Automation of Property Acquisition of Single Track Depositions Manufactured through Direct Energy Deposition

Abstract

Metallic additive manufacturing processes have been significantly developed since their inception with modern systems capable of manufacturing components for structural applications. However, successful processing through these methods requires extensive experimentation before optimised parameters can be found. In laser-based processes, such as direct energy deposition, it is common for single track beads to be deposited and subjected to analysis, yielding information on how the input parameters influence characteristics such as the output's adhesion to the substrate. These characteristics are often determined using specialised software, from images obtained by cross-section cutting the line beads. The proposed approach was based on a Python algorithm, using the scikit-image library and optical microscopy imaging from produced 18Ni300 Maraging steel on H13 tool steel, and it computes the relevant properties of DED-produced line beads, such as the track height, width, penetration, wettability angles, cross-section areas above and below the substrate and dilution proportion. 18Ni300 Maraging steel depositions were optimised with a laser power of 1550 <mml:semantics>W</mml:semantics>, feeding rate of 12 <mml:semantics>g</mml:semantics> <mml:semantics>min-1</mml:semantics>, scanning speed of 12 <mml:semantics>mm s-1</mml:semantics>, shielding gas flow rate of 25 <mml:semantics>L</mml:semantics> <mml:semantics>min-1</mml:semantics> and carrier gas flow rate of 4 <mml:semantics>L</mml:semantics> <mml:semantics>min-1</mml:semantics> for a laser spot diameter of <mml:semantics>2.1mm</mml:semantics>. Out of the cross-sectioned beads, their respective height, width and penetration were calculated with <mml:semantics>2.71%</mml:semantics>, <mml:semantics>4.01%</mml:semantics> and <mml:semantics>9.35%</mml:semantics> errors; the dilution proportion was computed with <mml:semantics>14.15%</mml:semantics> error, the area above the substrate with <mml:semantics>5.27%</mml:semantics> error and the area below the substrate with <mml:semantics>17.93%</mml:semantics> error. The average computational time for the processing of one image was <mml:semantics>12.7</mml:semantics> <mml:semantics>s</mml:semantics>. The developed approach was purely segmentational and could potentially benefit from machine-learning implementations.