一种基于PPCNN的金属激光熔化沉积熔池状态识别方法

A Method for Identifying Molten Pool State of Laser-based Direct Energy Deposition Based on PPCNN

金属激光熔化沉积质量与熔池状态具有密切的关联,根据熔池视觉特征对加工中的熔池状态进行识别,进而实现沉积质量的在线预测对金属激光熔化沉积技术具有重要意义。为构建上述映射关系,本文提出了一种基于金字塔池化卷积神经网络的金属激光熔化沉积熔池状态识别方法。针对所采集的熔池同轴图像,建立用于训练和测试的数据集;构建了金字塔池化卷积神经网络(Pyramid pooling convolutional neural networks,PPCNN),并进行网络关键参数的研究。结果表明:第一层卷积核尺寸为5×5,卷积层和金字塔池化模块含有64+8×3个卷积核使网络在识别准确率上达到最佳。所提方法取得了最高96.80%的识别准确率。

The quality of metal laser melting deposition is closely related to the state of the molten pool.Recognizing the state of the molten pool in processing according to the visual characteristics of the molten pool,and thus realizing the online prediction of the deposition quality is of great significance to the metal laser melting deposition technology.In order to construct the above mapping relationship,this paper proposes a method for identifying the molten pool state of metal laser melting deposition based on pyramid pooling convolutional neural network.Based on the collected coaxial images of the molten pool,a data set for training and testing was established;a pyramid pooling convolutional neural network(PPCNN)was constructed,and key network parameters were studied.The results show that the size of the first layer of convolution kernels is 5×5,and the convolutional layer and pyramid pooling module contain 64+8×3 convolution kernels to make the network achieve the best recognition accuracy.The proposed method achieved a maximum recognition accuracy of 96.80%.