Laser metal deposition (LMD) is one of the most significant additive manufacturing technologies thanks to many advantages. Since it is a highly flexible process, the building of complex-shaped parts is possible. In the production process, any specific mould is not required. Besides its advantages, there are some drawbacks. The LMD process is highly dependent on the process parameters that must be optimized. In addition, due to a different growth in the layers close to the substrate and a consequent change in the stand of distance some geometrical imprecisions such as over or under-built occurs. Although numerous researches were done, the process still has not been comprehended completely. Due to the drawbacks, monitoring the LMD process is highly important. Process monitoring methods are mostly used to acquire thermal and geometrical data from molten pool due to the relation between these data and process stability. Most important geometrical charecteristics for the LMD process stability is molten pool width and height. Image processing algorithms are developed to detect the molten pool to compute its geometrical characteristics mostly by using thresholding methods which are based on statistical calculations. Since thresholding method algorithms are not specifically designed for computing the molten pool characteristics, several methods are researched to find a proper method. Therefore, there is not a thresholding method which dominates the resources and gives significant benefits over the others. In this thesis, an image processing algorithm is developed to compute the molten pool geometrical characteristics by using the coaxially acquired NIR images which are obtained from video monitoring of the LMD process. Developed algorithm computes the deposition direction and the width, length, and area belonging to the molten pool with three different thresholding method which are Otsu, Isodata and multi-Otsu thresholding methods. The computed molten pool widths are compared with the track widths belonging LMD products to be able to choose the best method. Algorithm is created in MATLAB and tested in 2 experimental campaigns for both small laser spot and large laser spot. The small laser spot is more precise, and it is used for fine features, while the large laser spot is used for massive components, and it is used to improve the productivity and the deposition efficiency. The AISI 316L is used as selected material. First campaign which is thin wall is used to tune and analyze the algorithm by different layer number specimens. The newly deposited layers reheat the previous deposited ones and effect geometrical properties. To be able to sure that analyze of the algorithm is not affect from these phenomena, track width measurements are taken from the highest layer of each specimens in thin wall experiment. Second campaign which is hollow cylinder is used to validate the algorithm with varying the process parameters and determine the best thresholding method. Thin wall results shows that all the methods work properly with the algorithm and developed algorithm is capable to detect deposition angle as stable for small laser sport and with an exaptable variance for large laser spot. Hollow cylinder results show that developed algorithm is capable to react to different process parameters with a sufficient respond time. The best tested threshold method is observed as multi-Otsu for all the hollow cylinder experiments. The mean averaged error range of multi-Otsu is calculated as 3-10%. Otsu gives the second best results with 3-17% mean averaged error range. Isodata gives the worst results with 4-21% mean averaged error range. It is found that using multi-Otsu method with two thresholds is a successful approach to detect the molten pool geometry.
La deposizione laser di metalli (LMD) è una delle tecnologie di produzione additiva più significative grazie ai suoi numerosi vantaggi. Poiché si tratta di un processo altamente flessibile, è possibile la costruzione di parti di forma complessa. Nel processo produttivo non è richiesto nessuno stampo specifico. Inoltre, l'LMD provoca meno sprechi rispetto ai metodi tradizionali. Oltre ai vantaggi appena citati, ci sono tuttavia alcuni svantaggi che vanno considerati. L'LMD è un processo fortemente influenzato dai parametri di processo che devono pertanto essere ottimizzati. Inoltre la crescita degli strati depositati non è costante mostrando altezze di deposizione maggiore negli strati vicini al substrato di partenza. Tale fenomeno provoca una variazione della stand off distance e la generazione di geometri imprecise. Nonostante la ricerca si sia focalizzata si tale aspetto tale problematica non è ancora stata risolta. per tale ragione il monitoraggio dell'LMD è importante. I metodi di monitoraggio del processo sono utilizzati per acquisire informazioni termiche e geometriche dalla pozza fusa per la relazione che intercorrere tra tali dati e la stabilità del processo. Le più importanti caratteristiche geometriche legate alla stabilità del processo sono larghezza e altezza della pozza fusa. Algoritmi dedicarti sono stati sviluppati per analizzare la pozza fusa attraverso immagini coassiali nel NIR ottenute attraverso camere di monitoraggio. L'algoritmo sviluppato permette di analizzare la direzione di deposizione cos' come larghezza, lunghezza e area utilizzando tre metodi per l'identificazione della soglia quali Otsu, Isodata e multi-Otsu. I valori previsti della larghezza della pozza fusa vengono confrontati con le misurazioni della larghezza della traccia di deposizione per individuare il metodo migliore. L'algoritmo, reato in Matlab, è stato testato in due campagne sperimentali realizzate con spot laser piccoli per le lavorazioni in cui è richiesta maggiore precisione e con spot laser grandi per le lavorazioni che richiedono elevata produttività. L'acciaio inossidabile AISI 316 L è stato selezionato come materiale per la deposizione. La prima campagna sperimentale si è focalizzata sulla realizzazione di pareti sottili per studiare l'algoritmo su campioni a diverso numero di strati ed eliminare l'effetto dell'eccessiva crescita nei primi. LA seconda campagna ha focalizzato l'attenzione sulla stampa di tubi per la validazione dell'algoritmo al variare dei parametri di processo per la determinazione del metodo migliore. Le pareti sottili hanno mostrato che tutti e tre i metodi di soglia funzionano con l'algoritmo e lo stesso è in grado di determinare l'angolo di deposizione con spot piccoli, mentre si è dimostrato più affetto da variabilità nel caso di spot grandi. I risultati della seconda campagna mostrano che l'algoritmo è in grado di rispondere bene alla variazione dei parametri di processo. Il miglior metodo tuttavia si è rivelato essere il multi-Otsu per tutti gli esperimenti condotti. L'errore medio con tale metodo varia tra il 3% e 10%. Otsu da il secondo milgior risultato con errori tra il 3% e il 17%. Infine il metodo Isodata da il peggior risultato con errori fino al 21%. In conclusione il metodo multi-Otsu si è dimostrato soddisfacente per determinare le caratteristiche geometriche della pozza fusa.
Development of an algorithm for measuring molten pool geometrical characteristics from Coaxial NIR Imaging in Laser Metal Deposition
AKSOY, OVUNC
2021/2022
Abstract
Laser metal deposition (LMD) is one of the most significant additive manufacturing technologies thanks to many advantages. Since it is a highly flexible process, the building of complex-shaped parts is possible. In the production process, any specific mould is not required. Besides its advantages, there are some drawbacks. The LMD process is highly dependent on the process parameters that must be optimized. In addition, due to a different growth in the layers close to the substrate and a consequent change in the stand of distance some geometrical imprecisions such as over or under-built occurs. Although numerous researches were done, the process still has not been comprehended completely. Due to the drawbacks, monitoring the LMD process is highly important. Process monitoring methods are mostly used to acquire thermal and geometrical data from molten pool due to the relation between these data and process stability. Most important geometrical charecteristics for the LMD process stability is molten pool width and height. Image processing algorithms are developed to detect the molten pool to compute its geometrical characteristics mostly by using thresholding methods which are based on statistical calculations. Since thresholding method algorithms are not specifically designed for computing the molten pool characteristics, several methods are researched to find a proper method. Therefore, there is not a thresholding method which dominates the resources and gives significant benefits over the others. In this thesis, an image processing algorithm is developed to compute the molten pool geometrical characteristics by using the coaxially acquired NIR images which are obtained from video monitoring of the LMD process. Developed algorithm computes the deposition direction and the width, length, and area belonging to the molten pool with three different thresholding method which are Otsu, Isodata and multi-Otsu thresholding methods. The computed molten pool widths are compared with the track widths belonging LMD products to be able to choose the best method. Algorithm is created in MATLAB and tested in 2 experimental campaigns for both small laser spot and large laser spot. The small laser spot is more precise, and it is used for fine features, while the large laser spot is used for massive components, and it is used to improve the productivity and the deposition efficiency. The AISI 316L is used as selected material. First campaign which is thin wall is used to tune and analyze the algorithm by different layer number specimens. The newly deposited layers reheat the previous deposited ones and effect geometrical properties. To be able to sure that analyze of the algorithm is not affect from these phenomena, track width measurements are taken from the highest layer of each specimens in thin wall experiment. Second campaign which is hollow cylinder is used to validate the algorithm with varying the process parameters and determine the best thresholding method. Thin wall results shows that all the methods work properly with the algorithm and developed algorithm is capable to detect deposition angle as stable for small laser sport and with an exaptable variance for large laser spot. Hollow cylinder results show that developed algorithm is capable to react to different process parameters with a sufficient respond time. The best tested threshold method is observed as multi-Otsu for all the hollow cylinder experiments. The mean averaged error range of multi-Otsu is calculated as 3-10%. Otsu gives the second best results with 3-17% mean averaged error range. Isodata gives the worst results with 4-21% mean averaged error range. It is found that using multi-Otsu method with two thresholds is a successful approach to detect the molten pool geometry.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_07_Aksoy.pdf
non accessibile
Dimensione
17.78 MB
Formato
Adobe PDF
|
17.78 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/10589/190080