In industrial settings, continuous monitoring of the operation of assets generates a vast amount of data originating from a multitude of very diverse sources. This data allows to study and understand asset performance in real operating conditions, paving the way for failure prediction, machine setting optimisation and many other industrial applications. However, it is not always feasible and neither wise to approach data analytics for such applications by merging all the available data into a single data set, which often leads to information loss. The literature lacks methods to inspect asset performance based on splitting the data in different views corresponding to different types of monitored parameters. The multi-view data analysis method proposed in this work allows to extract operating modes for an industrial asset and subsequently, profile their performance. In this two-step approach, the endogeneous (internal working) data view is first exploited to detect and characterise distinct operating modes, while an exogeneous (operating context) data representation (disjoint with the endogeneous view) of these operating modes is subsequently used to derive prototypical performance profiles via non-negative matrix factorisation. The application potential and validity of the proposed method is illustrated based on real-world data from a wind turbine.
Dhont, M , Tsiporkova, E & Boeva, V 2022, Performance Profiling of Operating Modes via Multi-view Analysis Using Non-negative Matrix Factorisation . in Studies in Big Data. Studies in Big Data, vol. 106, Springer, pp. 289-316.
Dhont, M. , Tsiporkova, E., & Boeva, V. (2022). Performance Profiling of Operating Modes via Multi-view Analysis Using Non-negative Matrix Factorisation . In Studies in Big Data (pp. 289-316). (Studies in Big Data Vol. 106). Springer.
@inbook{4f5fe10c5e574bbcb2c5cf6e6ed15696,
title = " Performance Profiling of Operating Modes via Multi-view Analysis Using Non-negative Matrix Factorisation " ,
abstract = " In industrial settings, continuous monitoring of the operation of assets generates a vast amount of data originating from a multitude of very diverse sources. This data allows to study and understand asset performance in real operating conditions, paving the way for failure prediction, machine setting optimisation and many other industrial applications. However, it is not always feasible and neither wise to approach data analytics for such applications by merging all the available data into a single data set, which often leads to information loss. The literature lacks methods to inspect asset performance based on splitting the data in different views corresponding to different types of monitored parameters. The multi-view data analysis method proposed in this work allows to extract operating modes for an industrial asset and subsequently, profile their performance. In this two-step approach, the endogeneous (internal working) data view is first exploited to detect and characterise distinct operating modes, while an exogeneous (operating context) data representation (disjoint with the endogeneous view) of these operating modes is subsequently used to derive prototypical performance profiles via non-negative matrix factorisation. The application potential and validity of the proposed method is illustrated based on real-world data from a wind turbine. " ,
keywords = " Non-negative matrix factorisation, Performance profiling, Multi-view data, Multi-dimensional binning " ,
author = " Michiel Dhont and Elena Tsiporkova and Veselka Boeva " ,
note = " Funding Information: Funding: This research was subsidised through the projects MISTic and ReWind by the Brussels-Capital Region—Innoviris and received funding from the Flemish Government (AI Research Program). Publisher Copyright: { extcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG. Copyright: Copyright 2022 Elsevier B.V., All rights reserved. " ,
year = " 2022 " ,
month = may,
day = " 21 " ,
doi = " 10.1007/978-3-030-95239-6_11 " ,
language = " English " ,
isbn = " 9783030952389 " ,
series = " Studies in Big Data " ,
publisher = " Springer " ,
pages = " 289316 " ,
booktitle = " Studies in Big Data " ,
}