面向云边端协同的扶梯轴承边缘智能故障诊断方法

Cloud-edge Collaboration Oriented Intelligent Edge Diagnosis for Escalator Bearing

为实现面向云边端协同的扶梯轴承智能故障诊断,提出基于边缘计算和深度迁移的扶梯轴承故障诊断方法。提出了云边协同智能监测平台总体架构及其运行机制;针对扶梯轴承边缘诊断模块,根据故障数据较少、故障特征不稳定的特征,提出了离散Stockwell变换(DST)结合子域适应的边缘迁移诊断方法;进一步,提出了基于OpenVINO的扶梯轴承智能诊断的边缘部署方案并将该方案实施部署在边缘端。该方案采用线性运算融合、模型量化和模型剪枝的组合策略对模型作轻量化处理,以达到减少资源消耗的目的。通过迁移诊断实验,验证了所提边缘端智能迁移诊断方法具有良好的故障诊断能力。轻量化验证实验表明采用模型轻量化处理策略后,最大误诊率为3.13%,在保证诊断精度的同时,边缘推理的耗时轻量化后波动范围从60~80ms降低到20~40ms;单次边缘诊断的平均时间也从113.4ms降低到74.2ms,达到了实时诊断的要求。

To build an intelligent fault diagnosis system for escalator bearings based on cloud-edge collaboration technology,a diagnosis method for escalator bearing edge based on computing and deep transfer learning is proposed.The overall architecture and operating mechanism of the cloud-edge collaborative intelligent monitoring platform are presented.For the edge diagnosis module of escalator bearings,an edge transfer diagnosis method combining distributed Stockwell transform and subdomain adaptation is proposed to address the issues of limited fault data and unstable fault features.Furthermore,an edge deployment scheme for intelligent diagnosis of escalator bearings based on OpenVINO is proposed and deployed at the edge.This scheme adopts a combination strategy of linear operation fusion,model quantization,and model pruning to lighten the model and achieve the goal of reducing resource consumption.Through transfer diagnosis experiments,the proposed edge intelligent transfer diagnosis method demonstrates excellent fault diagnosis capability.Lightweight verification experiments show that after applying the model lightweighting strategy,the maximum misdiagnosis rate is 3.13%.While ensuring diagnosis accuracy,the time consumption of edge inference is lightweighted from a fluctuation range of 60~80 ms to 20~40 ms.The average time for a single edge diagnosis is reduced from 113.4 ms to 74.2 ms,meeting the requirement for real-time diagnosis.