A Composite Fault Diagnosis Method of Gearbox Combining with Convolution Neural Network and D-S Evidence Theory
-
摘要: 针对齿轮箱复合故障诊断问题,将深度卷积模型(CNN)和D-S证据理论相结合,对多传感器信息进行融合。首先,利用深度卷积模型对多个传感器信息进行自适应特征提取,经softmax进行初步分类。其次,将深度卷积模型的输出结果作为D-S证据理论的输入,计算出基本概率分配,根据Dempster合成法则进行决策融合。为验证此方法对齿轮箱复合故障诊断的有效性,使用BP神经网络与D-S证据理论模型作为对比,并对自适应提取的特征与人工特征进行了主成分分析(PCA)。实验结果表明,利用该方法对齿轮箱复合故障进行实验诊断,准确率达到84.58%。相比单一传感器,正确率提高了7.91%;相比BP神经网络与D-S证据理论模型,正确率提高了6.18%,验证了此方法的有效性。Abstract: For the problem of gearbox composite fault diagnosis, the multi-sensor information fusion is used based on the deep convolution neural network(CNN)and the D-S evidence theory method. First, the information features of multiple sensors are extracted adaptively based on the CNN model. The information is preliminarily classified based on softmax. Secondly, the output of the CNN model is used as the input of D-S evidence theory. After the basic probability distribution is calculated, decision fusion is made according to the Dempster synthesis rule. In order to validate the effectiveness of this method for composite fault diagnosis of gearbox, BP neural network and D-S evidence theory model are used as comparison. Principal component analysis(PCA)is carried out based on adaptive extraction of features and artificial features. Experiments show that the accuracy of the composite fault diagnosis of gearbox is 84.58%. Compared with single sensor, the accuracy is improved by 7.91%. Compared with BP neural network and D-S evidence theory model, the accuracy is improved by 6.18%. Thus this method is proved effective.
-
表 1 复合故障分类及其故障状态信息
故障类别 齿轮1 齿轮2 齿轮3 轴承1 轴承2 轴承3 输入轴转轴 输出轴转轴 1 正常 正常 正常 正常 正常 正常 正常 正常 2 纵向剥落 点蚀 正常 正常 正常 正常 正常 正常 3 正常 点蚀 正常 正常 正常 正常 正常 正常 4 正常 点蚀 断齿 滚动体故障 正常 正常 正常 正常 5 纵向剥落 点蚀 断齿 内圈故障 滚动体故障 外圈故障 正常 正常 6 正常 正常 断齿 内圈故障 滚动体故障 外圈故障 不平衡 正常 7 正常 正常 正常 内圈故障 正常 正常 正常 键剪断 8 正常 正常 正常 正常 滚动体故障 外圈故障 不平衡 正常 表 2 模型参数测试结果
组别 输入数据长度 滤波器尺寸 滤波器个数 正确率/% 1 1024 16 10、15 51.46 2 1024 32 10、15 55.42 3 1024 64 10、15 66.25 4 1024 128 10、15 55.21 5 2048 16 10、15 44.58 6 2048 32 10、15 68.75 7 2048 64 10、15 69.16 8 2048 128 10、15 82.91 9 2048 256 10、15 71.25 表 3 CNN模型最终参数
层数 结构类型 参数 1 卷积层 滤波器尺寸128, 滤波器个数10 2 池化层 池化块尺寸2 3 卷积层 滤波器尺寸128, 滤波器个数15 4 池化层 池化块尺寸2 5 全连接层 激活函数 Relu, 节点数10 6 Softmax 分类数8 表 4 CNN网络与BP神经网络故障诊断正确率
网络
类型传感器
类别测试
组数正确率/% 平均
正确率/%深度卷积网络 1 240 76.67 78.96 2 240 81.25 BP神经网络 1 240 73.48 72.89 2 240 72.30 表 5 两个深度卷积网络实际输出结果
CNN 实际输出 期望组别 诊断组别 结果 网络1 0.91693 0.01614 0.05847 0.00022 0.00018 0.00000 0.00540 0.00267 1 1 √ 0.00196 0.68290 0.30865 0.00079 0.00226 0.00000 0.00228 0.00116 2 2 √ 0.03155 0.00039 0.96792 0.00000 0.00000 0.00000 0.00007 0.00007 3 3 √ 0.10193 0.00077 0.12663 0.52960 0.00003 0.07439 0.05562 0.11101 4 4 √ 0.00215 0.47063 0.03778 0.10292 0.38234 0.00000 0.00001 0.00417 5 2 × 0.00290 0.00000 0.00152 0.02046 0.00000 0.77386 0.18275 0.01851 6 6 √ 0.05472 0.08768 0.22592 0.00325 0.00106 0.00001 0.62542 0.00194 7 7 √ 0.01407 0.04058 0.14334 0.68872 0.00817 0.00610 0.00026 0.09876 8 4 × 网络2 0.61256 0.00150 0.27435 0.00170 0.00649 0.00198 0.09804 0.00338 1 1 √ 0.09010 0.66193 0.20597 0.00448 0.03605 0.00028 0.00110 0.00009 2 2 √ 0.19378 0.00008 0.80609 0.00000 0.00002 0.00000 0.00003 0.00000 3 3 √ 0.04457 0.00010 0.43495 0.41977 0.00970 0.00720 0.00318 0.08052 4 3 × 0.04634 0.00598 0.62836 0.00982 0.30166 0.00000 0.00028 0.00757 5 3 × 0.00819 0.00000 0.00062 0.00035 0.00000 0.85036 0.13337 0.00710 6 6 √ 0.15673 0.00001 0.08389 0.00020 0.00022 0.00059 0.74414 0.01424 7 7 √ 0.00151 0.00000 0.00005 0.00016 0.00000 0.92520 0.03853 0.03456 8 6 × 表 6 D-S证据理论融合结果
期望组别 m(A1) m(A2) m(A3) m(A4) m(A5) m(A6) m(A7) m(A8) m(θ) 结果 1 0.96938 0.00014 0.02904 0.00001 0.00003 0.00001 0.00138 0.00000 0.00000 √ 2 0.00000 0.94088 0.00018 0.00000 0.05894 0.00000 0.00000 0.00000 0.00000 √ 3 0.00117 0.00000 0.99883 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 √ 4 0.03861 0.00016 0.24770 0.67957 0.00236 0.01598 0.01092 0.00471 0.00000 √ 5 0.00200 0.04136 0.17229 0.01183 0.77251 0.00000 0.00001 0.00001 0.00000 √ 6 0.00028 0.00000 0.00003 0.00023 0.00000 0.95971 0.03975 0.00000 0.00000 √ 7 0.01898 0.00245 0.04387 0.00009 0.00003 0.00000 0.93457 0.00000 0.00000 √ 8 0.02265 0.00019 0.11785 0.48140 0.00359 0.22686 0.01343 0.13403 0.00000 × -
[1] 俞昆, 谭继文, 李善.基于多传感器信息融合的滚动轴承故障诊断研究[J].仪表技术与传感器, 2016(7):97-102, 107 doi: 10.3969/j.issn.1002-1841.2016.07.027Yu K, Tan J W, Li S. Rolling bearing fault diagnosis research based on multi-sensor information fusion[J]. Instrument Technique and Sensor, 2016(7):97-102, 107(in Chinese) doi: 10.3969/j.issn.1002-1841.2016.07.027 [2] 徐卫晓, 谭继文, 文妍.基于BP神经网络和D-S证据理论的滚动轴承故障诊断方法研究[J].机床与液压, 2014, 42(23):188-191 doi: 10.3969/j.issn.1001-3881.2014.23.045Xu W X, Tan J W, Wen Y. Study on fault diagnosis methods of rolling bearing based on BP neural network and D-S evidence theory[J]. Machine Tool & Hydraulics, 2014, 42(23):188-191(in Chinese) doi: 10.3969/j.issn.1001-3881.2014.23.045 [3] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553):436-444 doi: 10.1038/nature14539 [4] 时永刚, 程坤, 刘志文.结合深度学习和支持向量机的海马子区图像分割[J].中国图象图形学报, 2018, 23(4):542-551 http://d.old.wanfangdata.com.cn/Periodical/zgtxtxxb-a201804009Shi Y G, Cheng K, Liu Z W. Segmentation of hippocampal subfields by using deep learning and support vector machine[J]. Journal of Image and Graphics, 2018, 23(4):542-551(in Chinese) http://d.old.wanfangdata.com.cn/Periodical/zgtxtxxb-a201804009 [5] Chen Z Q, Li C, Sanchez R V. Gearbox fault identification and classification with convolutional neural networks[J]. Shock and Vibration, 2015, 2015:390134 http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=Doaj000004007466 [6] Li C, Sánchez R V, Zurita G, et al. Fault diagnosis for rotating machinery using vibration measurement deep statistical feature learning[J]. Sensors, 2016, 16(6):895 doi: 10.3390/s16060895 [7] Jia F, Lei Y G, 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 and Signal Processing, 2016, 72-73:303-315 doi: 10.1016/j.ymssp.2015.10.025 [8] Li S B, Liu G K, Tang X H, et al. An ensemble deep convolutional neural network model with improved D-S evidence fusion for bearing fault diagnosis[J]. Sensors, 2017, 17(8):1729 doi: 10.3390/s17081729 [9] Hoang D T, Kang H J. Deep belief network and dempster-shafer evidence theory for bearing fault diagnosis[C]//Proceedings of the 27th International Symposium on Industrial Electronics. Cairns, QLD, Australia: IEEE, 2018 [10] Jing L Y, Zhao M, Li P, et al. A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox[J]. Measurement, 2017, 111:1-10 doi: 10.1016/j.measurement.2017.07.017 [11] 李善, 谭继文, 俞昆.神经网络和改进D-S证据理论相结合的滚动轴承复合故障诊断研究[J].机床与液压, 2018, 46(1):153-157, 184 doi: 10.3969/j.issn.1001-3881.2018.01.032Li S, Tan J W, Yu K. Composite fault diagnosis research of rolling bearing based on combination of neural network and improved D-S evidence theory[J]. Machine Tool & Hydraulics, 2018, 46(1):153-157, 184(in Chinese) doi: 10.3969/j.issn.1001-3881.2018.01.032 [12] Al-Atat H, Siegel D, Lee J. A systematic methodology for gearbox health assessment and fault classification[J]. International Journal of Prognostics and Health Management, 2011, 2(1):16 http://d.old.wanfangdata.com.cn/OAPaper/oai_doaj-articles_398dbcb378a1f5a4cffaad6c78ebec80 [13] Zhang C, Li P, Sun G Y, et al. Optimizing FPGA-based accelerator design for deep convolutional neural networks[C]//Proceedings of 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays. Monterey, California, USA: ACM, 2015: 161-170 [14] Jing L Y, Wang T Y, Zhao M, et al. An adaptive multi-sensor data fusion method based on deep convolutional neural networks for fault diagnosis of planetary gearbox[J]. Sensors, 2017, 17(2):414 doi: 10.3390/s17020414 [15] 唐贵基, 杨玉婧, 宋彩萌.基于神经网络的旋转机械振动故障诊断[J].机械工程师, 2012(1):40-42 doi: 10.3969/j.issn.1002-2333.2012.01.014Tang G J, Yang Y J, Song C M. Fault diagnosis of rotating machinery vibration based on neural network[J]. Mechanical Engineer, 2012(1):40-42(in Chinese) doi: 10.3969/j.issn.1002-2333.2012.01.014