都正尧,张志芬,秦锐,李耿,温广瑞,何卫锋.基于声发射与关键帧选择的LSP表面硬度监测[J].表面技术,2022,51(11):35-44.
DU Zheng-yao,ZHANG Zhi-fen,QIN Rui,LI Geng,WEN Guang-rui,HE Wei-feng.Surface Hardness Monitoring of Laser Shock Peening:Acoustic Emission and Key Frame Selection[J].Surface Technology,2022,51(11):35-44
基于声发射与关键帧选择的LSP表面硬度监测
Surface Hardness Monitoring of Laser Shock Peening:Acoustic Emission and Key Frame Selection
  
DOI:10.16490/j.cnki.issn.1001-3660.2022.11.004
中文关键词:  表面硬度  激光冲击强化  声发射  关键帧  LSTM
英文关键词:surface hardness  laser shock peening  acoustic emission  key frame  LSTM
基金项目:国家自然科学基金资助项目(52175433);国家科技重大专项(2019-Ⅶ-0019-0161)
作者单位
都正尧 西安交通大学,西安 710049 
张志芬 西安交通大学,西安 710049 
秦锐 西安交通大学,西安 710049 
李耿 西安交通大学,西安 710049 
温广瑞 西安交通大学,西安 710049 
何卫锋 空军工程大学,西安 710038 
AuthorInstitution
DU Zheng-yao Xi'an Jiaotong University, Xi'an 710049, China 
ZHANG Zhi-fen Xi'an Jiaotong University, Xi'an 710049, China 
QIN Rui Xi'an Jiaotong University, Xi'an 710049, China 
LI Geng Xi'an Jiaotong University, Xi'an 710049, China 
WEN Guang-rui Xi'an Jiaotong University, Xi'an 710049, China 
HE Wei-feng Air Force Engineering University, Xi'an 710038, China 
摘要点击次数:
全文下载次数:
中文摘要:
      目的 提高激光冲击强化(Laser Shock Peening,LSP)表面硬度的在线检测能力,探究声发射信号关键帧对LSP表面硬度分类识别性能的影响。方法 在LSP处理期间声发射弹性波(Acoustic Emission,AE)具有与材料内部晶格位错和塑性变形密切相关的动态信息,是激光冲击强化在线监测的一种极具潜力的手段。但其高采样频率导致大量的实时计算,对在线监测技术的工业应用提出了巨大的挑战。为解决这一问题,提出了注意力权重统计方法获取激光冲击强化过程中声发射信号的关键帧。结果 四通道传感器各自的关键帧信号长度相比原始信号的有效长度均大幅度减少,最大可减少83.74%,相比原始数据每一轮测试(350个冲击样本),最大可减少57.37%的测试时间。关键帧信号的模型识别准确率最高可达到97.04%,相比原始数据集提升了2.93%。结论 与原始声发射信号相比,关键帧信号得到了更高的测试准确率,同时有效地减少了数据量。基于关键帧数据集的最高准确率和最短测试时间,评价了4种不同传感器中信号采集的最佳传感器,其结果可作为LSP质量监测领域的参考。
英文摘要:
      Laser Shock Peening (LSP) is a new technology which can improve the surface hardness of target and change the surface properties of metal materials. It has the advantages of good controllability and high processing efficiency, and is widely used in surface strengthening treatment of key components such as aero-engine blades. In order to overcome the problems of long cycle and low efficiency of off-line methods such as hardness testing, it is urgent to improve the on-line testing ability of laser shock strengthened surface hardness, improve the reliability and consistency of LSP processing quality while improving the quality testing efficiency, and ensure the service performance of aviation equipment. During LSP treatment, Acoustic Emission (AE) has dynamic information closely related to lattice dislocation and plastic deformation in the material, and it is a promising method for on-line monitoring of LSP. Therefore, we can identify the surface hardness of LSP by exploring the information of AE signal. In this paper, the surface hardness of 7 kinds of samples under different laser energy impacts is collected through off-line characterization experiments, and it is used as the basis for online detection and classification. However, the high sampling frequency of AE signals leads to a large number of real-time calculations, which poses a huge challenge to the industrial application of online monitoring technology. To solve this problem, attention weight statistics (AWS) method is proposed to obtain the key frames of acoustic emission signals during LSP. The time dimension weight of attention mechanism in LSTM model is visualized. After collecting a large number of attention weight samples, we find that some categories of attention weight samples are multi-peak, and multiple peak positions average the weight distribution, while some categories of attention weight samples are single-peak, and the peak positions cover most of the weights. Faced with the different weight distribution characteristics of each category, AWS method can comprehensively reflect the attention weight of each category of AE signals, and can be used as a reference for extracting key frames. Compared with the effective length of the original signal, the four-channel sensor's key frame signal length can be greatly reduced by 83.74% at the maximum, and the test time can be reduced by 57.37% at the maximum compared with each round of testing of the original data (350 AE samples). The model classification accuracy of key frame signals can reach 97.04%, which is 2.93% higher than the original data set. Compared with the original AE signal, key frames set of the signal provide greater test accuracy while effectively reducing the amount of data. Based on the highest accuracy and the shortest test time of key frames set, the best sensors of signal acquisition in four different sensors are evaluated. The method proposed in this paper is affected by the number of experimental samples, and at present it can only represent the detection effect of the network model on the existing data. Although the quantitative results will change with the change of experimental conditions, the method itself has certain universality, and the results can be used as a reference in the field of LSP quality monitoring.
查看全文  查看/发表评论  下载PDF阅读器
关闭

关于我们 | 联系我们 | 投诉建议 | 隐私保护 | 用户协议

您是第20731576位访问者    渝ICP备15012534号-3

版权所有:《表面技术》编辑部 2014 surface-techj.com, All Rights Reserved

邮编:400039 电话:023-68792193传真:023-68792396 Email: bmjs@surface-techj.com

渝公网安备 50010702501715号