During the last decades, advances related to the Internet of Things technologies have led to the widespread adoption of condition monitoring approaches for failure diagnosis. This fundamental vision has resulted in the introduction of advanced maintenance techniques such as Condition-Based Maintenance and Predictive Maintenance. Within this context, one of the main challenges arises from pre-processing the data acquired through sensors and, more specifically, the denoising procedures. Choosing a solid tool to separate the noise signal from the true one is usually considered a difficult task. In addition, data are usually collected from several different sensors, each of which monitors a specific process variable. Reducing the number of features analyzed could be helpful to improve the accuracy of the subsequent diagnosis and, at the same time, in making it less time-consuming. As a result, this paper aims at presenting a diagnosis approach based on Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA) for noise removal and feature reduction respectively. Indeed, EMD is very suitable for highly dynamic and non-stationary signals, while PCA is a good renowned approach for reducing the dimension of a multi-variate signal. Subsequently, a supervised machine learning approach is employed to classify the acquired signal. To demonstrate the applicability of the methodology, a compressor operating within a geothermal plant is considered as a case study, while the selected failure mode is the surge. The developed approach could be exploited by maintenance engineers and asset managers to perform diagnoses on relevant equipment
Integration of Empirical Mode Decomposition and Machine Learning for operating condition classification: a proposal / L. Leoni, A. Cantini, F. De Carlo, M. Tucci. - ELETTRONICO. - (2022), pp. 1-9. (Intervento presentato al convegno 27th Summer School "Francesco Turco" tenutosi a Sanremo (IM) nel 7-9 September 2022).
Integration of Empirical Mode Decomposition and Machine Learning for operating condition classification: a proposal
L. Leoni
Investigation
;A. CantiniMethodology
;F. De CarloFormal Analysis
;M. TucciMethodology
2022
Abstract
During the last decades, advances related to the Internet of Things technologies have led to the widespread adoption of condition monitoring approaches for failure diagnosis. This fundamental vision has resulted in the introduction of advanced maintenance techniques such as Condition-Based Maintenance and Predictive Maintenance. Within this context, one of the main challenges arises from pre-processing the data acquired through sensors and, more specifically, the denoising procedures. Choosing a solid tool to separate the noise signal from the true one is usually considered a difficult task. In addition, data are usually collected from several different sensors, each of which monitors a specific process variable. Reducing the number of features analyzed could be helpful to improve the accuracy of the subsequent diagnosis and, at the same time, in making it less time-consuming. As a result, this paper aims at presenting a diagnosis approach based on Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA) for noise removal and feature reduction respectively. Indeed, EMD is very suitable for highly dynamic and non-stationary signals, while PCA is a good renowned approach for reducing the dimension of a multi-variate signal. Subsequently, a supervised machine learning approach is employed to classify the acquired signal. To demonstrate the applicability of the methodology, a compressor operating within a geothermal plant is considered as a case study, while the selected failure mode is the surge. The developed approach could be exploited by maintenance engineers and asset managers to perform diagnoses on relevant equipmentFile | Dimensione | Formato | |
---|---|---|---|
50.pdf
accesso aperto
Descrizione: Final
Tipologia:
Pdf editoriale (Version of record)
Licenza:
Open Access
Dimensione
523.11 kB
Formato
Adobe PDF
|
523.11 kB | Adobe PDF |
I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.