Convolutional Neural Network Based On-Line Defect Recognition in Additive Manufacturing Using Image Processing

For the Selective Laser Melting (SLM) of metal powders, it is important to design a method of machine learning for online issue identification that makes use of automated image processing. To allow the earliest possible discovery and rectification of faults in the material that originate from non-conformities in the manufacturing process. When analyzing the in-process pictures that are taken while the layer-by-layer SLM processing is taking place, bi-stream deep convolutional neural networks are used as the primary tool. It is conceivable, via the use of automated feature learning and feature fusion, to discover patterns that are related with SLM -unfavorable conditions. The significance of the machine learning approach for the online detection of defects that are the consequence of process non-conformities has been shown and validated via the use of experimental testing and analysis. This lays the foundation for the component quality assurance and adaptive SLM process management that will follow. When compared to the data sets that were used for training or validation, CNNs demonstrated improved generalizability. Furthermore, it was shown that the size of the defect has an effect on the accuracy of CNN, which raises the prospect that very little errors may be made with the use of procedures that do not modify the surface morphology of the build plate. These findings point to a connection between the beginning of anomalies, such as a severe absence of fusion, and the presence of process ejection in the body.