Low-latency network-based fast melt pool state recognition method.

Area of Science:

• Materials Science
• Manufacturing Engineering
• Computer Vision

Background:

• Real-time melt pool state recognition is critical for high-quality metal additive manufacturing (MAM).
• Existing deep learning methods for melt pool recognition exhibit high latency, hindering real-time monitoring.
• There is a need for faster and more accurate melt pool state identification in MAM.

Purpose of the Study:

• To develop a low-latency, lightweight convolutional neural network (CNN) for rapid melt pool state recognition.
• To improve the speed and accuracy of melt pool monitoring in laser melting deposition (LMD) processes.
• To enable real-time quality control in additive manufacturing.

Main Methods:

• Designed a high-speed fundamental network unit utilizing a selective convolution operator to enhance spatial feature extraction.
• Developed a lightweight CNN architecture incorporating the designed network unit for efficient melt pool image analysis.
• Collected a dataset of melt pool states using a custom laser melting deposition additive manufacturing system.

Main Results:

• The proposed low-latency network achieved a high melt pool state recognition accuracy of 98.50%.
• The method demonstrated a significantly reduced execution time of only 5.26 ms.
• Experimental results confirm the superiority of the proposed approach over existing methods.

Conclusions:

• The developed lightweight CNN offers a superior solution for real-time melt pool state recognition in MAM.
• The method's low latency and high accuracy are crucial for effective real-time monitoring and quality assurance.
• This advancement contributes to the development of more robust and efficient additive manufacturing processes.