Laser Additive Manufacturing (LAM) faces various technical challenges, encompassing issues with dimensional accuracy, mechanical properties, and processing-related defects, such as surface roughness, cracking, and residual porosity. Currently, the quality of safety-critical parts produced by LAM processes, which is still improved through trial-and-error adjustments of multiple process variables, have limited real-time monitoring capabilities. This study introduces an innovative method for real-time quality assessment of laser-directed energy deposition (LDED) procedures across multiple scales (macroscale, mesoscale, and microscale) utilizing a single sensor. This approach is founded on two fundamental components: the single-sensor system for data acquisition and the algorithm for simultaneous multi-scale quality monitoring. For the data acquisition system aspect, a single-sensor, high dynamic range (HDR), multi-scale information acquisition system has been developed to mitigate intense radiation, eliminate plume disturbances, and rectifies backlight shadows, facilitating clearer images of deposition contours and the molten pool region. In the realm of data monitoring algorithms, a physics model-driven supervisory algorithm is employed to enable monitoring of contour height instability, while a convolutional neural network (CNN), driven by image data, is utilized for porosity prediction. Then, this algorithm adeptly tackles the challenge of simultaneously monitoring macroscopic contour and mesoscale porosity. Regarding the validation of the method, the outcomes of single-sensor multi-scale quality monitoring in additive manufacturing of ceramic thin-wall parts indicate a 100% recognition rate for deposition contour and molten pools, contour feature recognition with a relative error of under 0.05%, porosity prediction accuracy surpassing 99%, and a monitoring time of 80.4 ms per frame. Single-sensor multi-scale real-time quality monitoring can serve as a methodological support for future real-time quality control in LDED.
Read full abstract