研究目的
To predict a laser power value from a melt-pool image in laser powder bed fusion using a convolutional neural network.
研究成果
The proposed CNN (CNN-F7K3-Sig) showed high inference success rates and was effective in predicting laser power values from melt-pool images, even for states not seen in training. It is expected to be useful for identifying problematic positions in additive-manufactured layers without destructive tests.
研究不足
The study is limited by the quality of melt-pool images acquired and the need to distinguish useful from useless images in a preprocessing step.
1：Experimental Design and Method Selection:
The study used a convolutional neural network (CNN) to predict laser power values from melt-pool images. The CNN was optimally configured by grid search of hyper-parameters.
2：Sample Selection and Data Sources:
199,473 melt-pool images were acquired from a monitoring system inside metal additive manufacturing equipment. Images were labeled with laser power values.
3：List of Experimental Equipment and Materials:
High-speed camera inside metal AM equipment, laser beam of 1075 nm wavelength, stainless steel (SUS316L) powders.
4：Experimental Procedures and Operational Workflow:
Images were clipped into 60 × 60 pixels to remove useless black areas. Summing up pixel intensities and calculating histograms were used to filter garbage images.
5：Data Analysis Methods:
Root mean squared error (RMSE) and the coefficient of determination (R2) were used to evaluate the performance of the CNN.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
