研究目的
To detect shapes with IR imagery via CNN and assess part quality in LPBF technologies.
研究成果
The implementation of CNN for detecting shapes with IR imagery was a success, but more work is needed to achieve the main goal of detecting defects and shapes. A larger and more diverse training set would improve the testing and validation set.
研究不足
The training set needs to be larger and more diverse for better results. The pre-processing method must be reviewed further to ensure the output is most appropriate for the neural network.
1：Experimental Design and Method Selection:
The research utilized a CAD designed part with specific geometries for feature detection using CNN on IR images from an SLM machine.
2：Sample Selection and Data Sources:
IR images were collected from an open architecture SLM machine during the build process.
3：List of Experimental Equipment and Materials:
An open architecture SLM machine and a Stratonics Global Heat Monitor infrared camera were used.
4：Experimental Procedures and Operational Workflow:
IR data was pre-processed using PCA and Laplacian filtering before being input into a CNN for feature detection.
5：Data Analysis Methods:
CNN was used to classify and validate geometric elements against the build file.
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
