摘要
提出了一种创新的基于深度学习的激光选区熔化(SLM)铺粉状态识别方法,以应对传统铺粉异常检测方法中效率低下和精度有限的问题。通过模拟5种常见的铺粉状态(正常铺粉、划痕、铺粉不均匀、粉末不足和粉末过多),构建了一个包含1327张多角度图像的数据集。采用VGG-16、ResNet-101和EfficientNetV2-XL三种深度学习模型进行训练和测试。实验结果表明,该方法在实时检测铺粉异常方面相比传统方法具有显著优势。其中,VGG-16模型在保持高精度的同时推理速度最快,达到68.24 frame/s,适用于工业实时监控;而ResNet-101和EfficientNetV2-XL模型则在处理更复杂的异常类型(如铺粉不均匀)时表现更为出色,但推理速度相对较慢。基于Grad-CAM技术的注意力区域分析结果显示,VGG-16模型更适合检测局部异常,而ResNet-101和EfficientNetV2-XL模型在处理复杂背景和大范围异常方面具有优势。多角度图像数据集的构建和深度学习模型的比较分析,为SLM工艺的实时质量监控提供了新的解决方案,并为未来的模型优化和多模态传感器集成应用指明了方向。
Objective Selective laser melting(SLM)is a critical additive manufacturing process.However,defects that occur during the powder spreading can significantly affect the final product quality.Realtime detection of these anomalies is crucial for maintaining high manufacturing standards.This research proposes a deep learningbased method to identify powder spreading anomalies during the SLM process.We develop a dataset containing images under five common powder bed conditions:normal,scratch,uneven coating,insufficient coating,and excessive coating.The purpose is to classify these conditions accurately,thereby enhancing realtime monitoring and quality control in industrial applications.Methods We construct the dataset using an SLM machine,simulating five distinct powder bed states of normal,scratch,uneven powder coating,insufficient powder coating,and excessive powder coating.A total of 1327 images are captured from various angles for each defect type to enhance dataset diversity.Three deep learning models,namely,VGG16,ResNet101,and EfficientNetV2-XL,are trained and evaluated on this dataset.The dataset is divided into training and testing sets,and image preprocessing is applied that includes resizing and normalization.Model performance is assessed based on metrics such as accuracy,precision,recall,and F1-score(Table 1 and Fig.2).Results and Discussions Experimental results show that all three models perform well in classifying powder bed defects,with VGG16 achieving the best balance between speed and accuracy.VGG16 attains a classification accuracy of 99.63%and frame rate of 68.24 frame/s[Fig.4(a)],making it suitable for realtime applications.ResNet101 and EfficientNetV2-XL also perform well,particularly in identifying complex defects such as uneven powder distribution(Table 3).However,ResNet101 exhibits slower inference speeds,restricting its use in scenarios requiring rapid detection.EfficientNetV2-XL demonstrates robustness in detecting larger and more complex defects,achieving an accuracy of 99.19%(Table 4),but its slower processing speed limits its suitability for realtime systems.Heatmap analysis using GradCAM(Fig.7)reveals that VGG16 focuses more on localized defect regions,which contributes to its enhanced capabilities in detecting minor defects such as scratches.By contrast,ResNet101 and EfficientNetV2-XL exhibit broader attention,making them more effective in handling images with complex backgrounds.These findings underscore the importance of model selection based on defect type and realtime processing requirements.Conclusions We establish a novel multiangle dataset capturing five powder bed states in the SLM process,significantly enhancing model training diversity and robustness.The VGG16 model outperforms the others in terms of speed and precision,making it more suitable for realtime monitoring in industrial applications.EfficientNetV2-XL shows better performance in handling complex defects but has a slower inference speed,making it more suitable for less timesensitive scenarios.Future research will focus on optimizing these models through techniques such as multiscale feature fusion and integrating multimodal sensor data to improve defect detection accuracy across a wider range of manufacturing conditions.
作者
左明轶
国洪轩
李怀学
Zuo Mingyi;Guo Hongxuan;Li Huaixue(College of Software Engineering,Southeast University,Nanjing 210018,Jiangsu,China;School of Integrated Circuits,Southeast University,Nanjing 210018,Jiangsu,China;Aeronautical Science and Technology Key Laboratory of Additive Manufacturing,AVIC Manufacturing Technology Institute,Beijing 100192,China)
出处
《中国激光》
北大核心
2025年第8期167-178,共12页
Chinese Journal of Lasers
基金
国家重点研发计划(2023YFB4604900)。
关键词
激光选区熔化
深度学习
铺粉异常检测
图像处理
实时监控
selective laser melting
deep learning
powder spreading defect detection
image processing
realtime monitoring
