基于VMD-CNN的高噪声动态生产过程质量异常监控
Abnormal quality monitoring for strong noise dynamic process based on VMD-CNN
刘玉敏 1王德园 2王宁 1田光杰1
作者信息
- 1. 郑州大学商学院,郑州 450001
- 2. 郑州大学商学院,郑州 450001;郑州大学管理学院,郑州 450001
- 折叠
摘要
针对由智能制造现场动态生产过程的复杂随机因素影响造成的高噪声和质量异常监控方法效率低等问题,将变分模态分解方法(variational mode decomposition,VMD)与深度卷积神经网络(convolutional neural network,CNN)相结合,提出一种基于VMD-CNN的实时质量监控新方法.首先,利用VMD方法,将高噪声动态过程原始数据分解为包含质量异常特征和噪声信息的两类本征模态函数,通过去除噪声数据的本征模态函数,消除动态生产过程的高噪声干扰;进而,采用灰度变换将保留原始质量异常特征的本征模型函数转化为质量异常图像,构建VMD-CNN模型对质量异常图像进行识别,并提出基于VMD-CNN的高噪声动态过程质量异常实时监控框架;最后,通过实验验证所提方法的有效性,并与小波去噪方法和CNN识别模型进行对比分析,实验结果显示所提方法的识别精确度显著优于现有的动态过程质量异常监控方法.
Abstract
For the problems of strong noise and low quality abnormal monitoring efficiency in dynamic process of intelligent manufacturing scene,the real-time abnormal quality monitoring method is proposed by combining the variational mode decomposition(VMD)and deep convolutional neural network(CNN).Firstly,the VMD is used to divide the original data into two kinds of intrinsic model functions which reflect the quality abnormal characteristics and noise data.By removing the noise intrinsic model function,strong noise is disposed in the dynamic production process.Furthermore,the intrinsic model functions that retain the original quality abnormal features are transformed into quality abnormal images by gray level transformation,and the VMD-CNN model is constructed to identify the images to monitor dynamic abnormal quality process.Finally,the simulation experiments are used to compare and analysis the CNN,wavelet denoising reconstruction and the proposed method.Results show that the performance of the proposed method is significantly better than the existing dynamic process abnormal quality monitoring methods.
关键词
高噪声动态过程/变分模态分解/卷积神经网络/质量异常特征识别/质量监控Key words
strong noise dynamic process/variational mode decomposition(VMD)/convolutional neural network(CNN)/abnormal quality feature recognition/quality monitoring引用本文复制引用
基金项目
国家自然科学基金重点项目(U1904211)
国家社会科学基金(20BTJ059)
河南省高等学校青年骨干教师培养专项(2021GGJS006)
河南省高等学校哲学社会科学创新人才支持项目(22HASTIT022)
郑州大学精尖学科支持项目(XKLMJX202201)
出版年
2024
NETL
NSTL
万方数据