DBN-based online identification of porosity regions during laser welding of aluminum alloys using coherent optical diagnosis

Abstract

As a typical serious defect during laser welding of aluminum alloys, porosity can significantly impair the performance of the weld seam. Real-time monitoring of porosity defect can provide guidance for feedback adjustment of welding quality to prevent the subsequent continued formation of porosity, which has attracted increasing attention. This paper implements the deep learning-based online identification of porosity regions using coherent optical diagnosis. Considering the formation mechanism of porosity, a monitoring platform including a coherent optical sensor is established to measure the keyhole 3D morphology and extract the keyhole depth as the coherent optical signal. The acquired keyhole depth signal is processed by ensemble empirical mode decomposition, and it is discovered that there is a segmentation threshold in the reconstructed signal to separate the porosity regions and the non-porosity regions. A deep belief network is constructed to complete the reception of the reconstructed keyhole depth signal and the response of the prediction state, realizing the online identification of the porosity regions. The results show that the identification model has high accuracy, and the validity verification demonstrates the reliability of the proposed method.