Abstract
Selective laser melting (SLM) is modernizing the production of highly complex metal parts across the manufacturing industry. However, achieving material homogeneity and controlling thermal deformation remain major challenges for metal-based, additive manufacturing. Therefore, adequate control systems are needed to monitor build processes, and ensure part quality throughout production. Traditionally, control designs relied on physics-based knowledge in analyzing, characterizing, and modeling complex, nonstationary patterns. Recent advancements in machine learning techniques harness the abundance of data to discover effective control designs. In this paper, we investigate the efficacy of a data-driven approach towards in-situ modeling of melt-pool geometry. Specifically, we propose a new methodology that uses a deep neural network architecture to predict melt pool geometries with linear regression models, which manifest during in-situ processes. Experimental results show that our deep neural network model with multiple input features produced 84% goodness of fit score, outperforming the model with a single feature that scored 37% for the given dataset, and the monitored regression models. These outcomes promote further investigation into new and efficient ways for acquiring real-time data from in-situ processes. Our contribution complements the way we understand properties of in-situ data, and predict patterns of melt pools, based on artificial cognition.
A New Lightweight In-situ Adversarial Sample Detector for Edge Deep Neural Network
