School of Mechanical Engineering, Xi'an University of Science and Technology No 6 Mine, Pingdingshan Tianan Coal Company
Limited

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。

According to the chain flying problems easily occurred at the bending section of the scraper conveyor, based on a convolution neural network and support vector machine, a s ound and signal identification model was provided. With the PCA whitening processing of the equipment sound from fully mechanized coal mining face, the sonogram of the operation sound would be an input. The depth CNN network would be applied to pick up the sound and signal features, the SVM classifier would be applied to realize the identification of the sound and signal and finally the diagnosis on the flying chain fault of the scraper conveyor could be realized. Meanwhile, based on the SVM as the depth of an output layer, the output layer of the CNN network model at an error reverse transmission during the training could be applied to derive a sensitivity function of the full connected layer. The test showed that when the input sound and signal had a different time cutting as the input of the model and would be affected to the identification of the CNN- SVM model. Finally, the comparison test verified that the model could have a higher identification than the conventional GMM-HMM model.

国家自然科学基金资助项目(U1361121)；