Additive Manufacturing (AM) is one of the major technological game changers of the last decades. It allows the production of complex-shaped, lightweight bioinspired parts with high thermomechanical performance in a large variety of materials ranging from metals to composites and ceramics. The significant investments of many companies in different sectors, namely aerospace, automotive, oil and gas, and healthcare, are clear proof of the technological relevance of AM technologies in the industrial panorama. Regardless of the specific AM technique, the lack of process stability and repeatability is still considered a major issue for the technological outbreak, as it can strongly impact process defectiveness, which is a crucial factor for poor mechanical performance, increased costs due to the expensive raw material and long processing times. For this reason, the attention of the scientific and industrial communities is currently focused on in-situ monitoring solutions, which can in principle prevent and mitigate material waste and part failure. This research field has been indeed expanding at a fast rate in the last few years, with exponential growth especially focusing on AM technologies that recently came under the spotlight. Among them, Extrusion-based AM processes are receiving renewed attention for their flexibility, and the possibility of printing large-scale products (higher than 1 m3) in a large variety of materials (techno-polymers, metals, and ceramics), primarily for machinery and aerospace sectors. So far, many studies on in-situ monitoring of Extrusion-based AM processes focused mainly on proposing new sensing acquisition setups with the main objective of studying the process evolution. Few preliminary works have combined big data mining with machine learning and statistical process monitoring solutions to fully address in-situ defect detection. Regardless of the AM technique, assuming a defect is automatically detected by the in-situ monitoring solution, few approaches have been proposed for inline defect mitigation, correction and/or removal. These research challenges are the bases for this Ph.D. thesis, which is meant to discuss novel solutions for defect detection in Extrusion-based AM processes and to present a brand-new patented prototype for inline defect correction. Two main strategies will be considered for statistical process monitoring and control for in-situ defect detection. The first one makes use of the dynamic process signatures (i.e., process features evolving during the printing process) to characterize the process stability as related to the temporal changes of the process fingerprint. The second dimension of research mainly focuses on the layerwise level, taking advantage of images collected at the end of the printing layer in order to analyze the part quality at each stage. The first approach for dynamic defect detection presents a new solution for modeling and controlling the spatial maps of the temperature evolution since this is a major driver of material bonding quality thus affecting the final mechanical performances. This work was partially carried out during the period abroad at the Massachusetts Institute of Technology – MIT, together with Prof. Anastasios John Hart and Gregory Dreyfus (MIT), working on a case study of Big Area Additive Manufacturing (BAAM) from the Oak Ridge National Laboratory. The second solution for SPC based on layerwise images consists of a novel method for automatic flaw detection when a stochastic textured surface has to be analyzed. Several new elements characterize this second research result, as out-of-control detection of local defects in highly textured surfaces can be a complex task that currently lacks appropriate solutions. In this case, a novel methodology based on random trees and distribution similarity measures will be combined with statistical process monitoring to face these open challenges. Although applied to Extrusion-based technology, defect detection of layerwise images is a general approach that can be applied to other AM processes, like Powder Bed Fusion (PBF). With reference to PBF, the Ph.D. thesis will eventually present a novel approach for in-situ and in-line defect correction, to move a step forward in the direction of zero-waste, first-time-right AM. The proposed solution originated from a newly patented prototype named “Penelope” representing the first highly-sensorized AM machine that integrates a rectilinear grinder devoted to defective layer removal. The capability of detecting major dimensional and geometrical defects will be first discussed and the in-situ acquisition system performances will be characterized. This part of the study will determine which magnitude of dimensional defects can be detected by the in-situ setups and what are the best illumination condition and segmentation algorithm for an accurate inline measurement. Finally, the effects of layer grinding will be discussed. More in detail, the possibility of restoring the powder bed, eliminating chip contamination, as well as the capability of preserving mechanical performances and product quality will be deepened. All the experimental campaigns were specifically designed and tested in the AddMe Lab laboratories of Politecnico di Milano Even if not included in the discussion of this Ph.D. thesis, a final chapter will briefly describe an additional activity carried out during the Ph.D., focusing on multimaterial L-PBF.
L’Additive Manufacturing (AM) è un'insieme di tecnologie produttive che hanno cambiato lo scenario manifatturiero mondiale degli ultimi 10 anni. Queste permettono la produzioni di parti funzionali dalla geometria complessa e topologicamente ottimizzata e dal peso ridotto, in una vasta gamma di materiali, quali leghe metalliche, ceramici, polimeri e compositi. I significativi investimenti provenienti dai settori aerospaziale e sanitario sono la prova tangibile della rilevanza tecnologica dell’AM. Per tutte le tecnologie AM esistenti, la scarsa stabilità e ripetibilità del processo sono considerate un limite importante per la loro diffusione, a causa del loro impatto sulla qualità finale del prodotto, e quindi sui costi e tempi di produzione, ispezione, e post-processing. Per queste ragioni, l’attenzione della cominità scientifica e industriale si è concentrata sulla realizzazione di soluzioni di monitoraggio in-situ, atte all’identificazione e alla mitigazione inline dei difetti. Questo stream di ricerca ha visto una crescita esponenziale negli ultimi anni, particolarmente per le tecnologie più richieste e promettenti per il mercato. Tra queste, le tecnologie Additive basate sull’estrusione di materiale (Material Extrusion – ME) sono cresciute particolarmente per la possibilità di stampare ampi volumi (sopra il m^3) e per l’estensione della gamma di materiali stampabili (tecnopolimeri, metalli, ceramici, etc.). Fino ad ora, gli studi focalizzati sul monitoraggio in-situ di questa classe di processi si è concentrata sull’implementazione di sistemi di acquisizione e su studi di fattibilità, ma pochi lavori hanno proposto tecniche statistiche e/o di machine learning per l’identificazione automatica dei difetti. Inoltre, a prescindere dalla tecnologia, pochissimi lavori presentano soluzioni per la mitigazione o la rimozione inline dei difetti. Questi gap nella ricerca sono i pilastri di questo lavoro di dottorato, il quale si concentrerà sulla realizzazione di tecniche in-situ per l’identificazione di difetti i processi ME, grazie all’osservazioni di caratteristiche di processo dinamiche (ovvero che cambiano continuamente lungo il processo) o statiche (osservabili in particolari istanti discreti del processo). Per quanto riguarda il primo approccio di monitoraggio, questo lavoro si concentrerà sulla realizzazionie di metodo statistici per l’analisi ed il monitoraggio della storia termica, in quanto driver principale dei meccanismo di fusione e legame del materiale. Questo lavoro è stato condotto in collaborazione con il Massachusetts Institute of Technology – MIT, insieme al Prof. Anastasios John Hart e Gregory Dreyfus (MIT), lavorando su un caso studio di Big Area Additive Manufacturing (BAAM) svolto presso l’Oak Ridge National Laboratory. Il secondo filone di ricerca si focalizzerà sull’analisi della texture del layer attraverso la combinazione di metodi SPC e di analisi immagini indirizzati alle Stochastic Textured Surfaces. La soluzione presentata permette di identificare automaticamente difetti di deposizione (sovra e sotto-estrusione) grazie ad una tecnica che combina la modellazione della texture grazie a un modello di Random Forest, il confronto dei residui attraverso l’indice Anderson- Darling e il monitoraggio dei risultati attraverso ad un indicatore sintetico basato su k-means clustering. Infine, il presente lavoro tratterà la possibilità di rimuovere in linea eventuali difetti, presentando il proof-of-concept Penelope, la prima macchina ibrida altamente sensorizzata che combina la tecnologia Lase Powder bed Fusion (L-PBF) con la rettifica lineare, indirizzata all’eliminazione dei layer difettosi. In questo lavoro, la caratterizzazione del sistema in-situ di acquisizione di immagini ad alta definizione e la fattibilità della rimozione di uno o più layer online sono state investigate. I risultati hanno mostrato che, l’accuratezza del sistema permette l’identificazione di di difetti geometrici e dimensionali in-plane e out-of plane, permettendo l’identificazione di una vasta gamma di difetti. Infine, lo studio del letto di polvere, della polvere riciclata, e delle performace meccaniche delle parti soggette a rimozione del layer confermano l’applicabilità del proof-of- concept in campo industriale.
In-situ monitoring for defects detection and correction in additive manufacturing processes
Caltanissetta, Fabio
2021/2022
Abstract
Additive Manufacturing (AM) is one of the major technological game changers of the last decades. It allows the production of complex-shaped, lightweight bioinspired parts with high thermomechanical performance in a large variety of materials ranging from metals to composites and ceramics. The significant investments of many companies in different sectors, namely aerospace, automotive, oil and gas, and healthcare, are clear proof of the technological relevance of AM technologies in the industrial panorama. Regardless of the specific AM technique, the lack of process stability and repeatability is still considered a major issue for the technological outbreak, as it can strongly impact process defectiveness, which is a crucial factor for poor mechanical performance, increased costs due to the expensive raw material and long processing times. For this reason, the attention of the scientific and industrial communities is currently focused on in-situ monitoring solutions, which can in principle prevent and mitigate material waste and part failure. This research field has been indeed expanding at a fast rate in the last few years, with exponential growth especially focusing on AM technologies that recently came under the spotlight. Among them, Extrusion-based AM processes are receiving renewed attention for their flexibility, and the possibility of printing large-scale products (higher than 1 m3) in a large variety of materials (techno-polymers, metals, and ceramics), primarily for machinery and aerospace sectors. So far, many studies on in-situ monitoring of Extrusion-based AM processes focused mainly on proposing new sensing acquisition setups with the main objective of studying the process evolution. Few preliminary works have combined big data mining with machine learning and statistical process monitoring solutions to fully address in-situ defect detection. Regardless of the AM technique, assuming a defect is automatically detected by the in-situ monitoring solution, few approaches have been proposed for inline defect mitigation, correction and/or removal. These research challenges are the bases for this Ph.D. thesis, which is meant to discuss novel solutions for defect detection in Extrusion-based AM processes and to present a brand-new patented prototype for inline defect correction. Two main strategies will be considered for statistical process monitoring and control for in-situ defect detection. The first one makes use of the dynamic process signatures (i.e., process features evolving during the printing process) to characterize the process stability as related to the temporal changes of the process fingerprint. The second dimension of research mainly focuses on the layerwise level, taking advantage of images collected at the end of the printing layer in order to analyze the part quality at each stage. The first approach for dynamic defect detection presents a new solution for modeling and controlling the spatial maps of the temperature evolution since this is a major driver of material bonding quality thus affecting the final mechanical performances. This work was partially carried out during the period abroad at the Massachusetts Institute of Technology – MIT, together with Prof. Anastasios John Hart and Gregory Dreyfus (MIT), working on a case study of Big Area Additive Manufacturing (BAAM) from the Oak Ridge National Laboratory. The second solution for SPC based on layerwise images consists of a novel method for automatic flaw detection when a stochastic textured surface has to be analyzed. Several new elements characterize this second research result, as out-of-control detection of local defects in highly textured surfaces can be a complex task that currently lacks appropriate solutions. In this case, a novel methodology based on random trees and distribution similarity measures will be combined with statistical process monitoring to face these open challenges. Although applied to Extrusion-based technology, defect detection of layerwise images is a general approach that can be applied to other AM processes, like Powder Bed Fusion (PBF). With reference to PBF, the Ph.D. thesis will eventually present a novel approach for in-situ and in-line defect correction, to move a step forward in the direction of zero-waste, first-time-right AM. The proposed solution originated from a newly patented prototype named “Penelope” representing the first highly-sensorized AM machine that integrates a rectilinear grinder devoted to defective layer removal. The capability of detecting major dimensional and geometrical defects will be first discussed and the in-situ acquisition system performances will be characterized. This part of the study will determine which magnitude of dimensional defects can be detected by the in-situ setups and what are the best illumination condition and segmentation algorithm for an accurate inline measurement. Finally, the effects of layer grinding will be discussed. More in detail, the possibility of restoring the powder bed, eliminating chip contamination, as well as the capability of preserving mechanical performances and product quality will be deepened. All the experimental campaigns were specifically designed and tested in the AddMe Lab laboratories of Politecnico di Milano Even if not included in the discussion of this Ph.D. thesis, a final chapter will briefly describe an additional activity carried out during the Ph.D., focusing on multimaterial L-PBF.| File | Dimensione | Formato | |
|---|---|---|---|
|
PhDthesis_Caltanissetta_2022_v30 - Copy.pdf
solo utenti autorizzati dal 29/08/2023
Descrizione: PhDthesis_Caltanissetta_2022
Dimensione
8.96 MB
Formato
Adobe PDF
|
8.96 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/10589/220337