24 Feb 2025

Through the development of innovative solutions in Big Data and Machine Learning, the “Data-driven IT Services for Sustainable and Efficient Manufacturing (DISSEM)” project, funded by a PR-FESR 2021-2027 grant and coordinated by Prof. Mauro Tortonesi of the IN4 laboratory, aims to create a widely applicable IT service platform for optimizing production processes, enabling efficient, waste-free, and defect-free Manufacturing 5.0.

Prof. Tortonesi, what is the status of the DISSEM project one year after its launch?

We are still in the early stages of the second year, with the project set to conclude in mid-2026, but we have already achieved significant milestones in both the development of innovative data analysis methodologies and the implementation of the DISSEM platform.

Thanks to the collaboration with our industrial partners Bonfiglioli Group and Carpigiani Group, who provided representative use cases and datasets collected from real-life equipment, we have developed advanced models based on Generative AI to generate synthetic data, addressing the common issue of imbalanced datasets in Industry 5.0 applications.

Additionally, we have designed an innovative methodology for anomaly detection in multidimensional time series, a type of data that is typical in Industry 5.0 and notoriously difficult to analyze. Our solution leverages neural networks to transform a multidimensional time series into a one-dimensional embedding, making it easier to process with Machine Learning (ML) algorithms.

Finally, we have created a highly innovative methodology, which we called “multi-milestone”, to enable efficient, real-time anomaly detection at multiple stages of a production process. This approach allows a single ML model to analyze data both at the beginning of the process—with limited accuracy but ample time for corrective actions—and in later stages, with progressively increasing accuracy.

In parallel, together with our partner laboratories CIRI ICT at the University of Bologna, CRIS at the University of Modena and Reggio Emilia, and our industry partner Imola Informatica, we are working on methodologies and tools for distributed learning and a scalable platform for ML model lifecycle management.

What benefits do your data analysis methodologies provide?

Our solutions offer companies widely applicable and easy-to-adopt tools that enable the development of sophisticated data analysis applications in Industry 5.0 with low CAPEX and reduced time-to-market, leading to high productivity.

DISSEM allows companies to access cutting-edge technologies and best practices in data analysis and Machine Learning without the need to develop in-house expertise—which would be costly and require significant investments. In the field of data analysis and Machine Learning, keeping up with the rapid pace of scientific and technological advancements is extremely challenging, and the gap between scientific research and applied research is narrowing.

For example, we presented our embedding-based anomaly detection methodology, developed within the DISSEM project, at the 2024 edition of ITADATA, the leading conference for the Italian data analysis research community. Our work was highly appreciated by the reviewers and the organizers, and even received the Best Paper Award (ex aequo). This achievement highlights the value of our team and the scientific relevance of the project.

What are your goals for the coming year?

We aim to further develop our work on distributed learning, advance the platform, and test our solutions with our industrial partners. Specifically, we are focusing on semi-supervised Federated Learning tools, which allow training models without centralizing data, improving efficiency and privacy. The goal is to enable on-device data analysis, as in the case of Carpigiani’s ice cream machines, which operate worldwide, often in areas with limited internet connectivity. This would allow the machine to detect if the operator has loaded an unbalanced mixture and automatically take corrective actions.

More broadly, DISSEM aims to revolutionize the management of industrial AI, making it more accessible, sustainable, and efficient, with a significant impact on both research and industry. The DISSEM platform, developed using open-source and Cloud-native technologies such as KubeFlow, is designed to be scalable, portable, and easy to use—even for SMEs.

For more information about the project or to get in touch with the research group at Ferrara Technopole, please contact us at: tecnopolo@unife.it.