Enhanced in-situ monitoring of metal deposition behaviour for pulsed wire arc directed energy deposition using integrated noncoaxial imaging and supervised deep learning framework

Computer vision is increasingly used in manufacturing to ensure high-quality products and reduce cost. In Directed Energy Deposition-Arc (DED-Arc) processes, challenges like dimensional inaccuracy, layered morphology, and metallurgical flaws require online monitoring. This paper presents a method to predict arc and melt pool areas during layer deposition using computer vision. Pareto optimization is applied to analyze how melt pool area changes with arc area. The strong and weak arc classification has been determined using Inception V3. The two different variations of arc and melt pool area in a pulsed DED-Arc process are considered by calculating the mean of both the arc/ melt pool area in each of the single pulse cycles. The segmentation accuracies of the arc and melt pool area are found as more than 95 percent, which is validated using the proposed Convolution Neural Network(CNN) architecture in this research.