基于一维卷积神经网络的自动扶梯机械故障分类研究

老旧扶梯机械故障较为隐蔽,定期检验不易发现,且对扶梯机械故障的智能分类的研究较少。自动扶梯振动信号复杂多变,数据量大,而采用传统机器学习算法对其机械故障进行诊断效果不佳。为实现自动扶梯机械故障的智能分类,在经典二维卷积神经网络的基础上,引入了卷积核的一维卷积神经网络,构建了自动扶梯机械故障的自动分类模型。首先为提高模型的泛化性能,融合凯斯西储大学轴承故障、东南大学齿轮故障和某大型商场自动扶梯梯级滚轮磨损故障的复合故障数据建立了数据集。然后用数据增强的方法对数据进行预处理,接着采用一维卷积神经网络,构建自动扶梯机械故障诊断模型。最后使用测试数据集对模型的分类精度进行了验证实验,结果表明该模型有着比传统机器学习算法自动化程度高、成本低、专业门槛低、步骤简单等明显优势,而且该模型能快速准确地对自动扶梯的机械故障进行自动诊断,实现了95%的诊断准确率,为下一步将该算法集成到检验仪器中打下了基础。

Research on Escalator Mechanical Fault Classification Based on One-Dimensional Convolutional Neural Network

Mechanical faults of old escalators are hidden,which are not easy to be found by regular inspection,and there is little research on intelligent classification of escalator mechanical faults.The escalator vibration signal is complex and changeable,and the data is large,but the traditional machine learning algorithm is not effective in diagnosing its mechanical fault.In order to realize the intelligent classification of escalator mechanical faults,an one-dimensional convolutional neural network with large convolution kernel is introduced,which is based on the classical two-dimensional convolutional neural network.Firstly,the bearing fault data sets of Case Western Reserve University and the gearbox fault data sets of Southeast University and the composite fault data of escalators in a large shopping mall are used to establish the data sets.Then the data is preprocessed by data enhancement method,and then the escalator mechanical fault diagnosis model is constructed by using one-dimensional convolutional neural network.Finally,the classification accuracy of the model is verified by the test data set.The results show that the model has obvious advantages such as higher automation,lower cost,lower professional threshold and simpler steps than traditional machine learning algorithms.And this model can automatically diagnose the mechanical faults of escalators quickly and accurately,and the diagnosis accuracy is 95%,which lays a foundation for integrating the algorithm into inspection instruments in the next step.