摘要
本文设计一款基于改进稀疏自编码器构建深度神经网络识别模型,可以准确感知精密机械零件运行状态。首先,基于谐波算法去除机械原始信号谐波分量,使用小波变换提取机械信号时频特征生成时域分析图;其次,构建稀疏自编码深度神经网络,为网络交叉熵损失函数增加权重衰减项与稀疏正则化项,实现自编码器稀疏约束,从而构建高性能的深度神经网络机械特征识别模型。测试结果显示:包含4层稀疏自编码器的深度神经网络模型识别机械运行特征效果最佳,误差在2.3%~4.0%之间,有效提升机械运行特征识别与分类精度,具有广阔的市场应用前景。
This paper designs a deep neural network recognition model based on improved sparse self encoder, which can accurately perceive the running state of precision mechanical parts.Firstly, the harmonic component of the original mechanical signal is removed based on the harmonic algorithm, and the time-frequency features of the mechanical signal are extracted by wavelet transform to generate the time-domain analysis diagram.Secondly, the sparse self coding depth neural network is constructed, the weight attenuation term and sparse regularization term are added to the network cross entropy loss function, and the sparse constraint of self coding is realized, so as to construct a high-performance depth neural network mechanical feature recognition model.The test results show that the deep neural network model including 4-layer sparse self encoder has the best effect in identifying mechanical operation features, with an error of 2.3% ~ 4.0%,which effectively improves the accuracy of mechanical operation feature recognition and classification, and has a broad market prospect.
作者
陈玉苗
CHEN Yu-miao(Hefei Gongda Vocational and Technical College,Hefei 231135,Anhui Province,China)
出处
《景德镇学院学报》
2022年第3期37-40,共4页
Journal of JingDeZhen University
关键词
深度神经网络
小波变换
稀疏
自编码器
机械
特征识别
deep neural network
wavelet transform
sparse
autoencoder
mechanics
feature recognition
