Development of Melt-pool Monitoring System based on Degree of Irregularity for Defect Diagnosis of Directed Energy Deposition Process
Jun Ho Kong, Sang Won Lee
Abstract
In this paper, a novel deep learning-based defect diagnosis system for the directed energy deposition (DED) process with melt-pool monitoring has been developed by proposing a new feature index, the degree of irregularity (DOI), which represents a degree of geometrical deviation of the melt-pool from its circular shape. During the DED process, time series data of the DOI are collected in real-time by the melt-pool monitoring module, and they are processed through the Hilbert-Huang transform (HHT) with ensemble empirical mode decomposition (EEMD). The processed features are input to the 2D convolutional neural networks (CNN) in order to diagnose the defect, more specifically, melting balling, in this paper. The final diagnostic accuracies for the normal and balling states were 76.65% and 96.92%, respectively, and they are about 10 to 15% better than those from the conventional short-time Fourier transform (STFT)-based 2D-CNN model. Thus, the proposed feature index (DOI) and associated processing approach (HHT with EEMD) have been validated for effective defect diagnosis in the DED process.