INSA Strasbourg:
INSA Strasbourg's project update emphasizes data collection and model building. The team has applied 15 temperature sensors across five chambers, alongside humidity and height sensors. Their data preprocessing includes resampling to aggregate hourly data, combining temperature readings for a consolidated input matrix. For their model building they apply regression tasks for quality prediction, utilizing architectures like LSTM and CNN. A notable strategy involves analyzing quality patterns against high-quality reference inputs to understand deviations and categorize quality degradation, further enriching their analysis with chronicles of these events.
INSA Rouen:
INSA Rouen is working on stream reasoning, utilizing RSP4J for dynamic data stream management and SPARQL for querying. Their work focuses on integrating ontologies with stream data to enhance query specificity and accuracy. Through Franco's PhD research, they've established three data streams, augmenting SPARQL queries with ontologies for targeted information extraction. Their efforts extend to ontology reconciliation, exploring various techniques like mappings and matchers to align and integrate different ontologies. This approach enables them to process data streams more effectively, with future work aimed at refining RDF table queries and advancing ontology integration practices.
HFU:
Rudolf and Christoph submitted a paper with the title "Machine Learning with Fault Tree Analysis for Explainable Failure Detection in Cloud Computing". This paper is currently under peer review. Their research tackles the complexities of fault diagnosis in complex systems, addressing the critical gap in interpretability. By employing machine learning to predict basic events within fault trees and using the fault tree to determine the overall failure, they aim to enhance root cause identification and failure mechanism understanding. This way, the fault tree acts as a surrogate model. Their results indicate the potential for improved failure diagnosis and interpretability using fault trees.
INSA Strasbourg:
INSA Strasbourg is focused on enhancing product quality prediction through data mining. The team faces challenges in finding suitable datasets, as most available are image-based, whereas their methodology requires time-series data for dynamic analysis. This data is crucial for identifying trends and making accurate predictions. Their work also extends to developing an explainable AI (XAI)-based model for understanding factors affecting product quality. The current approach involves reviewing existing XAI methods and creating models to predict and explain the State of Charge (SOC) of batteries, incorporating SHAP for explanations and Matrix Profile for anomaly detection.
INSA Rouen:
The team at INSA Rouen encountered limitations with OntopStream, particularly with aggregations and time windows, leading to the decision to transition to RSP4J for RDF Stream Processing. RSP4J's maintenance and flexibility offer significant advantages. Additionally, the team is working on merging ontologies using PrOM, focusing on preprocessing to detect inconsistencies, matching through analysis, and postprocessing to create linked ontologies. This process includes improving class hierarchy and avoiding duplicated classes, aiming to enhance the efficiency of merging sound ontologies for production domain applications, demonstrating a strategic pivot towards more robust and flexible technologies.
HFU:
HFU presented innovative approaches combining machine learning (ML) with fault trees (FT) to enhance predictive accuracy and efficiency. The team outlined five distinct methodologies: using ML to select FT, generate FT, both generate and select FT, and predict events. This interdisciplinary effort showcases HFU's commitment to integrating ML with established analytical techniques to improve reliability and decision-making processes. The approaches offer a glimpse into the potential for ML to revolutionize traditional fault analysis methods, providing a foundation for future research and development in predictive modeling and risk assessment.
The next meeting encompassed a warm welcome to new project members and a comprehensive overview of current progress from each partner:
INSA Strasbourg:
- Welcomed Amel HIDOURI (Postdoc) and Slimane Arbaoui (PhD) to the X-Quality project.
- Amel focuses on an explainable data mining approach for failure understanding, utilizing frequent pattern mining and emerging patterns.
- Slimane works on an explainable deep learning model using graphical patterns (Chronicles), employing CNN and LSTM models for degradation prediction, along with pattern mining-based approaches to explain model predictions.
INSA Rouen:
- Enrolled Léa Charbonnier as a new PhD student, focusing on semantic description and hybrid AI approaches for reliability engineering.
- Considering recruitment of an intern to investigate XAI in manufacturing and ontology-based AI methods.
- Engaged in a literature review and planning future work including ontology compilation and OLAF platform testing.
HFU:
- Presented a survey paper highlighting the role of AI and XAI in visual quality assessment across industries.
- Identified limited XAI adoption and emphasized the need for wider XAI adoption.
- Shared insights from a literature review and provided a table of 8 use cases discovered.
This meeting facilitated a deeper understanding of ongoing efforts and set the stage for future collaboration and progress.
For the upcoming meeting, several key topics are slated for discussion:
Data Management Plan (DMP): Cecilia Zanni-Merk has set up the platform https://dmp.opidor.fr for creating and editing a centralized DMP. A preliminary version of the DMP has been generated and reviewed. Discussion points will include considerations for the DMP, particularly regarding privacy requirements in data acquisition and storage. Topics such as defining the data storage timeframe, determining storage locations (e.g., HFU server), and implementing data encryption for enhanced security will be addressed.
Project Status - HFU: Rudolf Hoffmann has conducted an extensive systematic literature review (SLR) regarding AI and XAI for Visual Quality Assurance. A clustering analysis of the selected studies indicates that most works are focusing on quality control activities using AI, rather than on activities, such as predictive maintenance, process optimization or root cause analysis. Furthermore, the analysis indicates, that AI applications are widespread in this context, however XAI applications are still limited.
