Multi-View Data Exploration Recipes for Making Sense of Multi-Source Industrial Data

Le résumé fourni par la source

Industrial assets, such as wind turbines, pumps or heavy-duty vehicles, produce large data traces which can be used for diverse purposes. However, making sense of real-world industrial data is challenging, as it is typically produced from different, highly heterogeneous sources, such as sensor measurements, configuration specifications or log records stemming from different components and assets. Applying off-the-shelf artificial intelligence (AI) algorithms does not necessarily yield relevant insights, as these methods struggle to handle the complexity, variability, and domain-specific constraints of the data. In this paper, we bridge this gap, by formalising three data exploration strategies into standardised workflows (recipes) for effectively analysing, integrating, and deriving actionable insights from multi-source industrial data. The benefits of using these recipes for exploring industrial data are demonstrated on multi-source datasets collected during the in-the-field operations of four different types of industrial assets: feedwater pumps, compressors, windmills, and heavy-duty trucks. Our findings underline the importance of conceiving and validating formal data analysis methodologies enabling the optimal exploitation of industrial data without subjecting each new application context to an initial trial-and-error data exploration phase.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.