What if a simple camera image could stand in for a difficult temperature measurement during metal 3D printing? This study finds that the melt pool’s length tracks temperature better than its area. Directed energy deposition using laser beam and wire feed is a critical technology in high-productivity metal additive manufacturing, enabling precise deposition and fabrication of components with high quality and deposition rates.

The interaction between the laser beam and metal is highly temperature-sensitive, affecting metal transfer, melting, solidification, and phase transformations. Effective control of heat input is crucial for influencing the geometry, microstructure, and overall integrity of the fabricated parts.

This study investigates the correlation between the melt pool top surface boundary contour and the immediate solidification temperature, utilizing machine vision and radiation pyrometer measurements. The research highlights the significance of melt pool length as an indicator of temperature variations during deposition, and system identification techniques were employed to develop a predictive model of the dynamic relationship between melt pool length and temperature.

The approach substitutes off-axis pyrometry with a cost-effective coaxial vision camera integrated into the laser head, while comparative analysis demonstrates the superiority of length-based correlations with temperature data. Metal additive manufacturing technologies, particularly directed energy deposition with laser beam and wire, enable the direct manufacturing of components from digital designs.

Beyond design advantages, directed energy deposition with laser beam and wire minimizes material waste, enhances manufacturing control, and finds applications in industries requiring lightweight and precisely crafted components. Directed energy deposition with laser beam and wire introduces challenges in maintaining control over process conditions, particularly as deposition is used for complex geometries.

The management of multiple variables during deposition requires thorough understanding and exact control to provide consistent and high-quality results. In industrial settings, the number of feasible sensor systems is limited, calling for more compact and affordable solutions.

Heat-input control, specifically management of heat accumulation, is a significant challenge in directed energy deposition, leading to research on temperature measurement and heat-input regulation. Earlier work analyzed melt pool top surface area changes and their interactions with process parameters to support closed-loop control.

That work was developed further for real-time melt pool surface area control using infrared-camera frames and modulation of laser power, traverse speed, and feedstock wire speed. All examined control modes resulted in reduced energy density, but the effects of those settings on heating, cooling, and final product quality still need more exploration.

Figure one lays out the directed-energy-deposition cell from three perspectives: the six-axis robot and argon shield-gas tent, the laser head with integrated camera and pyrometer, and a schematic of their optical paths. The diagram shows laser light focused onto the workpiece, while vision cameras and a two-color pyrometer observe the deposition region.

This setup matters because it combines processing and in-situ sensing in the same tool, enabling monitoring of the melt area during deposition. Radiation pyrometers in directed energy deposition can be direction-dependent, prone to errors, noisy, and affected by uncertainty in target emissivity estimation; mounting and positioning them can also be complex and cumbersome.

This study replaces off-axis pyrometer measurements with a cost-effective coaxial vision camera integrated into the laser head, enabling melt pool length measurement with minimal dependence on deposition direction and reduced risk of field-of-view obstruction. The work systematically quantifies the correlation between melt pool geometry and immediate solidification temperature measured at the tail of the melt pool.

The present work uses a compact, calibration-free coaxial vision camera to measure melt pool length and quantitatively correlates that length with immediate solidification temperature in industrial-scale, continuous, multi-layer builds under realistic heat accumulation conditions.

A dedicated image-processing method removes oxide islands and prior-layer effects, enabling robust and repeatable correlation measurement. Depositions were performed in an in-house directed energy deposition with laser beam and wire cell, with the laser head mounted on a six-axis ABB robot.

The depositions took place in an argon-filled shielding gas chamber to prevent oxidation. The laser used a five-millimeter spot size with a top-hat power density distribution and a wavelength of one thousand thirty nanometers.

The laser head included a wire feeder and a hot-wire system that preheated the feedstock wire using electrical power, while an adaptive height controller maintained the vertical wire-entry distance relative to the molten pool. A spectrometer analyzed emissions from the laser optics, and the data showed a significant peak around one thousand four hundred nanometers.

Three cameras were tested, including a short-wave infrared camera with a one-thousand-four-hundred-nanometer bandpass filter and a Photonfocus camera selected for near-infrared imaging with a seven-hundred-fifty-nanometer bandpass filter. All cameras were installed coaxially with the laser head to capture a top view of the melt pool, because complicated geometries limit the field of view and off-axis installation is difficult in industrial applications.

The Photonfocus camera captured the whole melt pool region and a clear view of the wire tip without active illumination, so it was selected for the following experiments. The feedstock was duplex stainless steel 2209 wire with a diameter of one point two millimeters.

The chosen geometry was a cylindrical single-wall structure, two hundred millimeters in diameter, with a bead height of zero point nine millimeters and a thickness of five millimeters. The geometry was chosen to illustrate heat accumulation during deposition of multiple layers, and heat began to accumulate beyond a few layers because of inadequate heat dissipation.

Figure five lays out the image-processing pipeline, beginning with the camera image, I, and cropping it to a region of interest. The frames are then grouped and averaged, followed by an absolute-difference step, subtraction to suppress oxide artifacts, and thresholding to produce the binarized melt-pool geometry, Ibin.

This matters because the resulting binary image supports contour extraction and measurement of the melt pool’s length and area. The image-processing techniques were used to measure melt pool dimensions and analyze the camera and pyrometer signals. Camera frames were cropped first to remove irrelevant information and limit the number of pixels processed in later steps.

The cropped images were grouped, and an element-wise mean was applied to produce the grouped image mean. Absolute difference measurement and subtraction followed a thresholding technique, producing the binarized image as the outcome of the algorithm.

The method was not designed to capture transient melt pool dynamics at the beginning and end of deposition, so image processing was applied to images showing a fully developed melt pool. Images from the initial stages were excluded because the melt pool did not yet have a quasi-steady phase.

Temperature recording was aligned with image timestamps, and temperature measurements were resampled to match the camera’s higher recording frequency. For comparison, the desired number of samples was set at sixteen thousand two hundred sixty-two, and MATLAB resampling functions adjusted the pyrometer array’s sampling rate.

A single-input, single-output predictive model was constructed to capture the dynamics between melt pool length as the input and temperature as the output. Using the MATLAB System Identification Toolbox, the analysis covered linear and nonlinear dynamic models, including first-order and second-order linear models.

Figure ten tracks normalized temperature, melt pool length, area, laser power, and layer number during nine deposition layers, from layers ten through nineteen. Three pronounced responses appear around one point three, two point six, and four point two minutes, following step changes in laser power, with abrupt declines and later recovery.

The raw signals are visibly noisy, while filtered temperature and length clarify the underlying responses and the effects of heat accumulation across layers. The graph shows three pronounced spikes at one point three minutes, two point six minutes, and four point two minutes, caused by step changes in laser power; each spike is followed by an abrupt decline and then a growth phase.

With the addition of more layers, both melt pool length and measured temperature tend to increase. A reduction in laser power significantly impacts regulation of both melt pool length and measured temperature. Laser power modulation therefore acts as a tool for controlling and optimizing melt pool and temperature dynamics in directed energy deposition with laser beam and wire.

The central result is that melt pool length provides a strong temperature indicator, and a dynamic model reached an eighty-five point one percent fit. That could support coaxial, closed-loop heat control without a pyrometer.