The emergence of Industry 4.0 and the associated rise of sensing technologies in industrial equipment has made the adoption of Machine Learning (ML) solutions crucial in enhancing and making more efficient enterprise production processes. However, the high demand for ML models often clashes with the small number of professionals capable of handling such projects. For this reason, Automatic Machine Learning (AutoML) tools are considered high-value solutions, thanks to their capability to provide a suitable model for the provided data without the need for the intervention by a ML expert. Indeed, AutoML libraries are designed to be used in an easy way also for people with limited or no experience with ML. In this work, three important tasks involved in manufacturing, that are Anomaly Detection, Visual Anomaly Detection, and Remaining Useful Life Estimation, are considered. After analysing the most critical aspects of each task and the state of the art, possible solutions are proposed for the development of specialised AutoML modules. Additionally, given the increasing emphasis on the interpretability of MLmodels, part of the analysis performed aims at identifying explainability tools, which are particularly important for an AutoML library. In fact, they provide useful motivations for the model predictions, increasing also the user confidence in AutoML tools.

The emergence of Industry 4.0 and the associated rise of sensing technologies in industrial equipment has made the adoption of Machine Learning (ML) solutions crucial in enhancing and making more efficient enterprise production processes. However, the high demand for ML models often clashes with the small number of professionals capable of handling such projects. For this reason, Automatic Machine Learning (AutoML) tools are considered high-value solutions, thanks to their capability to provide a suitable model for the provided data without the need for the intervention by a ML expert. Indeed, AutoML libraries are designed to be used in an easy way also for people with limited or no experience with ML. In this work, three important tasks involved in manufacturing, that are Anomaly Detection, Visual Anomaly Detection, and Remaining Useful Life Estimation, are considered. After analysing the most critical aspects of each task and the state of the art, possible solutions are proposed for the development of specialised AutoML modules. Additionally, given the increasing emphasis on the interpretability of MLmodels, part of the analysis performed aims at identifying explainability tools, which are particularly important for an AutoML library. In fact, they provide useful motivations for the model predictions, increasing also the user confidence in AutoML tools.

AutoML for Advanced Monitoring in Digital Manufacturing and Industry 4.0

PERATONER, ALESSANDRO
2022/2023

Abstract

The emergence of Industry 4.0 and the associated rise of sensing technologies in industrial equipment has made the adoption of Machine Learning (ML) solutions crucial in enhancing and making more efficient enterprise production processes. However, the high demand for ML models often clashes with the small number of professionals capable of handling such projects. For this reason, Automatic Machine Learning (AutoML) tools are considered high-value solutions, thanks to their capability to provide a suitable model for the provided data without the need for the intervention by a ML expert. Indeed, AutoML libraries are designed to be used in an easy way also for people with limited or no experience with ML. In this work, three important tasks involved in manufacturing, that are Anomaly Detection, Visual Anomaly Detection, and Remaining Useful Life Estimation, are considered. After analysing the most critical aspects of each task and the state of the art, possible solutions are proposed for the development of specialised AutoML modules. Additionally, given the increasing emphasis on the interpretability of MLmodels, part of the analysis performed aims at identifying explainability tools, which are particularly important for an AutoML library. In fact, they provide useful motivations for the model predictions, increasing also the user confidence in AutoML tools.
2022
AutoML for Advanced Monitoring in Digital Manufacturing and Industry 4.0
The emergence of Industry 4.0 and the associated rise of sensing technologies in industrial equipment has made the adoption of Machine Learning (ML) solutions crucial in enhancing and making more efficient enterprise production processes. However, the high demand for ML models often clashes with the small number of professionals capable of handling such projects. For this reason, Automatic Machine Learning (AutoML) tools are considered high-value solutions, thanks to their capability to provide a suitable model for the provided data without the need for the intervention by a ML expert. Indeed, AutoML libraries are designed to be used in an easy way also for people with limited or no experience with ML. In this work, three important tasks involved in manufacturing, that are Anomaly Detection, Visual Anomaly Detection, and Remaining Useful Life Estimation, are considered. After analysing the most critical aspects of each task and the state of the art, possible solutions are proposed for the development of specialised AutoML modules. Additionally, given the increasing emphasis on the interpretability of MLmodels, part of the analysis performed aims at identifying explainability tools, which are particularly important for an AutoML library. In fact, they provide useful motivations for the model predictions, increasing also the user confidence in AutoML tools.
AutoML
Industry 4.0
RUL Estimation
Anomaly Detection
Visual AD
File in questo prodotto:
File Dimensione Formato  
Peratoner_Alessandro.pdf

accesso aperto

Dimensione 12.7 MB
Formato Adobe PDF
12.7 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/43335