首页|基于多尺度一维卷积神经网络的弯管冲蚀损伤智能检测方法

基于多尺度一维卷积神经网络的弯管冲蚀损伤智能检测方法

扫码查看
针对高压管汇损伤需要提高检测效率和准确率的问题,提出一种基于多尺度一维卷积神经网络(multi-scale one-di-mensional convolutional neural network,MS-1 DCNN)的弯管冲蚀损伤智能检测新方法,即用多尺度卷积层代替传统的单一尺度卷积层.在MS-1DCNN模型中,把通过模拟实验所得弯管冲蚀损伤原始时域信号作为多尺度一维卷积神经网络的输入,这样能解决传统方法依赖人工提取特征和专家知识的问题;然后,通过多尺度卷积层和池化层的交替连接对输入信号进行特征提取;最后,经由输出层输出弯管冲蚀损伤分类结果.模型试验结果表明:基于MS-1DCNN弯管冲蚀损伤检测方法可以有效检测出弯管冲蚀损伤,且平均检测准确率达到99.18%.研究可为高压管汇冲蚀损伤智能检测提供一种新思路.
Intelligent Detection Method for Bending Pipe Erosion Damage Based on Multi-scale One-dimensional Convolutional Neural Network
In order to improve the detection efficiency and accuracy of high pressure manifold damage,a new intelligent detection method for bending erosion damage based on a multi-scale one-dimensional convolutional neural network(MS-1 DCNN)was proposed,and a multi-scale convolution layer was used to replace the traditional single-scale convolution layer.In the MS-1 DCNN model,the original time domain signal of bending erosion damage obtained through simulation experiments was used as the input of multi-scale one-dimensional convolutional neural networks,which can solve the problem that traditional methods rely on manual feature extraction and expert knowledge.Then,the feature extraction of input signals was carried out by alternating connection of multi-scale convolution layer and pooling layer.Finally,the classification results of erosion damage of bent pipe were output through the output layer.The model test results show that the erosion damage detection method based on MS-1 DCNN can effectively detect erosion damage of bent pipes,and the average detection accuracy is 99.18%.Research can provide a new approach for quantitative intelligent detection of erosion damage in high-pressure manifolds.

high pressure manifolderosion damageone-dimensional convolutional neural networkmulti-scaleintelligent detection

陈传智、李宁、王畅、陈家梁、罗锦达

展开 >

长江大学机械结构强度与振动研究所,荆州 434023

高压管汇 冲蚀损伤 一维卷积神经网络 多尺度 智能检测

中国石油科技创新基金江苏省油(气)井设备工程技术研究中心开放基金油气藏地质及开发工程国家重点实验室(西南石油大学)项目

2020D-5007-0503HT202102PLN2022-26

2024

科学技术与工程
中国技术经济学会

科学技术与工程

CSTPCD北大核心
影响因子:0.338
ISSN:1671-1815
年,卷(期):2024.24(5)
  • 22