Particle swarm wavelet convolution for cross-domain multimodal analog circuit fault diagnosis

FENG Yifei; GUO Tao; QIN Haojie; HUANG Zhenyu; ZHAO Hang

Chinese

您当前的位置：

首页 >

文章列表页 >

Particle swarm wavelet convolution for cross-domain multimodal analog circuit fault diagnosis

Research & Approach | 更新时间：2026-05-18

- Particle swarm wavelet convolution for cross-domain multimodal analog circuit fault diagnosis
- Vol. 45, Issue 5, Pages: 52-58(2026)
- 作者机构：
  
  1. 中北大学极限环境光电动态测试技术与仪器全国重点实验室,山西,太原,030051
  2. 中北大学仪器与电子学院,山西,太原,030051
  3. 华南师范大学电子科学与工程学院（微电子学院）,广东,佛山,528225
  4. 山西江阳工程爆破有限公司,山西,太原,030041
  5. 大唐黑龙江发电有限公司龙唐供热分公司,黑龙江,哈尔滨,150077
- 作者简介：
- 基金信息：
- DOI：
  CLC： TP218
- Published：2026
- 稿件说明：
移动端阅览
FENG Yifei, GUO Tao, QIN Haojie, et al. Particle swarm wavelet convolution for cross-domain multimodal analog circuit fault diagnosis[J]. 2026, 45(5): 52-58.
DOI：

FENG Yifei, GUO Tao, QIN Haojie, et al. Particle swarm wavelet convolution for cross-domain multimodal analog circuit fault diagnosis[J]. 2026, 45(5): 52-58. DOI：

摘要

为突破传统模拟电路故障诊断依赖单模态数据的局限，提出一种粒子群小波卷积跨域多模态模拟电路故障诊断。该方法构建时域、频域和时频域三模态特征空间，利用小波包变换提取精细化时频特征，并借助粒子群优化（PSO）算法优化小波基函数参数。设计三路径混合编码器，其中并行的两路一维卷积神经网络(1D-CNN)和二维卷积神经网络(2D-CNN)分支分别处理原始时间信号，经过快速傅里叶变换的频域信号与小波时频图，引入注意力机制

实现跨模态特征交互。同时，通过级联空洞卷积、残差连接和自适应池化层构建多尺度残差结构，增强局部特征提取能力。

Abstract

To overcome the limitation that traditional analog circuit fault diagnosis relies on single-modal data

a particle swarm wavelet convolution for cross-domain multimodal analog circuit fault diagnosis method is proposed.This method constructs a three-modal feature space covering the time

frequency

and time-frequency domains

adopts wavelet packet transform to extract refined time-frequency features and optimizes wavelet basis function parameters with the assistance of the particle swarm optimization(PSO) algorithm.A three-path hybrid encoder is designed

in which two parallel branches of one-dimensional convolutional neural network(1D-CNN) and two-dimensional convolutional neural network(2D-CNN) respectively process the original time-domain signal

frequency-domain signal converted by fast Fourier transform and wavelet time-frequency maps.An attention mechanism is introduced to realize cross-modal feature interaction.Meanwhile

a multi-scale residual structure is constructed by cascading dilated convolutions

residual connections and adaptive pooling layers to enhance the local feature extraction capability.

关键词

Keywords

references

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Rigid-flexible coupled two-degree-of-freedom rehabilitation exoskeleton based on sEMG

Research on load recognition of FBG tactile  sensing array based on CNN

Collaborative facial expression recognition method integrating image synthesis and feature learning

CNN-LSTM lower limb joint prediction method based on attention mechanism

Power grid relay protection fault detection based on multi-sensor imaging and CNN

Related Author

LI Xingshan

XU Shuai

ZHANG Jiaqi

SHEN Zhihui

XU Yiming

LU Guan

REN Xiwei

XIAO Man

Related Institution

School of Mechanical Engineering,Shenyang Ligong University

School of Electrical Engineering and Automation,Nantong University

School of Mechanical Engineering,Nantong University

School of Electronic Information and Artificial Intelligence,Shaanxi University of Science & Technology

School of Mechanical Engineering,North China University of Science and Technology

⁰