Quality monitoring in wire-arc additive manufacturing based on cooperative awareness of spectrum and vision

Abstract

This paper presents a multi-source classification method based on cooperative awareness method of spectrum, vision and electrical parameter for the quality monitoring in wire-arc additive manufacturing. Triggered by the field programmable gate array (FPGA), the spectrum was collected in the peak current, and a weld pool image was captured in the base current. In this way, we acquired the multi-time information about both the spectrum with abundant information and the weld pool image with low interference within one welding current period, and achieved the cooperative awareness. We proposed a k-nearest neighbor (KNN) classification algorithm based on contour curve feature (CC-KNN) in vision and two classification methods -priori threshold and KNN based on locality preserving projection (LPP-KNN) -in spectral analysis. The combination of vision and spectrum can simultaneously monitor the unusual states of process parameters and quality defects. Our method is not limited to one welding process, and experimental results of three wire materials in cold metal transfer (CMT) welding have verified the superiority of our method on the number of monitoring objects, accuracy and stability.