Eventos Anais de eventos
COBEF 2023
12th Brazilian Congress on Manufacturing Engineering
A COMPREHENSIVE EVALUATION OF THE LASER POWER AND POWDER FEED RATE FOR THE DIRECTED ENERGY DEPOSITION PROCESS USING PRINCIPAL COMPONENTS ANALYSIS
Submission Author:
Vincent Wong , SP
Co-Authors:
Vincent Wong, Gustavo Jose Giardini Lahr, Glauco Caurin, Alessandro Rodrigues, Reginaldo Coelho
Presenter: Vincent Wong
doi://10.26678/ABCM.COBEF2023.COF23-0380
Abstract
Laser Powder Directed Energy Deposition (LP-DED) is a metal manufacturing technique that employs a high-powered laser and powder distribution system. The process begins with the powder transport using an inert gas from the powder distribution system to a coaxial nozzle. Then the powder leaves the coaxial nozzle, and the laser action allows it to melt powder in a substrate to create lines and up layer by layer until creating the final product. The final product's quality heavily depends on the process parameters, which must be adjusted based on the morphology and particle size distribution of each material. These parameters are closely linked to phenomena that occur during the deposition process, such as growth rate, fluid flow in the molten pool, and thermal gradient. Ultimately, these factors determine the mechanical behavior and microstructural features of the final product. One of the challenges observed in the literature is the number of parameters and phenomena involved. These aspects create a multivariate dataset that is difficult to understand from an analytical and experimental point of view. This paper aims to analyze the influence of laser power and powder feed rate on bead geometry using exploratory data analysis and Principal Components Analysis (PCA) to reduce the dataset's dimensionality. To evaluate the deposition quality, we create 49 Single Scan Tracks (SSTs) and then extract four main features: bead height, substrate penetration depth, width, and penetration depth area. The dataset analyzed comprises 194 cross-sections obtained from SSTs containing laser power, powder feed rate, global density energy, line mass, heat input, powder capture efficiency, and dimensional features of cross sections from AISI 316L Single Scan Tracks. Matrix correlation was performed to investigate the relationship of each variable, followed by principal components analysis. The proportion of the total variance showed that two components were sufficient to explain >90% of the dataset. The Random Forest (RF) algorithm was then used to estimate SST quality. PCA improved the model's performance, increasing accuracy from 87% using the original dataset to 91% when principal components were used. This research showed that the geometrical features of SSTs are highly dependent on process parameters. The principal component analysis enabled dimensional reduction to construct relevant features through a linear combination of the original features, improving the dataset's quality and making this approach a useful tool to enhance the performance of machine learning algorithms.
Keywords
Laser Powder Directed energy deposition (LP-DED), Principal component analysis (PCA), AISI 316L, Process parameters influence
