带式输送机机械故障深度迁移学习诊断方法

Deep Transfer Learning Diagnosis Method for Mechanical Fault of Belt Conveyor

带式输送机早期机械故障诊断面临现场环境噪声干扰和工况多变等问题。提出了基于深度迁移学习的带式输送机早期机械故障诊断模型。首先利用集成经验模态分解(EEMD)对振动信号进行预处理,得到振动信号的时频特征组图和时频统计特征参数集;然后构建了一种ResNet与Transformer相结合的融合特征提取网络,提取振动信号的局部和全局时频特征;设计了基于域对抗神经网络的深度迁移方法对融合特征提取网络进行优化,以提升模型对变工况场景的适应能力;最后设计了基于物联网的带式输送机早期故障诊断和预测性维护系统,能够有效地提升带式输送机的生产效率和管理水平。

The belt conveyor early mechanical fault diagnosis faces challenges such as environmental noise interference and variable operating conditions.Proposed an early mechanical fault diagnosis model based on deep transfer learning.Firstly,vibration signals were preprocessed by using ensemble empirical mode decomposition(EEMD)to obtain time-frequency feature spectrograms and statistical feature parameter sets.Then a fusion feature extraction network combining ResNet and Transformer was constructed to extract both local and global time-frequency features from the vibration signals.A deep transfer method based on domain adversarial neural networks was designed to optimize the fusion feature extraction network,so as to enhance the model adaptability to changing operational conditions.Finally,an IoT-based early fault diagnosis and predictive maintenance system for belt conveyor was developed,which can effectively improve the productivity and management level of belt conveyor.