基于CBAM-RCNN的乏燃料剪切机刀具故障诊断

Fault Diagnosis of Spent Fuel Shears Based on CBAM-RCNN Model

乏燃料剪切机是后处理厂的关键设备,一旦刀具磨损严重,后处理厂工作将受到干扰,乏燃料剪切机刀具的监测和故障诊断对后处理有重要意义.将工作噪声信号采用时移对数据进行增强,并转换为梅尔频谱图,运用SpecAugment技术对梅尔频谱第二次数据增强,作为模型的输入.采用全局平均池化层替代全连接层的卷积神经网络作为噪声提取分类的基础模型,将卷积块注意力模块(CBAM)和提高网络训练稳定性的残差网络(ResNet)融合进卷积神经网络(CNN),构建CBAN-RCNN模型.实验表明,该模型故障诊断准确率达到96.85%,相对于单独的CNN和CBAM-CNN分别提高了4.42%和3.18%,具有更好的诊断能力.

The spent fuel shears are the key equipment in the reprocessing plant.Once the tools are worn seriously,the reprocessing plant will be disturbed.Monitoring and fault diagnosis of the spent fuel shears tools are of great significance for reprocessing.The working noise signal is enhanced with time shift,and converted into Mel spectrum.SpecAugment technology is used to enhance the second data of Mel spectrum as the input of the model.The convolutional neural network with global average pooling layer instead of full connection layer is used as the basic model for noise extraction and classification.Then CBAM and residual block are integrated into CNN to build CBAN-RCNN model.The experimental results show that the fault diagnosis accuracy of the model reaches 96.85%,which is 4.42%and 3.18% higher than that of CNN and CBAM-CNN,respectively.The model has better diagnostic ability.