基于视觉传感的热丝激光金属沉积熔滴—熔池多特征信息同步监测

Feature information extraction of hot wire laser metal deposition process based on object and key point detection

为了提高热丝激光金属沉积(HW-LMD)过程中的质量稳定性和实现熔滴—熔池多特征信息的同步实时监测,采用基于高动态视觉相机结合YOLO v8深度学习神经网络的高精度实时监控方法,通过相机捕捉HW-LMD过程中的动态变化,并利用YOLO v8神经网络对过渡方式和熔池行为进行同步监测,首先判断沉积过程是否为稳定的液桥过渡,然后在液桥过渡模式下提取熔池尺寸的关键点信息.结果表明,YOLO v8神经网络在检测沉积过程过渡方式和熔池关键点信息方面具有高精确度,精确率分别达到了98.8%和99.9%,熔池宽度的平均误差为4.1%,且推理时间平均仅为12 ms/帧,满足了HWLMD过程实时监控的需求.

In order to improve the quality stability and achieve simultaneous real-time monitoring of droplet-melting pool multifeature information during the hot-wire laser metal deposition(HW-LMD)process,this study adopts a high-precision real-time monitoring method based on a high-dynamic vision camera combined with a YOLO v8 deep learning neural network.The dynamic changes in the HW-LMD process are captured by a camera,and the transition mode and melt pool behaviour are monitored synchronously using the YOLO v8 neural network,which firstly determines whether the deposition process is a stable liquid bridge transition or not,and then extracts the key point information of the melt pool dimensions in the liquid bridge transition mode.The results show that the YOLO v8 neural network has high accuracy in detecting the transition mode and melt pool key point information of the deposition process,with an accuracy rate of 98.8%and 99.9%,respectively,and an average error of 4.1%for the melt pool width,and the inference time is only 12 ms per frame on average,which meets the demand for real-time monitoring of the HW-LMD process.