Development of a CNN-based real-time monitoring algorithm for additively manufactured molybdenum

Abstract

A convolutional neural network (CNN)-based real-time monitoring algorithm is present to detect an abnormal wire + arc additive manufacturing (WAAM) process for molybdenum. The proposed algorithm consists of three modules: image conversion, CNN prediction, and real-time monitoring. The image conversion module changes the form of a time-series voltage waveform data into voltage image data. The CNN prediction module classifies each voltage image into a normal or abnormal image. The real-time monitoring module expresses the results of the CNN prediction model on a real-time dashboard. Experiments for single beads of molybdenum materials were performed to validate the performance of the proposed algorithm. It was observed that abnormal WAAM processes are detected in real-time with high accuracy. In addition, a sensitivity analysis with respect to different intervals and bandwidths of the voltage image data was conducted, which are the main input parameters of the proposed method. Based on this investigation, guidelines for setting the interval and bandwidth were established. Finally, the effectiveness of the CNN classifiers was validated by applying a class-activation mapping method. It was concluded that the CNN classifiers were adequately trained because they captured the critical regions in the voltage images for both normal and abnormal cases.