刮板输送机飘链故障诊断技术研究

发布时间：2018-03-22 23:02

本文选题：刮板输送机　切入点：卷积神经网络　出处：《煤炭科学技术》2017年05期 　论文类型：期刊论文

【摘要】：针对刮板输送机在其弯曲区段容易发生的飘链问题,提出了一种基于卷积神经网络和支持向量机的声音信号识别模型,该模型以经过PCA白化处理的综采工作面设备声音运行声音的声谱图为输入,由深度CNN网络提取声音信号的特征,并以SVM分类器实现对声音信号的识别,最终实现对刮板输送机飘链故障的诊断。同时推导了以SVM为输出层的深度CNN网络模型在训练时误差反向传播时输出层对全连接层的敏感度函数,并通过试验发现了对输入的声音信号进行不同时长的切分作为模型输入时,对CNN-SVM模型识别率产生影响的规律,最后通过对比试验验证了此模型确实比传统的GMM-HMM模型具有更高的识别准确率。
[Abstract]:Aiming at the floating chain problem of scraper conveyor in its bending section, a sound signal recognition model based on convolution neural network and support vector machine is proposed. In this model, the sound spectrum of the sound running sound of the equipment in the fully mechanized mining face after PCA whitening is taken as input, and the feature of the sound signal is extracted from the depth CNN network, and the recognition of the sound signal is realized by using the SVM classifier. Finally, the fault diagnosis of floating chain of scraper conveyor is realized. At the same time, the sensitivity function of the output layer to the full connection layer is derived when the error is back propagated by the depth CNN network model with SVM as the output layer. And through the experiment, we find the rule that the CNN-SVM model recognition rate is influenced by the different time segmentation of the input sound signal as the model input. Finally, the comparison experiment shows that the model has higher recognition accuracy than the traditional GMM-HMM model.
【作者单位】：西安科技大学机械工程学院;平顶山天安煤业股份有限公司六矿;
【基金】：国家自然科学基金资助项目(U1361121)
【分类号】：TD528.3;TN912.34

【参考文献】