A fast method for monitoring molten pool in infrared image streams using gravitational superpixels.

Additive manufacturing (AM) is one of the most trending areas in production that allows creating three-dimensional objects according to a predetermined design. AM finds its application in all kinds of niches, from medicine to the aerospace industry, although there are still several technological barriers that must be addressed. For example, monitoring techniques that guarantee decision-making to guarantee quality and repeatability of processes. An imaging-based methodology is presented to monitor and extract thermal and geometric characteristics of molten pool in real-time. A superpixel-based approach is proposed to reduce the dimensionality of the infrared images and facilitate the segmentation and tracking tasks. This algorithm is called gravitational superpixels. Using the color and temperature features, our algorithm groups the pixels. These superpixels have better adherence to the structures that form the images. Facilitating the segmentation tasks. Our algorithm is compared against superpixel-based and saliency-based already reported works. To validate the performance, infrared-image streams (LMD process) and standard datasets are using. The proposed algorithm has a molten pool segmentation uncertainty of $$0.1\; mm$$ . Reported results show that the performance of our proposal is applicable for tasks that require good precision when segmenting and fast runtime. It is important to highlight the relevance of this work for additive metal manufacturing processes.