液压与气动2025，Vol.49Issue(7)：43-52,10.DOI:10.11832/j.issn.1000-4858.2025.07.005
基于多源异构传感信息与有效通道注意力—卷积神经网络的液压起重机迁移故障诊断
Fault Diagnosis of Hydraulic Cranes via Transfer Learning Based on Multi-source Heterogeneous Sensor Information and Effective Channel Attention-convolutional Neural Network
摘要
Abstract
The dynamic pressure signals of hydraulic systems have the characteristics of nonlinearity,multi-source coupling,and sensitivity to operating conditions.This results in high signal complexity and low feature discernibility,rendering traditional diagnostic methods ineffective.A deep learning diagnostic framework based on multi-sensor collaborative perception is proposed to address these issues.Multiple heterogeneous sensor signals are mapped into a multi-channel input tensor through spatial topology mapping,which preserves independent sensor features while achieving joint representation of multimodal information.A parallel convolutional architecture extracts spatiotemporal features from each channel,and an effective channel attention mechanism enhances fault-sensitive information,optimizing cross-modal features for precise classification.Experimental results show that the proposed method achieves over 95%accuracy in diagnosis of hydraulic pump leakage fault on the UCI standard hydraulic dataset.By introducing transfer learning theory,the pre trained model trained on the UCI standard hydraulic dataset is transferred to the forklift lifting hydraulic system,and the model still maintains an accuracy of 97.65%.These results confirm the model's strong generalization ability across different scenarios and provide an effective solution for fault diagnosis in complex hydraulic systems.关键词
液压系统/多传感器信息融合/故障诊断/深度学习/有效通道注意力Key words
hydraulic system/multi-sensor information fusion/fault diagnosis/deep learning/effective channel attention分类
机械制造引用本文复制引用
郭媛,王成龙,湛从昌,夏欢..基于多源异构传感信息与有效通道注意力—卷积神经网络的液压起重机迁移故障诊断[J].液压与气动,2025,49(7):43-52,10.基金项目
国家重点研发计划(2021YFB2011200) （2021YFB2011200）
国家自然科学基金(52475210) （52475210）
