Image Synthesis Using Conditional GANs for Selective Laser Melting Additive Manufacturing

In-situ process monitoring for metals additive manufacturing is paramount to the successful build of an object for application in extreme or high stress environments. Yet in selective laser melting additive manufacturing, it is extremely difficult to evaluate the build process. The difficulty is that obtaining enough variety of data to quantify the internal microstructures for the evaluation of its physical properties is problematic, as the laser passes at high speeds over powder grains at a micrometer scale. Using generative models, a type of machine learning, has been shown here to provide new artificially generated data with the same properties as the experimental images. The Generative Adversarial Network (GAN) synthesized new computationally derived data through a process that learns the underlying features of images that correspond to the different laser process parameters in a generator network. While this technique was effective at delivering high-quality images that closely matched the training data when tested against holdout samples, modifications to the general form of the network through a conditional generative adversarial network (CGAN) showed improved capabilities at creating these new images. Using multiple evaluation metrics, it has been shown that generative models can be used to create new data for various laser process parameter combinations, thereby allowing a more comprehensive evaluation of ideal laser conditions for any particular build. The new data can supplement the experimental data, thereby growing the overall knowledge framework for build characteristics.