基于一维改进SqueezeNet和WSNs节点计算的轴承故障诊断方法

Bearing Fault Diagnosis Method Based on One-dimensional Improved SqueezeNet and WSNS Node Computation

为充分利用Squeeze Net的轻量、高效以及无线传感器网络(Wireless sensor networks,WSNs)易于安装、成本低的优点,同时为弥补WSNs带宽受限的缺陷,提出了一种基于一维改进SqueezeNet模型和WSNs节点计算的轴承故障诊断方法。通过简化模型结构、改进fire模块、添加BN模块和旁路结构的方法,对经典SqueezeNet模型进行了改进,得到了一维改进SqueezeNet-BN-Bypass模型。实验表明,该模型的参数量和计算量明显低于SqueezeNet模型,适于嵌入资源受限的WSNs节点;其故障诊断准确率的仿真值为99.1%,接近经典SqueezeNet模型的仿真值(99.3%);该模型嵌入节点后的准确率约为98.8%。此外,与直接传送原始数据相比,该方法可减少99.9%的数据传输量。

To take full advantage of SqueezeNet’s lightweight,high efficiency and the ease of installation and low cost of wireless sensor networks(WSNS),and to compensate for the bandwidth limitations of WSNs,a bearing fault diagnosis method based on one-dimensional improved SqueezeNet model and WSNS node calculation is proposed.By simplifying model structure,improving fire module,adding BN module and bypass structure,the classical SqueezeNet model is improved,and the one-dimensional improved SqueezeNet-BN-bypass model is obtained.The experimental results show that the parameters and computation of the model are lower than those of the SqueezeNet model,and the model is suitable for embedding resource-constrained WSNS nodes,and the simulation value of the fault diagnosis accuracy is 99.1%,which is close to the simulation value of the classical SqueezeNet model(99.3%)The accuracy of the model is about 98.8%after embedding nodes.In addition,the method can reduce the amount of data transmission by 99.9%compared with direct transmission of raw data.