轻量化卷积神经网络在调门油动机故障诊断中的应用OA北大核心

Aimed at the difficulties in fault diagnosis and low maintenance efficiency in adjustment hydraulic servomotor,this study utilizes the high parallelism and low fragmentation characteristics of the ShuffleNetV2 network structure to one-dimensional extract its basic feature modules,and constructs a 1D_ShuffleNetV2 lightweighting network for the task of 10 classification of fault diagnosis of adjustment hydraulic servomotor.Based on the one-dimensional vibration signal of the adjustment hydraulic servomotor,this study conducts comparative experiments between 1D_ShuffleNetV2,1D_MobileNetV3,1D_ShuffleNetV1 and traditional one-dimensional residual network models.The results show that 1D_ShuffleNetV2 has the highest degree of lightness,the fastest convergence speed during training,the highest stability,and is able to effectively improve the data processing speed while maintaining high classification accuracy.This provides a new technical solution for health monitoring and fault diagnosis of adjustment hydraulic servomotors,which can reduce the hardware resource demand for edge-side devices while ensuring the diagnostic accuracy of adjustment hydraulic servomotors.