A monitoring method for local defects in laser additive manufacturing process based on molten pool spatiotemporal information fusion

Abstract

Online monitoring is a key technology for improving the quality of laser additive manufacturing (AM). However, current online monitoring techniques primarily focus on the transient spatial features of process state information and insufficiently account for the spatiotemporal features contained in the evolution of the molten pool in the layer-by-layer deposition process of laser AM. To address this problem, this paper proposes an online monitoring method based on molten pool spatiotemporal information fusion to predict local defects in the laser AM process. We utilized a coaxially integrated Charge-Coupled Device (CCD) camera to capture the molten pool information throughout the printing process. Based on the spatiotemporal correspondence between these molten pool images and local defects, we constructed an experimental dataset. In addition, considering the physical process of layer-by-layer deposition, we propose a spatiotemporal fusion neural network (STFNN) to establish a mapping relationship between the spatiotemporal information contained in the molten pool image sequences and local defects. A temporal information extraction module is designed to capture the spatiotemporal characteristics contained in molten pool images within the same layer and across different layers during the process. Concurrently, a spatial information extraction module is introduced to extract transient spatial features from process images, and a feature fusion module is implemented to integrate high-level features. Compared to methods that extract transient spatial features from the molten pool image, the STFNN model exhibits a significant improvement in defect prediction accuracy. Furthermore, experimental results show that the monitoring method considering both intra-layer and inter-layer spatiotemporal information contained in the molten pool has better porosity detection than those considering only intra-layer or inter-layer spatiotemporal features.