In industrial contexts, prognostics refers to the ability of predicting the Remaining Useful Life (RUL) of critical components and machines that are part of complex production systems. This is the rationale of a new maintenance strategy, named Predictive Maintenance (PM), whose ultimate goal is a near-zero downtimes situation. PM can be implemented through the so-called Prognostics Health Management (PHM) program, which is based on the monitoring of relevant parameters produced by the equipment, e.g., vibrations, in order to assess its health status. Based on the Condition Monitoring (CM) data and according to the PHM paradigm, the degradation process leading to an identified fault condition can be modelled and the RUL computed as the difference between the current time instant and the instant in which the monitored parameter is predicted to overcome a fixed Failure Threshold (FT). The supervised and offline approaches exploited to that purpose have important drawbacks limiting their application. In this paper, a data-driven methodology for the PHM implementation is introduced, which directly processes data in streaming and in an unsupervised way to detect anomalies, automatically find data partitions and compute the RUL as new data is available, after degradation models for clusters representing fault conditions have been identified. In this way, learning algorithms do not require large amount of historical data for training and never-seen incipient failures can be recognized during the machine functioning. The methodology has been applied to a rolling bearing vibration signal, whose degrading behavior until the fault has been simulated. Results show that the methodology is able to recognize a deviating behavior and anticipate the occurrence of the failure with sufficient time. The proposed case study also highlighted the necessity of a supervised and offline part for the FT and degradation model definition.

Data-driven prognostics: From an offline and supervised analysis to an innovative, online and unsupervised methodology / Calabrese, F.; Gamberi, M.; Margelli, S.; Pilati, F.; Regattieri, A.. - In: ...SUMMER SCHOOL FRANCESCO TURCO. PROCEEDINGS. - ISSN 2283-8996. - 2019:(2019), pp. 74-80. ((Intervento presentato al convegno 24th Summer School Francesco Turco, 2019 tenutosi a Brescia nel 11th-13th September 2019.

Data-driven prognostics: From an offline and supervised analysis to an innovative, online and unsupervised methodology

Pilati F.;
2019-01-01

Abstract

In industrial contexts, prognostics refers to the ability of predicting the Remaining Useful Life (RUL) of critical components and machines that are part of complex production systems. This is the rationale of a new maintenance strategy, named Predictive Maintenance (PM), whose ultimate goal is a near-zero downtimes situation. PM can be implemented through the so-called Prognostics Health Management (PHM) program, which is based on the monitoring of relevant parameters produced by the equipment, e.g., vibrations, in order to assess its health status. Based on the Condition Monitoring (CM) data and according to the PHM paradigm, the degradation process leading to an identified fault condition can be modelled and the RUL computed as the difference between the current time instant and the instant in which the monitored parameter is predicted to overcome a fixed Failure Threshold (FT). The supervised and offline approaches exploited to that purpose have important drawbacks limiting their application. In this paper, a data-driven methodology for the PHM implementation is introduced, which directly processes data in streaming and in an unsupervised way to detect anomalies, automatically find data partitions and compute the RUL as new data is available, after degradation models for clusters representing fault conditions have been identified. In this way, learning algorithms do not require large amount of historical data for training and never-seen incipient failures can be recognized during the machine functioning. The methodology has been applied to a rolling bearing vibration signal, whose degrading behavior until the fault has been simulated. Results show that the methodology is able to recognize a deviating behavior and anticipate the occurrence of the failure with sufficient time. The proposed case study also highlighted the necessity of a supervised and offline part for the FT and degradation model definition.
Augmented knowledge: a new era of industrial systems engineering
Roma
AIDI - Italian Association of Industrial Operations Professors
Calabrese, F.; Gamberi, M.; Margelli, S.; Pilati, F.; Regattieri, A.
Data-driven prognostics: From an offline and supervised analysis to an innovative, online and unsupervised methodology / Calabrese, F.; Gamberi, M.; Margelli, S.; Pilati, F.; Regattieri, A.. - In: ...SUMMER SCHOOL FRANCESCO TURCO. PROCEEDINGS. - ISSN 2283-8996. - 2019:(2019), pp. 74-80. ((Intervento presentato al convegno 24th Summer School Francesco Turco, 2019 tenutosi a Brescia nel 11th-13th September 2019.
File in questo prodotto:
File Dimensione Formato  
data drive prognostic.pdf

Solo gestori archivio

Tipologia: Versione editoriale (Publisher’s layout)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 951.61 kB
Formato Adobe PDF
951.61 kB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11572/286356
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? ND
social impact