Predictive maintenance (PM) is the maintenance policy where the inefficient use of resources is reduced, due to unnecessary preventive or corrective actions, and the operations costs are optimized thanks to advanced monitoring and diagnosis of components. The growing spread of real-time control systems in manufacturing companies increases the amount of available data about the state of monitored components. This paper investigates the use of machine learning (ML) techniques in monitoring the global health-state of machineries and facilities in general. For ensuring the success of the ML predictive model it is important to select and extract the coherent set of features that describes asset operations. The literature analysis shows that a statistical approach for this feature selection is preferred. However, the practice suggests that it is necessary to know the asset engineering model and its failure mode for establishing the architecture of data collection. Furthermore, the development of maintenance strategy is constrained by quality and maturity of dataset. The aim of this paper is to introduce a framework that will guide maintenance engineers to choose and validate the correct set of features starting from the engineering model. An illustrative case study will show the application of this framework and its benefits. Since this research is at the preliminary stage, finally the future steps are proposed.
Machine learning for predictive maintenance: a methodological framework
eleonora florian
;fabio sgarbossa;ilenia zennaro
2019
Abstract
Predictive maintenance (PM) is the maintenance policy where the inefficient use of resources is reduced, due to unnecessary preventive or corrective actions, and the operations costs are optimized thanks to advanced monitoring and diagnosis of components. The growing spread of real-time control systems in manufacturing companies increases the amount of available data about the state of monitored components. This paper investigates the use of machine learning (ML) techniques in monitoring the global health-state of machineries and facilities in general. For ensuring the success of the ML predictive model it is important to select and extract the coherent set of features that describes asset operations. The literature analysis shows that a statistical approach for this feature selection is preferred. However, the practice suggests that it is necessary to know the asset engineering model and its failure mode for establishing the architecture of data collection. Furthermore, the development of maintenance strategy is constrained by quality and maturity of dataset. The aim of this paper is to introduce a framework that will guide maintenance engineers to choose and validate the correct set of features starting from the engineering model. An illustrative case study will show the application of this framework and its benefits. Since this research is at the preliminary stage, finally the future steps are proposed.File | Dimensione | Formato | |
---|---|---|---|
paper summer school 2019.pdf
accesso aperto
Descrizione: articolo
Tipologia:
Published (Publisher's Version of Record)
Licenza:
Accesso gratuito
Dimensione
1.04 MB
Formato
Adobe PDF
|
1.04 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.