Enhancing image processing in single-camera two-wavelength imaging pyrometry for advanced in-situ melt pool measurement in laser powder bed fusion

Abstract

For in-situ measurement of the melt pool (MP) temperature profile in the laser powder bed fusion (LBPF) additive manufacturing (AM) process, a new technology is the single-camera two-wavelength imaging pyrometry (STWIP). Accurate temporally and spatially resolved MP temperature field measurement using this STWIP method requires a precise profiling of pixel-wise two-wavelength intensity ratio, which is highly dependent on optical alignment, and camera's spectral sensitivity, among other factors. Thus, it is essential to develop an accurate, robust, and fast transformation method for reliable and effective mapping of two-wavelength images acquired from the STWIP system. In this work we propose a Blob analysis-based MP guided Image Transformation (BMPIT) method as opposed to the typical feature detector descriptor-based image transformation approach like KAZE. The BMPIT's performance is assessed and compared with the KAZE in terms of efficiency, execution time, accuracy, and robustness. An experiment using a standard calibrated tungsten filament strip lamp is done to validate the effectiveness of BMPIT. Compared to the KAZE, the BMPIT successfully transformed 100 % of the MP images with higher accuracy and faster speed. It is also shown that the BMPIT is a robust technique for image transformation, unaffected by the image size, MP position, and surrounding noise. Moreover, experimental ground truth data collected using Type C thermocouples implanted into an Inconel-718 build plate are used to further validate the LPBF MP temperature estimation accuracy of BMPIT-aided STWIP. Unlike KAZE, temperature estimated by BMPIT agrees well (error <5 %) with both the lamp and thermocouple experiments. BMPIT is an appealing alternative for online measurement due to its reduced execution time, it takes only one fifth of the time that KAZE takes to transform two-wavelength images. In addition, the BMPIT can be used to calculate MP width, which is validated by comparing with ex-situ characterization. It enables a high level of agreement (with an error less than 1.89 %) between MP images of two wavelengths. Overall, the BMPIT greatly improves STWIP image processing, allowing for measuring MP temperature and morphology more rapidly, accurately, and precisely. The developed BMPIT approach can be employed as part of a STWIP-feedback LPBF process control system to improve the quality of additively manufactured metal products.