Near Infrared Thermal Imaging for Process Monitoring in Additive Manufacturing

This work presents the design and development of a near infrared thermal imaging system
specifically designed for process monitoring of additive manufacturing. The overall aims of
the work were to use in situ thermal imaging to develop methods for monitoring process
parameters of additive manufacturing processes. The main motivations are the recent
growth in use of additive manufacturing and the underutilisation of near infrared camera
technology in thermal imaging. The combination of these two technologies presents
opportunities for unique process monitoring methods which are demonstrated here.
A thermal imaging system was designed for monitoring the electron beam melting process
of an Arcam S12. With this system a new method of dynamic emissivity correction based
on tracking the melted material is shown. This allows for the automatic application of
emissivity values to previously melted areas of a layer image. This reduces the potential
temperature error in the thermal image caused by incorrect emissivity values or the
assumption of a single value for a whole image. Methods for determining materials
properties such as porosity and tensile strength from the in situ thermal imaging are also
shown. This kind of analysis from in situ images is the groundwork for allowing part
properties to be tuned at build time and could remove the need for post build testing that
would determine if it is suitable for use.
The system was also used to image electron beam welding and gas tungsten arc welding.
With the electron beam welding of dissimilar metals, the thermal images were able to
show the preheating effect that the melt pool had on the materials, the suspected reason
for the process’s success. For the gas tungsten arc welding process analysis methods that
would predict weld quality were developed, with the aim of later integrating these into the
robotic welding process. Methods for detecting the freezing point of the weld bead and
tracking slag spots were developed, both of which could be used as indicators of weld
quality or defects. A machine learning algorithm was also applied to images of pipe
welding on this process. The aim of this was to develop an image segmentation algorithm
that could be used to measure parts of the weld in process and inform other analysis, like
those above.