View source for AM Tomography Image Processing Algorithm

<strong>Image processing module for extracting the melt pool area and center coordinates from the additive manufacturing tomography. </strong> [[File:Manufacturing tomography.png|thumb|right|<div style="font-size:88%;line-height: 1.5em">Image 1: Outputs of the image processing module.</div>]]

== Asset Description ==
<p style="line-height: 1.5em">
Processing algorithm that automatically calculates the melt pool area and the melt pool center coordinates, extracting these values from the corresponding melt pool tomography image of the additive manufacturing process.
</p>

== Asset Details ==
=== Dataset Information ===
<p style="line-height: 1.5em">
The implementation of this algorithm is framed within the execution of an experiment that focuses on the provision of real-time monitoring through AI at the Edge in an additive manufacturing (AM) process for early defect detection. In particular, geometrical deformations in metal pieces that are manufactured.</p>

<p style="line-height: 1.5em">
The main reason why defects occur in this particular manufacturing process is the accumulation of the heat applied as layers on the metal piece increase. For this reason, besides the analysis and monitoring of parameters such as the laser power, position angles, position coordinates, etc., the analysis of the melt pool tomography image of the AM process, together with the evolution of the z coordinate, is critical. In particular, the melt pool area and the melt pool center offset are highly relevant variables. Therefore, in order to apply tabular-data AI/ML models to implement this kind of anomaly detection in the manufacturing process, it was required to extract from that tomography image the corresponding melt pool area and melt pool centroid coordinates.</p>

Regarding the technical implementation of this algorithm, it must be taken into account that the format expected for the images to be processed is bytearray, since they are collected from an OPC-UA server. Due to this, before processing the information retrieved from the OPC-UA server, the images need to be converted to a format such that the OpenCV library can detect and extract the right mask from it. Currently this is performed by using the Image library, reading the image in RGBA and converting it to a numpy array afterwards.

Described in a simplified form, the image processing algorithm follows the following steps:
1.	Conversion of the image to grayscale.
2.	Enhancement of the area definition by using a GaussianBlur function.
3.	Detection of the ROI (region of interest) by defining the limit values from which each pixel will be considered as non-background.
4.	Extraction of the area by checking the number of available white pixels. This is made by adding all the values, as the image only has two values: black (0) and white (255). After this, the obtained result is divided by 255 so the result is in pixels.
5.	Finally, the centroid of the melt pool shape is computed, obtaining the Cx and Cy coordinates.
</p>

== Usage ==
<p style="line-height: 1.5em">
The usage of this tool is detailed in the corresponding GitHub repository. However, for now the license applicable to this asset is Proprietary, so the code has not been publicly published.

The algorithm is prepared to be deployed as a Docker container, reading the input image data from a RabbitMQ queue and publishing the melt pool area and melt pool center coordinates in another.

</p>

=== Maturity===
<p style="line-height: 1.5em">
PoC ready, Ongoing Development </p>

=== Licence===
Proprietary

== Resources ==
<ul>
<li>For further information, please contact with mmarquez@gradiant.org.</li>
<li>Provided by [https://gradiant.org/en/ GRADIANT]</li>
</ul>

==== Acknowledgement====
<p style="font-size:90%;line-height: 1.5em">
''The algorithm was created in the framework of the AI REDGIO 5.0 project. It will be used in the DF XIV experiment (AI at the Edge for real-time monitoring of an additive manufacturing cell), which is being developed by Gradiant''
</p>

== Relevant Categories==
[[Category:Jupyter Notebook]][[Category:Docker Container]][[Category:Quality Control]][[Category:Edge AI]]