基于深度卷积迁移学习的机车齿轮箱故障诊断

Fault Diagnosis of Gearbox Based on Deep Convolutional Neural Network and Transfer Learning

齿轮箱作为系统传动的核心部件,确保其健康状态对于旋转机械有效运行至关重要。然而,目前大多数传统故障诊断方法通常难以充分挖掘故障相关特征信息,且常见模型在变工况服役条件下通用性差。与此同时,实际工程应用中往往难以获取充足标签数据。针对上述问题,提出一种基于深度卷积迁移学习的变工况机车齿轮箱故障诊断方法。首先,考虑到单一通道所含信息往往存在严重局限性,将多通道特征信息进行有机融合作为输入,搭建深度卷积网络自适应挖掘多通道深度特征,得到源域诊断模型。进一步将不同工况下多通道信号作为输入训练源域模型以增强其感知能力以及泛化性,由源域向目标域做迁移映射,从而实现变工况下的齿轮箱故障诊断。采用齿轮箱故障实验声学数据进行验证分析,结果表明:该方法能在不同的工况下实现知识迁移,增强诊断模型的通用性,准确高效地实现齿轮箱故障诊断,诊断准确率超过99%;对比其他传统故障诊断方法,所提方法有更好的时效性和泛化性。

As the core component of the system transmission,the gearbox is essential for the efficient operation of the rotating machinery to ensure its health.However,most of the current traditional fault diagnosis methods are usually difficult to fully mine the fault-related feature information,and the common models have poor versatility under variable working conditions and service conditions.At the same time,it is often difficult to obtain sufficient label data in practical engineering applications.In view of the above problems,a fault diagnosis method for a locomotive gearbox with variable working conditions based on DCNN-TL is proposed.First,considering that the information contained in a single channel often has serious limitations,the multi-channel feature information is organically fused as input,and a deep convolutional network is built to adaptively mine multi-channel deep features to obtain a source domain diagnosis model.Furthermore,multi-channel signals under different working conditions are used as input to train the source domain model to enhance its perception ability and generalizability,and transfer mapping is performed from the source domain to the target domain,so as to realize gearbox fault diagnosis under various working conditions.The acoustic data of the gearbox fault experiment is used for verification and analysis.The results show that the method can realize knowledge transfer under different working conditions,enhance the versatility of the diagnostic model,and accurately and efficiently realize gearbox fault diagnosis.The diagnostic accuracy rate exceeds 99%.Compared with other traditional fault diagnosis methods,the proposed method is faster and generalizable.