条件分布域适应下数模混动齿轮箱故障诊断  

Fault diagnosis of digital-analog hybrid transmission gearbox under conditional distribution domain adaptation

作　　者：王冉 韩海保 颜福成 余亮 WANG Ran;HAN Haibao;YAN Fucheng;YU Liang(School of Logistics Engineering,Shanghai Maritime University,Shanghai 201306,China;School of Civil Aviation,Northwestern Polytechnical University,Xi’an 710072,China;State Key Laboratory of Airliner Integration Technology and Flight Simulation,Shanghai 200126,China)

机构地区：[1]上海海事大学物流工程学院,上海201306 [2]西北工业大学民航学院,西安710072 [3]大型客机集成技术与模拟飞行全国重点实验室,上海200126

出　　处：《振动与冲击》2025年第3期182-190,209,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金项目(51505277);上海市自然科学基金(23ZR1426700);机械系统与振动国家重点实验室开放基金课题资助(MSV202305)。

摘　　要：齿轮箱的故障诊断对于确保机械系统的可靠性、安全性和经济可行性至关重要。在工业实际中,齿轮箱通常运行在正常状态下,因此故障状态发生较少,且由于获取有标签的故障数据的成本较高,导致齿轮箱的健康状态监测面临着有标签故障数据稀缺的问题。然而,现有的深度迁移诊断方法存在数据生成质量不均匀和过度依赖少数类信息等局限性。为了克服这一挑战,提出条件分布域适应下数模混动齿轮箱故障诊断方法。首先,基于集中参数法构建不同齿轮故障的动力学模型以扩充少标签源域的故障数据;其次,类条件分布最大均值差异(class-conditional maximum mean discrepancy,CMMD)被嵌入诊断模型中,在再生希尔伯特核空间中(reproducing kernel Hilbert space,RKHS)显式构建了故障特征与故障标签的关系,以减小源域数据和目标域数据的分布差异;同时,为保证目标域样本建立可靠的伪标签,熵损失被引入模型训练过程中;最后,通过两个试验验证了所提出方法的有效性和可行性。Fault diagnosis of gearbox is crucial for ensuring the reliability,safety and economic feasibility of mechanical systems.In industrial practice,gearboxes usually operate under normal conditions,fewer fault states appear.Due to higher cost of obtaining labeled fault data,health monitoring of gearboxes faces the problem of scarce labeled fault data.However,existing deep transfer diagnosis methods have limitations of uneven data generation quality and excessive reliance on minority class information.Here,to overcome this challenge,a fault diagnosis method for digital-analog hybrid transmission gearbox under conditional distribution domain adaptation was proposed.Firstly,based on the lumped parameter method,dynamic models of different gear faults were constructed to expand fault data in less labeled source domain.Secondly,the class-conditional maximum mean discrepancy(CMMD)was embedded into diagnosis model,and the relation between fault features and fault labels was explicitly constructed in reproducing kernel Hilbert space(RKHS)to reduce distribution differences between source domain data and target domain data.At the same time,to ensure establishing reliable pseudo labels for target domain samples,entropy loss was introduced into model training process.Finally,the effectiveness and feasibility of the proposed method were verified with two experiments.Diagnostic model to explicitly establish the relationship between fault features and fault labels in the reproducing kernel hilbert space(RKHS),reducing the distribution discrepancy between the source and target domain data.Meanwhile,to ensure reliable pseudo-labels for target domain samples,entropy loss is introduced during model training.Finally,the effectiveness and feasibility of the proposed method are validated through two experiments.

关 键 词：齿轮箱故障诊断 动力学建模 条件最大均值差异 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3