融合PSO优化的相关变模态分解与深度学习的旋转机械早期故障智能分类方法

Early Fault Intelligent Classification Method of Rotating Machinery Based on PSO-Relevant Variational Mode Decomposition and Deep Learning

针对旋转机械早期故障信号呈现微弱、相互干扰,易导致故障智能分类精度低的现状,提出一种融合优化的PSORVMD(particle swarm optimization-relevant variational mode decomposition)与SAE(stacked autoencoder)的旋转机械早期故障分类方法;智能分类方法主要由信号增强与智能分类两阶段组成;首先该方法利用所改进的PSO-RVMD分解电机-轴承系统的早期故障振动信号,通过定义的相关能量比概念计算各分量信号(IMFs)与原始信号之间的相关程度,筛选并重构相关程度高的分量,去除冗余与不相干的干扰与噪声成分,实现信号增强;最后,将增强的早期微弱信号输入到SAE模型中进行训练;利用SAE模型提取高层、抽象且利于分类的深度特征且在最后一层添加BP层,直接对提取的深度特征进行故障分类;通过仿真与实际电机-轴承系统振动信号验证了该方法的有效性,结果表明该方法能快速的实现旋转机械早期微弱故障的精确识别与诊断,提高故障特征学习与自动分类程度。

Aiming at the weakness and mutual interference of the early failure signals of rotating machinery,it is easy to cause the intelligent fault classification with low accuracy.An early fault classification method of rotating machinery based on PSO-RVMD(Particle Swarm Optimization-Related Variational Mode Decomposition)and SAE(Stacked AutoEncoder)is proposed.The main methods of intelligent classification are two phases of signal enhancement and intelligent classification.Firstly,Improved PSORVMD motor breakdown.-Early fault vibration signals of the bearing system,the correlation between each component signal(IMF component)and the original signal is calculated through the definition of the correlation energy ratio concept,the high correlation component is screened and reconstructed,eliminate the redundant and irrelevant interference and noise components,to achieve signal enhancement.Finally,the enhanced early weak signal is input into the SAE model for training.The SAE model is used to extract the high-level,abstract and class-specific depth features,and the BP layer is added on the last layer.The extracted deep features are directly simulated for fault classification with the motor.The bearing system vibration signal verifies the effectiveness of this method.The method can quickly identify and diagnose the early weak faults of rotating machinery,and improve the learning and automatic classification of fault features.