In recent years, the application of additive manufacturing processes in various industries has continued to increase. In this framework, ensuring the quality and repeatability of additively manufactured parts is a basic requirement to meet the strict requirements of leading industries. Several methodologies for in-situ process monitoring and control have been proposed recently, however, there is still a lack of embedded intelligent methods suitable to use the high amount of process data gathered and to detect anomalies and defects robustly and effectively. This thesis work presents a case study based on an open science collaboration project between TRUMPF GmbH, one of major Additive Manufacturing system developers, and Politecnico di Milano. The case study relies on an open dataset including in-line and in-situ signals gathered during the laser powder bed fusion of specimens of aluminum parts on a TRUMPF TruPrint5000 industrial machine. The signals were acquired by means of two photodiodes installed co-axially to the laser path. The dataset is specifically designed to support the development of novel in-situ monitoring methodologies for fast and robust anomaly detection while the part is being built. This thesis work presents novel Statistical Process Monitoring methodologies to detect poor quality layers in the process and detect anomalies as soon as possible. A layer wise statistical monitoring approach as well as innovative methods based on the use of spatial statistics (Moran’s Index) are proposed and preliminary results are presented. The different methodologies have been studied and compared, highlighting advantages and disadvantages of each of them, and a benchmark is done by means of a specificity and sensitivity analysis. Further improvements and other techniques based on advanced clustering methods are then discussed as future research possibilities.
Negli ultimi anni, l'applicazione dei processi di additive maufacturing in varie industrie ha continuato ad aumentare. Garantire la qualità e la ripetibilità delle parti prodotte in modo additivo è un requisito fondamentale per soddisfare i severi requisiti delle industrie leader. Diverse metodologie per il monitoraggio ed il controllo del processo in-situ sono state proposte recentemente, tuttavia, c'è ancora una mancanza di metodi intelligenti incorporati adatti a utilizzare l'elevata quantità di dati di processo raccolti e a rilevare anomalie e difetti in modo robusto ed efficace. Questa tesi presenta un case study basato su un progetto di collaborazione open science tra TRUMPF GmbH, uno dei maggiori sviluppatori di sistemi di Additive Manufacturing, e il Politecnico di Milano. Il case study si basa su un dataset aperto che include segnali in linea e in situ raccolti durante la produzione di campioni alluminio su una macchina industriale TRUMPF TruPrint5000. I segnali sono acquisiti per mezzo di due fotodiodi installati coassialmente al percorso del laser. Il dataset è progettato per supportare lo sviluppo di nuove metodologie di monitoraggio in-situ per il rilevamento rapido e robusto delle anomalie durante la costruzione del pezzo. Questa tesi presenta nuove metodologie di monitoraggio statistico del processo per rilevare le anomalie il prima possibile. Vengono proposti un approccio di monitoraggio statistico layer-wise e metodi innovativi basati sull'uso di statistiche spaziali (indice di Moran) presentandone risultati preliminari. Le diverse metodologie sono studiate e confrontate, evidenziando vantaggi e svantaggi di ciascuna di esse, infine viene fatto un benchmark. Ulteriori miglioramenti e tecniche basate su metodi di clustering avanzati sono poi discussi come possibilità di ricerca futura.
In-situ monitoring of the LBPF process via synthetic and spatial statistics : the TRUMPF open science case study
GRANITO, EMIDIO
2020/2021
Abstract
In recent years, the application of additive manufacturing processes in various industries has continued to increase. In this framework, ensuring the quality and repeatability of additively manufactured parts is a basic requirement to meet the strict requirements of leading industries. Several methodologies for in-situ process monitoring and control have been proposed recently, however, there is still a lack of embedded intelligent methods suitable to use the high amount of process data gathered and to detect anomalies and defects robustly and effectively. This thesis work presents a case study based on an open science collaboration project between TRUMPF GmbH, one of major Additive Manufacturing system developers, and Politecnico di Milano. The case study relies on an open dataset including in-line and in-situ signals gathered during the laser powder bed fusion of specimens of aluminum parts on a TRUMPF TruPrint5000 industrial machine. The signals were acquired by means of two photodiodes installed co-axially to the laser path. The dataset is specifically designed to support the development of novel in-situ monitoring methodologies for fast and robust anomaly detection while the part is being built. This thesis work presents novel Statistical Process Monitoring methodologies to detect poor quality layers in the process and detect anomalies as soon as possible. A layer wise statistical monitoring approach as well as innovative methods based on the use of spatial statistics (Moran’s Index) are proposed and preliminary results are presented. The different methodologies have been studied and compared, highlighting advantages and disadvantages of each of them, and a benchmark is done by means of a specificity and sensitivity analysis. Further improvements and other techniques based on advanced clustering methods are then discussed as future research possibilities.| File | Dimensione | Formato | |
|---|---|---|---|
|
2021_10_Granito.pdf
non accessibile
Dimensione
7.56 MB
Formato
Adobe PDF
|
7.56 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/10589/180153