摘要
针对激光金属熔丝沉积过程不稳定、易导致缺陷、工艺参数调整存在一定模糊性且无法同时对多个熔池特征进行控制的问题,提出一种结合两种深度学习模型的工艺参数预测方法,将综合的熔池特征直接映射为工艺参数,为过程控制提供直接工艺参数指导。通过全因子实验和中心复合实验建立两种熔池数据集,采用YOLO模型对图像进行分割并提取丝材摆动幅度,结合卷积神经网络模型实现工艺参数预测。结果表明,YOLO模型能快速且准确地完成熔池和丝材分割,所建立的工艺参数预测模型在置信度为90%时对激光功率、扫描速度和送丝速率的预测准确率分别达99.94%、99.25%和99.64%,在置信度为98%时分别达81.07%、75.85%和90.24%。此外,新沉积单道的验证实验表明,该模型具有高准确性和适用性。
Laser wire melting deposition is an efficient additive manufacturing technique,but its process instability often causes defects.Current adjustments of process parameters remain somewhat ambiguous and lack the capability to simultaneously control multiple melt pool features.To address this issue,this paper proposes a process parameter prediction method based on a combination of two deep learning models,which directly maps comprehensive melt pool features to process parameters,providing direct guidance for process control.Two melt pool datasets are constructed through a full factorial experiment and a central composite experiment.A YOLO model is employed for image segmentation and to extract the wire oscillation amplitude,followed by a convolutional neural network(CNN)to predict process parameters.Results show that the YOLO model can quickly and accurately segment the melt pool and wire.The established prediction model achieved accuracies of 99.94%,99.25%and 99.64%for laser power,scanning speed and wire feeding speed respectively at 90%confidence level,and 81.07%,75.85%and 90.24%respectively at 98%confidence level.Validation experiments on newly deposited single tracks further confirm the model's strong accuracy and applicability.
作者
王江玲
石拓
孙承峰
魏超
张荣伟
谢宇
马烁焜
WANG Jiangling;SHI Tuo;SUN Chengfeng;WEI Chao;ZHANG Rongwei;XIE Yu;MA Shuokun(School of Optoelectronic Science and Engineering,Soochow University,Suzhou 215021,China;School of Mechanical and Electrical Engineering,Soochow University,Suzhou 215021,China)
出处
《电加工与模具》
北大核心
2025年第6期53-60,共8页
Electromachining & Mould
基金
国家自然科学基金项目(62173239)。
关键词
激光熔丝沉积
YOLO模型
深度学习
工艺参数
预测
laser wire melting deposition
YOLO model
deep learning
process parameter
prediction
