Skip to content

knowlEdge Newsletter — 2023

Dear Reader,

We are pleased to announce the newsletter for the third year of the EU project knowlEdge. As in previous years, we achieved many project innovations and highlights in order to realize the project’s overall vision.

Looking back on the last year, the knowlEdge project, whose aim is to develop a new generation of AI methods, systems, and data management infrastructure, has successfully continued with 12 partners from across 7 EU countries.

Among the many highlights, the knowlEdge platform architecture has been utilized among the different consortium partners. The conceptual architecture of the knowledge platform epitomizes the edge-to-cloud continuum of the project and comprises five different layers ensuring trustworthiness, security, distributed data analytics, AI model sharing and human-machine interaction. To connect the underlying (manufacturing) assets, the data integration and management layer is endowed with the data collection platform which connects shopfloor data to higher-level components. In addition, this layer provides the functionality of coping with historical and live data by means of the historical data storage service and real-time brokering. The knowlEdge discovery engine is used in the AI and data analytics layer to foster exploratory data analysis from heterogeneous sources to prepare data for AI model generation. Generated AI models are made available in the knowlEdge management layer, which provides AI model storage and sharing through the knowlEdge repository and knowlEdge marketplace. The smart decision layer enables access to the AI-analytical functionalities via the human-AI collaboration component and the decision support system. The aforementioned layers are supported by the platform services layer.

One of the innovations that have been advanced during the project is the knowlEdge discovery engine, which provides a rich set of exploratory data analytics functionality. As its core functionality, the knowledge discovery engine allows a user to identify explicit and latent data characteristics by making use of unsupervised data analysis methods.
One example of the aforementioned data characteristics is the occurrence of anomalies in time series data. Unsupervised models of the knowlEdge discovery engine can be trained to learn expected data characteristics and predict whether a given data point is anomalous or not. For this purpose, a user can choose between several available models and configure the task to his or her needs. In addition, a domain expert can label analysis results to enrich the data available for supervised approaches.

In the knowlEdge project, there is a special emphasis on how humans interact with AI models generated within the AI and data analytics layer, either by the knowlEdge discovery engine or the AI model generation component. To this end, we developed and implemented a human-AI collaboration process, which enables domain experts to choose models, set parameters and optimize model parameters on top of the selected data source. As illustrated in the figure, the human-AI collaboration interface empowers domain experts to seamlessly monitor and adjust all pertinent parameters through an intuitive graphical user interface. Once the expert has reached a decision on the preferred model, parameters, and values to proceed with, the UI offers the capability to export the results, enabling their seamless integration into the decision support system for further utilization.

Alongside technical advancements, numerous online and in-person meetings have been conducted. The second physical consortium meeting of the knowlEdge project took place from January 16th to 19th, 2023, in Turin, Italy, hosted by the project partner LINKS Foundation. Comprehensive discussions were held regarding project progress, achievements, and forthcoming plans for the remainder of the year. Nearly 30  participants representing the project partners were able to attend, fostering collaborative dialogue and strategic planning.

Another physical meeting of the knowlEdge project consortium was held from May 3 to May 5, 2023. The event was hosted by the coordinating project partner VTT. The consortium provided a comprehensive status overview of the various components that make up the knowlEdge platform, including steps completed. The workflow, architecture and next steps of development of each component were also put under  discussion. The meeting also focused on the pilot implementations. Checkpoints and planning for the deployment of functionalities were covered, and roadmaps were developed for the coming months up to the finalization of the project

The last technical knowlEdge consortium meeting in 2023 took place in October at Barcelona Supercomputing Center. It concentrated on “dotting the i’s and crossing the t’s”, i.e. ensuring that all project’s objectives will be met. During the first two days all technical Work Packages were reviewed together with the pilots and use cases. The last day finally focused on dissemination and exploitation activities. The group photo was taken in front of the Mare Nostrum 4 supercomputer, which is installed in the deconsecrated Chapel Torre Girona at the Polytechnic University of Catalonia in Barcelona.

Among the many activities, the knowlEdge project has achieved several scientific publications. A particular highlight is the joint journal paper on “AI Lifecycle Zero-Touch Orchestration within the Edge-to-Cloud Continuum for Industry 5.0” which appeared in the special issue on Manufacturing and Service Systems for Industry 4.0/5.0 in the journal Systems and can be found online at https://doi.org/10.3390/systems12020048. As another highlight, we were able to contribute six chapters to the book Artificial Intelligence in Manufacturing – Enabling Intelligent, Flexible and Cost-Effective Production Through AI, which can be found online at https://doi.org/10.1007/978-3-031-46452-2. Our topics include the design of a marketplace to exchange AI models for industry 5.0, the human-AI interaction for semantic knowledge enrichment of AI model output, a manufacturing digital twin framework, advancing networked production through decentralized technical intelligence, an AI model generation framework for boosting AutoML and XAI in manufacturing as well as our approach to anomaly detection in manufacturing. More information can be found online on the project website. We are convinced that our open access publication efforts have contributed to a transparent scientific transfer of knowledge gathered during the knowlEdge project.

We are looking forward to share further highlights and final project outcomes with you in the final year of the project.