滚动轴承的正常运转是确保机械设备可以稳定工作的前提条件，而滚动轴承也是最容易出现损坏的零件。若轴承长期在恶劣且高负荷条件下持续运转工作，则极易发生故障。因此，轴承故障状态的检测在机械运转方面是十分有必要的。由于传统方法在轴承故障信号特征提取时，容易对该领域的专家知识和经验产生依赖，对算法的通用性较差，提取过程较为繁琐，因此，本文以滚动轴承故障信号为研究内容，总结相关的信号分析方法和技术，对轴承的故障特征做深入研究，并提出了相关的改进算法模型，在提高故障分类正确率上有一定的效果。论文的主要研究内容如下：

首先，为了可以更大程度的保留滚动轴承故障信号的本征特征，并对本征特征进行进一步的特征提取，对滚动轴承的基本组成及故障形成的原因和表现形式进行了详细分析，根据轴承各部件的振动频率，通过常用的三种信号分析方法进行实验对比，得出适合本实验的滚动轴承故障信号分析方法。

其次，针对滚动轴承故障识别的传统方法效果不理想，且对领域内专家经验的过分依赖问题，提出一种基于融合特征的T-CNN滚动轴承故障识别方法。该方法首先使用小波变换将原始故障信号转换成对应的时频图，然后将原始故障信号和时频图作为输入数据，一同输入到各自的网络模型中，将各自提取到的特征在汇聚层进行特征融合，生成新的T-CNN网络模型，并使用Softmax分类器对轴承信号进行故障识别。研究表明，该方法在识别不同故障状态的轴承方面，识别准确率均在99%左右。

最后，为了使融合特征的双通道模型识别出大批量滚动轴承数据，并提高故障信号的识别效率，提出一种基于T-BNCNN的滚动轴承故障识别方法。该方法采用一维BNCNN模型与二维BNCNN模型相结合的方法，分别将滚动轴承的原始一维故障信号和其对应信号经过小波变换后的二维时频图作为对应的输入数据，在两层卷积层和池化层后分别引入BN层，对输入的轴承数据进行故障特征提取，并对提取后的向量进行特征融合，并使用Softmax分类器对轴承实现故障诊断。分别采用小样本和大样本对实验进行验证，结果具有较高的识别效率和准确率。

In order to ensure that the mechanical equipment can work stably, the rolling bearing needs to run normally, and rolling bearings are also the most vulnerable parts. If the bearing is operated continuously for a long time under harsh and high load conditions, it is easy to fail. Therefore, the fault detection of rolling bearings is essential. Traditional signal detection methods usually rely on the expert knowledge and experience in the field when extracting the features of bearing faults, the algorithm of poor universality, extraction process is relatively complicated, therefore, based on the rolling bearing fault signal as the research content, summarizes the related signal analysis methods and techniques, characteristics of rolling bearing fault to do in-depth research, and put forward the further improved algorithm model, have certain effect on improving fault classification accuracy. The main research contents of this paper on bearing fault identification are as follows:

Firstly, to a greater degree of preserve the intrinsic characteristics of rolling bearing fault signal, and this feature for further feature extraction, the basic composition and the fault of rolling bearing formation reasons and forms are analyzed in detail, according to the bearing vibration frequency of each parts through the commonly used three kinds of signal analysis method through experiment contrast, The method of rolling bearing fault signal analysis suitable for this experiment is obtained.

Secondly, in order to solve the problems that traditional methods are not ideal in fault identification of rolling bearings and over-reliance on the experience of experts in the field, a T-CNN fault identification method for rolling bearings based on fusion features was proposed. This method firstly USES cmor3-3 wavelet transform to convert the original fault signal to the corresponding time-frequency diagram, then the original fault signal and the time-frequency diagrams of the network model, the input to the respective will each be extracted to features in the output layer of fusion, generate new T - CNN network model, finally, Softmax classifier is used to identify bearing fault signals. The results show that the accuracy of this method in the identification of bearings with different fault states is about 99%.

Finally, a rolling bearing fault identification method based on T-BNCNN was proposed in order to make the feature fusion two-channel model recognize a large number of rolling bearing data and improve the identification efficiency of fault signals.The method using a one-dimensional BNCNN model combined with two-dimensional BNCNN model method, respectively, will be the original one-dimensional fault signal of rolling bearing and its corresponding signals after wavelet transform, the 2-d time-frequency diagram as the corresponding input data, after the two layers of convolution and pooling layer respectively introduces BN layer, fault feature extraction of the input data, and to extract feature vector of the fusion, finally Softmax classifier is used to implement fault diagnosis. The experiment is verified by using small samples and large samples respectively, and the results have high recognition efficiency and accuracy.

[1] 王奉涛, 苏文胜. 滚动轴承故障诊断与寿命预测[M]. 北京：科学出版社, 2018.

[2] Li-Ming Wang,Yi-Min Shao.Crack Fault Classification for Planetary Gearbox Based on Feature Selection Technique and K-means Clustering Method[J].Chinese Journal of Mechanical Engineering,2018,31(01):242-252.

[3] 孙鲜明, 刘欢, 赵新光, 周勃. 基于瞬时包络尺度谱熵的滚动轴承早期故障奇异点识别及特征提取[J]. 机械工程学报, 2017,53(03):73-80.

[4] H k Jiang,Chengliang Li,Huaxing Li. An improved EEMD with multiwavelet packet for rotating machinery multi-fault diagnosis[J]. Mechanical Systems and Signal Processing,2013,36(2).

[5] “十一五”国家科技基础条件平台建设实施意见[J].中国科技信息,2006(07):8-11.

[6] Witten I H, Frank E, Hall M A, et al. Data Mining: Practical machine learning tools and

techniques[M]. Morgan Kaufmann, 2016.

[7] Johan A.K. Suykens. Support Vector Machines: A Nonlinear Modelling and Control Perspective[J]. European Journal of Control,2001,7(2-3).

[8] Duarte-Carvajalino, J.M, Yu G, Carin, L, et al. Task-Driven Adaptive Statistical Compressive Sensing of Gaussian Mixture Models[J]. IEEE Transactions on Signal Processing, 2011, 59(12):5842 - 5858.

[9] Harrington Peter. 机器学习实战[M].人民邮电出版社, 2015.

[10] 张建明，詹智财，成科扬等. 深度学习的研究与发展[J].江苏大学学报:自然科学版,

2015(2):191-200.

[11] 余凯, 贾磊, 陈雨强,等. 深度学习的昨天、今天和明天[J].计算机研究与发展, 2013,

50(9):1799-1804.

[12] 孙志远, 鲁成祥, 史忠植,等.深度学习研究与进展[J].计算机科学, 2016(2).

[13] HORRIGAN M, TURNER W J N, O’DONNELL J. A statistically-based fault detection approach for environmental and energy management in buildings[J]. Energy & Buildings, 2017, 158: 56-73.

[14] 郑君, 张冬泉. 故障诊断技术[J]. 电气时代, 2008,1(5): 96-98．

[15] 范宇中. 智能信息系统中的知识获取研究[D].武汉大学,2004.

[16] MARTIN R. Detection of ball bearing malfunctions [J]. Instruments and Control Systems, 1970, 1(12): 79-82.

[17] 安国庆. 异步电动机早期故障特征检测技术的研究[D], 河北工业大学, 2013.

[18] HARTING D R. Demodulated resonance analysis systems[P]. U. S. Patent: 3. 842. 663, 1974-10-22

[19] MC FADDEN P D, SMITH J D. The condition monitoring of rolling element bearings by the high frequency resonance technique—a review[J]. Tribology International, 1984, 17(1):3-10．

[20] Wang Dong, Tse Peter W. Tsui Kwok Leung. An enhanced Kurtogram method for fault diagnosis of rolling element bearings[J].Mechanical Systems & Signal Processing,2013,35(1-2):176-199.

[21] D. H. Pandya, S.H. Upadhyay, S.P Harsha. Fault diagnosis of rolling element bearing with intrinsic mode function of acoustic emission data using APF-KNN[J].Expert Systems with Applications,2013,40(10):4137-4145.

[22] Liu Huanhuan Han Minghong.A fault diagnosis method based on local mean decomposition and multi-scale entropy for roller bearings[J].Mechanism & Machine Theory,2014,75(5):67-78.

[23] Rajeswari C, Sathiyabhama B, Devendiran S, et al. Bearing Fault Diagnosis using Wavelet Packet Transform, Hybrid PSO and Support Vector Machine[J].Procedia Engineering Packet Transform, Hybrid PSO and Support Vector Machine[J].Procedia Engineering,2014,97:1772-1783.

[24] 杨青,孙佰聪,朱美臣,等.基于小波包熵和聚类分析的滚动轴承故障诊断方法[J].南京理工大学学报（自然科学版）,2013,37(4):51.

[25] 隋文涛,张丹.基于EMD和MKD的滚动轴承故障诊断方法[J].振动与冲击,2015,34(9):55-59.

[26] 王建国,李健,万旭东.基于奇异值分解和局域均值分解的滚动轴承故障特征提取方法[J].机械工程学报,2015,51(3):104-110.

[27] H.O.A. Ahmed, M.L.D. Wong, A. K. Nandi. Intelligent condition monitoring method for bearing faults from highly compressed measurements using sparse over-complete features[J]. Mechanical Systems and Signal Processing,2018,99.

[28] Akkad, MA Khaled. A deep belief network based approach for bearing fault diagnosis[J]. Chicago: The University of Illinois at Chicago,2016.

[29] Tra V, Kim j, Khan S A, et al. Bearing fault diagnosis under variable speed using convolutional neural networks and the stochastic diagonal levenberg-marquardt Algorithm[J].Sensors,2017,17(12):2834.

[30] Appana R D K, Kim J M. Reliable fault diagnosis of bearings using deep learning technique[C]. Korea: The Engineering and Arts Society in Korea,2017.

[31] Duong B P, Kim J M. Non-mutually exclusive deep neural network classifier for combined modes of bearing fault diagnosis[J].Sensors,2018,18(4):1129.

[32] Haidong Shao, Hongkai Jiang, Fuan Wang, Yanan Wang. Rolling bearing fault diagnosis using adaptive deep belief network with dual-tree complex wavelet packet[J]. ISA Transactions,2017,69.

[33] 姜涛,袁胜发.基于改进小波神经网络的滚动轴承故障诊断[J].华中农业大学学报,2014,33(1):131-136.

[34] Zhiqiang Chen, Shengcai Deng, Xudong Chen, Chuan Li, René-Vinicio Sanchez, Huafeng Qin. Deep neural networks-based rolling bearing fault diagnosis[J]. Microelectronics Reliability,2017,75.

[35] Xiaojie Guo,Liang Chen,Changqing Shen. Hierarchical adaptive deep convolution neural network and its application to bearing fault diagnosis[J]. Measurement,2016,93.

[36] Xiaojie Guo, Shen C, Chen L. Deep fault recognizer:an integrated model to denoise and extract features for fault diagnosis in rotating machinert[J].Applied Sciences,2016,7(1):41.

[37] 佘道明,贾民平.一种新型多层深度卷积神经网络的滚动轴承故障诊断方法[C].全国振动理论及应用学术会议,2017.

[38] Jia F,Lei Y,Lin J,et al.Deep neural networks:a promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data[J].Mechanical Systems & Signal Processing,2016,72-73:303-315.

[39] Rai A, Upadhyay S H. A review on signal processing techniques utilized in the fault diagnosis of rolling element bearings[J].Tribology International, 2016, 96:289-306.

[40] 林超．结合异常检测算法的轴承故障检测研究[D]. 浙江大学,2017.

[41] Hinton G E , Osindero S , Teh Y W . A Fast Learning Algorithm for Deep Belief Nets[J]. Neural Computation, 2014, 18(7):1527-1554.

[42] 刘林凡. 深度学习在故障诊断中的研究综述[J]. 新型工业化, 2017, 7(4):45-48.

[43] Rawat W, Wang Z. Deep Convolutional Neural Networks for Image Classification: A Comprehensive Review[J]. Neural Computation, 2017,29(9):2352-2449.

[44] Bergstra J. Theano: A CPU and GPU Math Compiler in Python[J].2012.

[45] Chan Tsung-Han,Jia Kui, Gao Shenghua,Lu Jiwen, Zeng Zinan,Ma Yi. PCANet: A Simple Deep Learning Baseline for Image Classification?[J]. IEEE transactions on image processing : a publication of the IEEE Signal Processing Society,2015,24(12):5017-5023.

[46] 黄驰城. 结合时频分析和卷积神经网络的滚动轴承故障诊断优化方法研究[D].浙江大学,2019.

[47] 童浩然. 基于深度学习的图像分类的研究[D], 2018.

[48] 刘建伟, 崔立鹏, 刘泽宇, 等. 正则化稀疏模型[J]. 计算机学报, 2015(07):1307-1325.

[49] Moore C M, MaWhinney S, Forster J E, etal. Accounting for dropout reason in longitudinal studies with nonignorable dropout[J]. Statistical Methods in Medical Research, 2015,26(4):1854-1866.

[50] 张毅. Bagging RCSP脑电特征提取算法[J], 2017.

[51] 李兴珂, 何云涛. 相干合成中的随机并行梯度下降算法性能研究[J]．光学学报, 2016(12):8-15.

[52] Jansson J, Li Z, Sung W. On finding the Adams consensus tree[J]. Information and Computation, 2017, 256: 334-347.

[53] Hadgu A T, Nigam A, Diaz-Aviles E. Large-scale learning with AdaGrad on Spark, 2015[C].IEEE, 2015.

[54] 张慧. 深度学习中优化算法的研究与改进[D]. 北京邮电大学, 2018.

[55] Diederik P. Kingma J L B. Adam: A Method for Stochastic Optimization[J], 2015.

[56] W. A. Smith, R. B. Randall. Rolling element bearing diagnostics using the case western reserveuniversity data: A benchmark study[J]. Mechanical Systems and Signal Processing, 2015, 64­65:100­131.

[57] B. Li, M. . Chow, Y. Tipsuwan, et al. Neural­network­based motor rolling bearing fault diagno­sis[J]. IEEE Transactions on Industrial Electronics, 2000, 47(5): 1060­1069.

[58] R. B. Randall, J. Antoni. Rolling element bearing diagnostics—a tutorial[J]. Mechanical Systemsand Signal Processing, 2011, 25(2): 485 ­ 520.

[59] F. Jia, Y. Lei, J. Lin, et al. Deep neural networks: A promising tool for fault characteristic miningand intelligent diagnosis of rotating machinery with massive data[J]. Mechanical Systems andSignal Processing, 2016, 72­73: 303­315.

[60] SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning[J]. 2016.

[61] SIMONYAN K, ZISSERMAN A. Very Deep Convolutional Networks for Large-Scale Image Recognition[J]. Computer Science, 2014.