A team of researchers from our EluciDATA Lab wins Best Industry Paper Award at IEEE Conference on Prognostics and Health Management (Montreal, Québec, June 5 - June 7, 2023)Investigating the performance of industrial assets poses a challenge due to the complex dynamic operating environment that machines typically are exposed to under in real-world conditions. In order to avoid unplanned and expensive downtime it is crucial to properly assess the asset’s health status. Our award-winning paper, titled "Multi-view Contextual Performance Profiling in Rotating Machinery," addresses this challenge by introducing an innovative unsupervised approach. As part of AI-ICON CONSCIOUS project, and in collaboration with I-Care and Engie, our research confronts the task of accurately estimating an asset's performance by considering the varying contexts to which industrial assets are exposed during operation. Our work emphasizes the importance of understanding the factors that shape these contexts and account for them during asset condition monitoring and profiling. We validated the approach on a real-world dataset of feedwater pumps, demonstrating its tangible applicability to industrial problems. In this article we will sketch how our insightful approach can enhance asset management strategies.
The proposed methodology leverages a multi-view representation and matrix decomposition to derive specific vibration fingerprints characterizing the asset’s behavior in a context-sensitive fashion. The approach processes data in two separate views: the process view and the vibration view.
In the process view, variables related to the asset's internal workings are analyzed and partitioned, with each measurement point associated with a specific label representing the operating modes of the asset. On the other hand, the vibration view focuses on extracting vibration profiles through non-negative matrix decomposition. By linking these two views together, the researchers can derive characteristic fingerprints using a suitable contextual representation and performance-related indicators. This methodology was validated using real-world industrial data from feedwater pumps, comprising vibration and operational sensor measurements. The results obtained from the validation demonstrate the efficacy of the profiling methodology in delivering meaningful risk assessment estimations associated with different operating contexts. At the 2023 IEEE Conference on Prognostics and Health Management our paper was awarded with the Best Industry Paper Award recognizing the continuous effort we put in our day-to-day work at the EluciDATA Lab to bridge the gap between AI research and practical applications for the industry.
The results obtained from the validation demonstrate the efficacy of the profiling methodology in delivering meaningful risk assessment estimations associated with different operating contexts. This award can be seen as an affirmation of our day-to-day work at the EluciDATA Lab to bridge the gap between AI research and practical applications for the industry.
Authors: EluciDATA LabPermanent URL