

confidence: 99%

Digital Shadows: Infrastructuring the Internet of Production

Aalst

¹

,

Jarke

²

,

Koren

³

et al. 2023

Interdisciplinary Excellence Accelerator Series

0

View full text Add to dashboard Cite

Digitization in the field of production is fragmented in very different domains, ranging from materials to production technology to process and business models. Each domain comes with specialized knowledge, often incorporated into mathematical models. This heterogeneity makes it hard to naively exploit advances in data-driven machine learning that could facilitate situation adaptation and experience transfer. Innovative combinations of model-driven and data-driven solutions must be invented but also made comparable and interoperable to avoid ending up in information silos. In future World Wide Labs (WWLs), experiences can be shared, aggregated, and used for innovation. WWLs will be complex, evolving socio-technical networks of interconnected devices, software, data stores, and humans as users and contributors of expert knowledge and feedback. Integrating a large number of research labs, engineering, and production sites requires a capable cross-domain Internet of Production (IoP) infrastructure. The IoP project claims Digital Shadows (DSs) to offer a shared conceptual foundation for infrastructuring the IoP. In engineering, DSs were introduced as the data provision link to Digital Twins, whereas in computer science, DSs generalize the well-established concept of database views. In this chapter, we elaborate on the roles of DSs in infrastructuring the IoP from three perspectives: analytic functionality, conceptual organization, and technical networking. As an example where an integrative DS-like approach is already highly successful, we showcase the approach and infrastructure of the process mining field.

show abstract

Improving Manufacturing Efficiency for Discontinuous Processes by Methodological Cross-Domain Knowledge Transfer

Lockner

¹

,

Buske

²

,

Rudack

³

et al. 2023

Interdisciplinary Excellence Accelerator Series

1

0

View full text Add to dashboard Cite

Discontinuous processes face common tasks when implementing modeling and optimization techniques for process optimization. While domain data may be unequal, knowledge about approaches for each step toward the solution, e.g., data gathering, model reduction, and model optimization, may be useful across different processes. A joint development of methodologies for machine learning methods, among other things, ultimately supports fast advances in cross-domain production technologies. In this work, an overview of common maturation stages of data-intensive modeling approaches for production efficiency enhancement is given. The stages are analyzed and communal challenges are elaborated. The used approaches include both physically motivated surrogate modeling as well as the advanced use of machine learning technologies. Apt research is depicted for each stage based on demonstrator work for diverse production technologies, among them high-pressure die casting, surface engineering, plastics injection molding, open-die forging, and automated tape placement. Finally, a holistic and general framework is illustrated covering the main concepts regarding the transfer of mature models into production environments on the example of laser technologies.Increasing customer requirements regarding process stability, transparency and product quality as well as desired high production efficiency in diverse manufacturing processes pose high demands on production technologies. The further development of digital support systems for manufacturing technologies can contribute to meet these demands in various production settings. Especially for discontinuous production, such as injection molding and laser cutting, the joint research for different technologies helps to identify common challenges, ranging from problem identification to knowledge perpetuation after successfully installing digital tools. Workstream CRD-B2.II “Discontinuous Production” confronts this research task by use case-based joint development of transferable methods. Based on the joint definition of a standard pipeline to solve problems with digital support, various stages of this pipeline, such as data generation and collection, model training, optimization, and the development and deployment of assistance systems are actively being researched. Regarding data generation, e.g., for the high-pressure die-casting process, data acquisition and extraction approaches for machines and production lines using OPC UA are investigated to get detailed process insights. For diverse discontinuous processes and use cases, relevant production data is not directly available in sufficient quality and needs to be preprocessed. For vision systems, ptychographic methods may improve recorded data by enhancing the picture sharpness to enable the usage of inline or low-cost equipment to detect small defects. Further down the pipeline, several research activities concern the domain-specific model training and optimization tasks. Within the realm of surface technologies, machine learning is applied to predict process behavior, e.g., by predicting the particle properties in plasma spraying process or plasma intensities in the physical vapor deposition process. The injection molding process can also be modeled by data-based approaches. The modeling efficiency based on the used amount of data can furthermore be effectively reduced by using transfer learning to transfer knowledge stored in artificial neural networks from one process to the next. Successful modeling approaches can then be transferred prototypically into production. On the examples of vision-based defect classification in the tape-laying process and a process optimization assistance system in open-die forging, the realization of prototypical support systems is demonstrated. Once mature, research results and consequent digital services must be made available for integrated usage in specific production settings using relevant architecture. By the example of a microservice-based infrastructure for laser technology, a suitable and flexible implementation of a service framework is realized. The connectivity to production assets is guaranteed by state-of-the-art communication protocols. This chapter illustrates the state of research for use-case-driven development of joint approaches.

show abstract

Improving Manufacturing Efficiency for Discontinuous Processes by Methodological Cross-Domain Knowledge Transfer

Lockner,

Buske,

Rudack

et al. 2023

Interdisciplinary Excellence Accelerator Series

0

View full text Add to dashboard Cite

Discontinuous processes face common tasks when implementing modeling and optimization techniques for process optimization. While domain data may be unequal, knowledge about approaches for each step toward the solution, e.g., data gathering, model reduction, and model optimization, may be useful across different processes. A joint development of methodologies for machine learning methods, among other things, ultimately supports fast advances in cross-domain production technologies. In this work, an overview of common maturation stages of data-intensive modeling approaches for production efficiency enhancement is given. The stages are analyzed and communal challenges are elaborated. The used approaches include both physically motivated surrogate modeling as well as the advanced use of machine learning technologies. Apt research is depicted for each stage based on demonstrator work for diverse production technologies, among them high-pressure die casting, surface engineering, plastics injection molding, open-die forging, and automated tape placement. Finally, a holistic and general framework is illustrated covering the main concepts regarding the transfer of mature models into production environments on the example of laser technologies.Increasing customer requirements regarding process stability, transparency and product quality as well as desired high production efficiency in diverse manufacturing processes pose high demands on production technologies. The further development of digital support systems for manufacturing technologies can contribute to meet these demands in various production settings. Especially for discontinuous production, such as injection molding and laser cutting, the joint research for different technologies helps to identify common challenges, ranging from problem identification to knowledge perpetuation after successfully installing digital tools. Workstream CRD-B2.II “Discontinuous Production” confronts this research task by use case-based joint development of transferable methods. Based on the joint definition of a standard pipeline to solve problems with digital support, various stages of this pipeline, such as data generation and collection, model training, optimization, and the development and deployment of assistance systems are actively being researched. Regarding data generation, e.g., for the high-pressure die-casting process, data acquisition and extraction approaches for machines and production lines using OPC UA are investigated to get detailed process insights. For diverse discontinuous processes and use cases, relevant production data is not directly available in sufficient quality and needs to be preprocessed. For vision systems, ptychographic methods may improve recorded data by enhancing the picture sharpness to enable the usage of inline or low-cost equipment to detect small defects. Further down the pipeline, several research activities concern the domain-specific model training and optimization tasks. Within the realm of surface technologies, machine learning is applied to predict process behavior, e.g., by predicting the particle properties in plasma spraying process or plasma intensities in the physical vapor deposition process. The injection molding process can also be modeled by data-based approaches. The modeling efficiency based on the used amount of data can furthermore be effectively reduced by using transfer learning to transfer knowledge stored in artificial neural networks from one process to the next. Successful modeling approaches can then be transferred prototypically into production. On the examples of vision-based defect classification in the tape-laying process and a process optimization assistance system in open-die forging, the realization of prototypical support systems is demonstrated. Once mature, research results and consequent digital services must be made available for integrated usage in specific production settings using relevant architecture. By the example of a microservice-based infrastructure for laser technology, a suitable and flexible implementation of a service framework is realized. The connectivity to production assets is guaranteed by state-of-the-art communication protocols. This chapter illustrates the state of research for use-case-driven development of joint approaches.

show abstract

Efficient Modeling of Digital Shadows for Production Processes: A Case Study for Quality Prediction in High Pressure Die Casting Processes

Cited by 7 publications

References 11 publications

Digital Shadows: Infrastructuring the Internet of Production

Digital Shadows: Infrastructuring the Internet of Production

Improving Manufacturing Efficiency for Discontinuous Processes by Methodological Cross-Domain Knowledge Transfer

Improving Manufacturing Efficiency for Discontinuous Processes by Methodological Cross-Domain Knowledge Transfer

Contact Info

Product

Resources

About