top

知識分享

Study on porosity prediction on L-PBF based upon machine learning

建立日期：2021/12/24
  • 作者： 成大機械工程系/Ze-Hong Lin, Trong-Nhan Le, Yu-Lung Lo* and Hong-Chung Tran
  • 出處： 2021 AOI論壇與展覽
  • 內容： 本篇為「2021 AOI論壇與展覽」論文集，摘要如下：

    The primary factors that influence the porosity of the 3D parts by L-PBF process are the variations of processing parameters (the laser parameters, the powder layer thickness, the surface roughness of the previous layer) due to system instability and the stochastic nature of L-PBF process. One of the most effective methods to control the quality of the fabricated part is the defect detection via a combination of an in-situ monitoring system and machine learning techniques. This study proposed an Artificial Intelligent-based approach utilizing the melt-pool shape and spatter angle features to predict the formation of pores during L-PBF process. The core component of the monitoring module is the high-speed and high-resolution CMOS camera and a set of in-house algorithms for melt-pool shape and spatter angle detections. For the pore-prediction module, self-organizing map (SOM) method was deployed to explore the correlation between melt-pool shape and spatter angle features and the possibility of pore formation.
    The final predicted outputs of the proposed system are the location and size of the porosity inside the built part. The prediction results are assessed using micro-CT results to determine the accuracy of the proposed method. As a result, the accuracy in pore position predication is 79.17 % and that in pore volume predication is 97.42%. As our best knowledge, this is the first attempt in the field to use both the melt-pool shape and spatter angle features in one unified framework for porosity prediction.
  • 檔案下載：全文下載
  • 下載次數：9
回上頁