Alarm forecasting plays a pivotal role within industrial systems and processes across various companies, as it closely ties into the concept of predictive maintenance. By successfully anticipating alarms within a future timeframe, the potential for abrupt component failures within industrial processes can be mitigated. Predictive maintenance revolves around preemptively addressing issues before they manifest, which aligns closely with the artificial intelligence and time series forecasting fields. This thesis aims to draw a comparison between machine learning approaches and time series forecasting techniques in addressing the challenge of alarm forecasting. The central focus lies in highlighting the unique performance attributes and resource requirements of each approach. Specifically, we are going to dig into how these methods perform when it comes to the computational power they need for training and making predictions, what is their accuracy on predictions and how much data is needed for training to have more predictive power. \textbf{Keywords:} alarm forecasting, machine learning, time series forecasting methods, performance comparison
La previsione degli allarmi svolge un ruolo fondamentale nei sistemi e nei processi industriali di varie aziende, in quanto è strettamente legata al concetto di manutenzione predittiva. Anticipando con successo gli allarmi in un arco di tempo futuro, è possibile ridurre la possibilità di guasti improvvisi dei componenti nei processi industriali. La manutenzione predittiva si basa sull'affrontare preventivamente i problemi prima che si manifestino, il che si allinea strettamente con i campi dell'intelligenza artificiale e della modellazione statistica. Questa tesi si propone di affrontare un confronto tra gli approcci di apprendimento automatico e le tecniche di modellazione statistica per affrontare la sfida della previsione degli allarmi. L'obiettivo principale è quello di evidenziare le prestazioni dei vari algoritmi e i requisiti di risorse di ciascun approccio. In particolare, analizzeremo le prestazioni di questi metodi in termini di potenza computazionale necessaria per l'addestramento e la formulazione delle previsioni, l'accuratezza delle previsioni e la quantità di dati necessari per l'addestramento al fine di ottenere un maggiore potere predittivo.
A Comparative Analysis of Machine Learning and Time Series Models for Alarm Forecasting
Alparone, Andrea
2022/2023
Abstract
Alarm forecasting plays a pivotal role within industrial systems and processes across various companies, as it closely ties into the concept of predictive maintenance. By successfully anticipating alarms within a future timeframe, the potential for abrupt component failures within industrial processes can be mitigated. Predictive maintenance revolves around preemptively addressing issues before they manifest, which aligns closely with the artificial intelligence and time series forecasting fields. This thesis aims to draw a comparison between machine learning approaches and time series forecasting techniques in addressing the challenge of alarm forecasting. The central focus lies in highlighting the unique performance attributes and resource requirements of each approach. Specifically, we are going to dig into how these methods perform when it comes to the computational power they need for training and making predictions, what is their accuracy on predictions and how much data is needed for training to have more predictive power. \textbf{Keywords:} alarm forecasting, machine learning, time series forecasting methods, performance comparisonFile | Dimensione | Formato | |
---|---|---|---|
Andrea_Alparone_Thesis_Final.pdf
Open Access dal 19/09/2024
Dimensione
1.74 MB
Formato
Adobe PDF
|
1.74 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/210427