Unraveling the potential of multi-sensor fusion of acoustic signals and melt pool geometric images towards defect identification in additive manufacturing

With its advantages of high efficiency and low cost, directed energy deposition-arc (DED-arc) has shown great potential in aerospace, automotive manufacturing and other industrial fields. However, the presence of multiple defects poses a considerable challenge to the quality of the formed parts, limiting the further development of the DED-arc in industry. This study proposes an in-situ monitoring system for defects identification based on multi-sensor fusion at the data and decision levels. Three different yet complementary signals were acquired for accurate and timely monitoring of the forming geometry to avoid discontinuity and surface pore defects in DED-arc. In this study, only discontinuity and pore defects were identified to verify the validity of the method and to provide a basis for identifying other complex defects in the future. The acoustic signals carried the unique acoustic characteristics of the manufacturing process, whilst the melt pool width images showed the lateral expansion of the molten material, and the height images revealed the vertical dimensions and dynamics of the melt pool. A detailed analysis of the defect signals revealed that discontinuity defects are most distinguishable from normal and surface pore defects, but distinguishing between normal and pore defects is difficult. The study systematically compared the performance of single-sensor, dual-sensor, and three-sensor defect monitoring models via data-level fusion and decision-level fusion strategies. The findings indicate that the multi-sensor fusion model has been shown to outperform the single-sensor model in both data-level fusion and decision-level fusion. The decision-level fusion effect exhibits superiority over the data-level fusion effect in the study, which have accuracy of 100 % and 99.55 %, respectively. The comparative analysis of these two fusion strategies yields critical insights, demonstrating that the multi-sensor fusion approach enhances the accuracy of defect identification in DED-arc manufacturing.