针对单一特征对粉末床缺陷表达不明确导致检测效果不佳的问题，提出了一种基于特征融合的增材制造过程粉末床缺陷视觉检测方法。该算法分别使用SIFT方法、灰度共生矩阵和Hu不变矩提取尺度空间特征、纹理特征和几何特征，借助词袋模型对每张图像构建3组视觉单词直方图，通过串行融合3组视觉单词直方图得到新的特征矩阵，采用特征选择对融合后特征矩阵进行降维，并传入随机森林分类器中进行训练。实验结果表明，不同特征对粉末床不同类型缺陷检测具有不同的贡献，优化特征融合参数后，算法平均准确率达到97.46%，缺陷检测效果明显提升。
A visual detection method for powder bed defects in the additive manufacturing process based on feature fusion is proposed to improve the poor detection effect due to unclear expression of single features for powder bed defects. The algorithm firstly extracts the scale-space features, texture features, and geometric features from each powder bed image using methods of SIFT, gray-level co-occurrence matrix and Hu invariant moments; then, these images are translated to the feature matrix by serially fusing the three visual word histograms constructed from the three kinds of features by the bag-of-words model; finally, the feature matrix of each image is reduced by feature selection, and passed to the random forest classifier for model training. Experimental results show that different features have different contributions to the detection of different types of defects; through optimizing the feature fusion parameters, the average accuracy of the proposed algorithm reaches 97.46%, significantly improving the defect detection effect.
[1] MARCO G, VITTORIO L, QUIRICO S, et al. In-process monitoring of selective laser melting:Spatial detection of defects via image data analysis[J]. Journal of Manufacturing Science and Engineering,2017,139(5):051001.
[2] KAHNERT M, LUTZMANN S, ZAEH M F. Layer formations in electron beam sintering[C]//Solid Freeform Fabrication Symposium. 2007:88-99.
[3] Z? H M F, LUTZMANN S. Modelling and simulation of electron beam melting[J]. Production Engineering, 2010,4(1):15-23.
[4] RAUSCH A M, KVNG V E, POBEL C, et al. Predictive simulation of process windows for powder bed fusion additive manufacturing:influence of the powder bulk density[J]. Materials,2017,10(10):1117.
[5] MURR L E, MARTINEZ E, GAYTAN S M, et al. Microstructural architecture, microstructures, and mechanical properties for a nickel-base superalloy fabricated by electron beam melting[J]. Metallurgical and Materials Transactions A, 2011,42(11):3491-3508.
[6] CUNNINGHAM R, NARRA S P, OZTURK T, et al. Evaluating the effect of processing parameters on porosity in electron beam melted Ti-6Al-4V via synchrotron X-ray microtomography[J]. JOM, 2016,68(3):765-771.
[7] DEPOND P J, GUSS G, LY S, et al. In situ measurements of layer roughness during laser powder bed fusion additive manufacturing using low coherence scanning interferometry[J]. Materials & Design, 2018, 154:347-359.
[8] CASATI R, LEMKE J, VEDANI M. Microstructure and fracture behavior of 316L austenitic stainless steel produced by selective laser melting[J]. Journal of Materials Science & Technology,2016,32(8):738-744.
[9] PUEBLA K, MURR L E, GAYTAN S M, et al. Effect of melt scan rate on microstructure and macrostructure for electron beam melting of Ti-6Al-4V[J]. Materials Sciences and Applications, 2012,3(5):259-264.
[10] CRAEGHS T, CLIJSTERS S, YASA E, et al. Online quality control of selective laser melting[C]//Solid Freeform Fabrication Proceedings, 2011:212-226.
[11] KLESZCZYNSKI S, JACOBSMVHLEN J Z, REINARZ B, et al. Improving process stability of laser beam melting systems[C]//Fraunhofer Direct Digital Manufacturing Conference, 2014:187-192.
[12] JACOBSMUHLEN J Z, KLESZCZYNSKI S, SCHNEIDER D, et al. High resolution imaging for inspection of Laser Beam Melting systems[C]//Instrumentation and Measurement Technology Conference (I2MTC), 2013:707-712.
[13] NEEF A, SEYDA V, HERZOG D, et al. Low coherence interferometry in selective laser melting[J]. Physics Procedia, 2014, 56:82-89.
[14] SCIME L, BEUTH J. Anomaly detection and classification in a laser powder bed additive manufacturing process using a trained computer vision algorithm[J]. Additive Manufacturing, 2018,19:114-126.
[15] SCIME L, BEUTH J. A multi-scale convolutional neural network for autonomous anomaly detection and classification in a laser powder bed fusion additive manufacturing process[J]. Additive Manufacturing, 2018, 24:273-286.
[16] 林椹尠, 李妮, 惠小强. 基于空间金字塔的BoW模型图像分类方法[J]. 西安邮电大学学报, 2018, 23(3):31-37. LIN Z X, LI N, HUI X Q. An image classification method of BoW model based on spatial pyramid[J]. Journal of Xi'an University of Posts and Telecommunications, 2018, 23(3):31-37(in Chinese).
[17] 闵信军. 基于灰度共生矩阵和视觉信息的布匹瑕疵检测方法研究[D]. 镇江:江苏大学, 2018:23-26. MIN X J. Research on detection methods of fabric defects based on gray level co-occurrence matrix and visual information[D]. Zhenjiang:Jiangsu University, 2018:23-26(in Chinese).
[18] 孙枭文. 基于纹理特征和Hu不变矩的KELM滤光片缺陷识别研究[J]. 甘肃科学学报, 2019, 31(5):17-22. SUN X W. Study on the identification of KELM filter defects based on texture characteristics and hu invariant moment[J].Journal of Gansu Sciences, 2019, 31(5):17-22(in Chinese).
[19] 陈静, 张艳新, 姜媛媛. 融合多特征与随机森林的纹理图像分类方法[J]. 传感器与微系统, 2019, 38(12):58-61. CHEN J, ZHANG Y X, JIANG Y Y. Texture image classification method combining multi-features and random forest[J]. Sensors and Microsystems, 2019, 38(12):58-61(in Chinese).
[20] 肖旎旖. 基于相关性和冗余性分析的特征选择算法研究[D]. 大连:大连理工大学, 2013:5-6. XIAO Y N. The research of feature selection algorithm based on analysis of relevancy and redundancy[D]. Dalian:Dalian University of Technology, 2013:5-6(in Chinese).
[21] 张鹏. 基于视觉的激光选区熔化成形铺粉质量在线监控系统研究[D]. 武汉:华中科技大学, 2017:36-51. ZHANG P. Research of a motoring system of deposition quality in selective laser melting based on machine vision[D]. Wuhan:Huazhong University of Science & Technology, 2017:36-51(in Chinese).
CJA