Process optimization of metal 3D printing through thermographic analysis of the melt pool

Abstract

Laser powder bed fusion (LPBF) is one of the most industrially used additive manufacturing technologies for metallic components, but the quality of printed parts is highly sensitive to the process parameters and their effect on the melt pool. Since the melt pool geometry and temperature distribution are related to the formation of process induced defects, residual stresses, and microstructure, their reliable in-situ characterization is essential for both process development and quality assurance. Among the available monitoring techniques, thermal cameras offer the most direct access to the melt pool surface, however, single-wavelength systems require prior knowledge of the material's emissivity, a property that cannot be reliably determined by a single factor under the transient, multi-phase conditions of the LPBF process. Even dual-wavelength cameras, which in principle eliminate emissivity dependence through the ratio of spectral radiances at two close wavelengths, rely in practice on calibrations performed using tungsten filament lamps. However, spectral characteristics of such lamps differ significantly from those of a laser-formed metallic melt pool, producing physically inconsistent temperature readings when applied to LPBF conditions. These limitations have so far confined most thermal imaging to qualitative process observation rather than quantitative melt pool characterization. This thesis develops and validates a methodology for quantitative in-situ melt pool monitoring using a dual-wavelength infrared thermal camera calibrated through ex-situ single-track metallography. The dual-wavelength approach eliminates the need for emissivity characterization by utilizing the ratio of spectral radiances measured at two wavelengths (725 and 900 nm), while metallography provides independent melt pool width measurements that serve as both calibration and cross-validation reference. This calibration approach, based on the known liquidus temperatures of five materials (316L, IN625, CoCr, W, and Mo), revealed temperature deviations of up to 450 °C compared to the tungsten filament calibration, confirming that the latter cannot reliably convert radiance emitted into temperature. The complete methodology, combining calibrated thermal imaging with metallographic validation, is initially applied to three alloys (316L, IN625, and CoCr) across a wide range of laser powers and scanning speeds on an EOS M280 system, and further validated on Ti-6Al-4V. The calibrated camera enables emissivity-independent extraction of melt pool width, length, and surface temperature, with cross-validation between camera and metallographic width measurements showing discrepancies of only 6–9% across all materials. Combined with the metallographic melt pool depth data, this provides a complete three-dimensional melt pool dataset used to assess the reliability of an analytical thermal conduction model. The model proved to be reliable for predicting surface dimensions within the conduction melting regime (global R²adj > 0.70 for length, > 0.52 for width) but systematically underestimates melt pool depth in the keyhole regime due to the neglect of recoil pressure and vapor dynamics, with a consistent linear deviation (slope ≈ 0.30) observed across all three materials.

Date	21 Aug 2026
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Alena Kreitcberg (Supervisor) & Vladimir Brailovski (Co-supervisor)