<i>In-situ</i> monitoring of laser powder bed fusion energy input based on a hybrid feature pyramid network and multi-sensor fusion

Abstract One of the core challenges in laser powder bed fusion (LPBF) technology is to achieve quality feedback control during the printing process, and the development of a dependable in-situ monitoring system is crucial to attaining this goal. In this paper, a multi-sensor feature fusion network model (HFF-net) based on a hybrid feature pyramid network (FPN) architecture is proposed to utilize multi-source sensing signals to achieve energy input state monitoring of the LPBF process. In this study, the signals from the LPBF process were initially captured using a photodiode and a microphone, the collected signals were then classified into three categories based on the differences in part porosity. Subsequently, preprocessing operations were conducted on the two collected signals and the matching of the different signals was achieved to generate the model training data. Following this, four innovative network models were developed, drawing inspiration from the ConvNeXt baseline model. In the proposed model, the HFF-net model achieved a recognition accuracy of 99.45% for the energy input state with a recognition time of 0.33 ms, demonstrating the superiority of the hybrid FPN architecture. Finally, when compared to other classical baseline models, the proposed HFF-net model framework showed a competitive advantage in both single-sensor and multi-sensor fusion monitoring. These results underscore the effectiveness of the feature fusion model proposed in this paper for conducting in-situ monitoring of energy input states in LPBF process, thereby providing a viable approach for quality feedback control.