首页|小样本情况下基于深度学习的设备健康状态识别研究

小样本情况下基于深度学习的设备健康状态识别研究

扫码查看
原文链接
  • 万方数据
  • 维普
针对工业生产中难以大量获取已标注设备状态数据的问题,文中提出了一种基于深度学习的设备健康状态识别模型,以实现小样本情况下的状态识别.首先,通过离散小波变换去除一维振动信号的噪声,通过格拉姆角场(GAF)将其转换为二维图像,并进行灰度化处理,以简化矩阵并提高深度学习的运算效率;其次,提出了一种深度学习模型,该模型运用Siamese架构,将深层残差结构作为子模块,并采用SimAM自注意力机制对残差结构进行改进,以SRes-Si-mAM205进行模型指代;最后,为了更快更准确地找到最优值以获得更高的精度,使用OneCycleLR函数自适应调整学习率,为了验证所构建方法的有效性,选择使用中南大学提供的齿轮数据集进行算例分析.试验结果表明:所提出的方法可以更好地提取特征并获得99.9％的识别精度,具有良好的泛化能力和适应性.
Research on state recognition of device conditions in small-sample scenarios based on deep learning
Since it is difficult to acquire a large amount of data on the annotated device conditions in industrial production,in this article a model is set up for state recognition of device conditions based on deep learning in small-sample scenarios.First-ly,noise in the one-dimensional vibration signals is removed by means of discrete wavelet transformation.These signals are then transformed into the two-dimensional images with the help of Gramian Angular Field(GAF)and processed in grayscale to simplify the matrix and improve the efficiency of deep learning in computation.Secondly,a model of deep learning is introduced.By means of the Siamese architecture,the deep residual structure as a sub-network is subject to improvement through the SimAM self-attention mechanism,which is referred to as SRes-SimAM205.Thirdly,in order to improve both the speed and the accuracy in identifying the optimal value for higher precision,the OneCycleLR function is used to adjust the learning rate in a self-adaptive manner.Finally,in order to validate this method,a gear dataset provided by Central South University is used to conduct the case analysis.The results demonstrate that this method can better extract features and achieve the recognition accuracy of 99.9％,with excellent generalization and adaptability.

state recognitiondeep learningsiamese architectureresidual structureattention mechanism

陈佳宁、刘勤明、谢世锐

展开 >

上海理工大学 管理学院,上海 200093

状态识别 深度学习 孪生架构 残差结构 注意力机制

国家自然科学基金国家自然科学基金上海市科技创新行动计划宝山转型发展科技专项(2021)上海市大学生创新创业训练计划(2023)上海理工大学科技发展项目

716320087184000321SQBS01404SH20230722020KJFZ038

2024

机械设计
中国机械工程学会,天津市机械工程学会,天津市机电工业科技信息研究所

机械设计

CSTPCD北大核心
影响因子：0.638
ISSN：1001-2354
年,卷(期)：2024.41(2)
  • 20