齿轮箱故障边缘智能诊断方法及应用研究

针对齿轮箱运行状态监测数据量大而数据价值密度低导致的数据传输和存储困难、受到带宽影响导致的故障辨识实时 性差以及大而深的深度学习模型难以有效部署至边缘端硬件等问题,本文提出了一种基于乘法-卷积网络(MCN)的齿轮箱故 障边缘智能诊断方法。 首先,综合考虑信号滤波在特征表征以及深度学习在特征提取的优势,设计了一种轻量化的 MCN 模型, 同时在嵌入式微处理器搭建了一套端侧边缘智能处理原型与系统。 该系统可以直接部署于齿轮箱边缘,通过云服务器训练和 更新 MCN 模型参数并部署至边缘端,于边缘端完成数据采集、处理和故障状态辨识等功能,将大量传感器数据直接消耗在边缘 端。 实验结果显示 MCN 具有 99. 75% 的平均识别精度,且部署 MCN 的齿轮箱故障边缘智能诊断系统可以在 0. 696 s 内准确识 别出故障状态。

Edge intelligent fault diagnosis method in the application of gearbox

To address the problems such as difficult data transmission and storage due to the large amount of operational status monitoring low-value density data, poor real-time performance of fault identification due to bandwidth impact, and the difficulty of deploying effectively large and deep learning models to edge-side hardware, this study proposes a gearbox edge intelligent fault diagnosis method based on multiplicative-convolutional network (MCN). Firstly, motivated by the merits of feature representation in signal filtering and feature extraction in deep learning, a lightweight MCN model is formulated. Secondly, a set of end-side edge intelligent processing unit prototype is made by using the embedded microcontroller unit. The system can be deployed directly at the edge of the gearbox, where the parameters of the MCN-based edge model can be trained and updated on the cloud side and deployed to the edge. The edge-side completes data acquisition, processing,and fault status identification, which can consume a large amount of sensor data directly. The experimental results show that MCN has an average recognition accuracy of 99. 75% , and the gearbox edge intelligent diagnosis system deployed with MCN can accurately identify the fault state at 0. 696 s.