针对旋转机械零部件进行故障诊断的方法包括传统方法和深度学习,传统方法往往需要大量的专家经验,且诊断精度欠佳,提出一种注意力机制改进多尺度深度卷积神经网络(multi-scale attention deep convolutional neural network,MADCNN)的...针对旋转机械零部件进行故障诊断的方法包括传统方法和深度学习,传统方法往往需要大量的专家经验,且诊断精度欠佳,提出一种注意力机制改进多尺度深度卷积神经网络(multi-scale attention deep convolutional neural network,MADCNN)的故障诊断方法。MADCNN方法提供3个卷积通道,每个通道差异化的核尺寸原理有效拓宽网络,实现了对原始时域数据的多尺度特征提取。同时,CBAM对提取的特征进一步赋予权重,增强了模型对不同类型故障的区分度。采用凯斯西储大学(Case Western Reserve University,CWRU)轴承故障数据和行星齿轮箱实验台故障数据分别进行实验验证,与传统深度卷积模型相比,验证集准确率提高7.76%。实验结果表明,该方法的诊断精度高,泛化性能好。展开更多
Fatigue performance of hot-rolled ribbed-steel bar with the yield strength of 500 MPa （HRB500） was stud- ied with bend-rotating fatigue test at a stress ratio of R = -1. It is determined by staircase method that its...Fatigue performance of hot-rolled ribbed-steel bar with the yield strength of 500 MPa （HRB500） was stud- ied with bend-rotating fatigue test at a stress ratio of R = -1. It is determined by staircase method that its fatigue strength for 107 cycles is 451 MPa, which is higher than that of common carbon structural steel. This should be at- tributed to the fine-grain strengthening resulting from the high content of alloy element V and Thermo-Mechanical Control Process （TMCP）. The S-N curve function is also obtained by nonlinear regression with three parameters power function. The fatigue fractures of the specimen were further analyzed with Scanning Electron Microscopy （SEM） and Energy Disperse Spectroscopy （EDS） to study the fracture mechanism. Taking into account microstruc- ture, hardness and cleanliness of the material, it implies that the fatigue fractures of HRB500 rebar all arise from surface substrates in which many brittle inclusions are contained, and that the fatigue crack propagation is principally based on the mechanism of quasi-cleavage fracture, because of the intracrystalline hard spots leading to stress con- centration and thus to the cracks. Moreover, the transient breaking area exhibits microvoid coalescence of ductile fracture due to the existing abundant inclusions.展开更多
文摘针对旋转机械零部件进行故障诊断的方法包括传统方法和深度学习,传统方法往往需要大量的专家经验,且诊断精度欠佳,提出一种注意力机制改进多尺度深度卷积神经网络(multi-scale attention deep convolutional neural network,MADCNN)的故障诊断方法。MADCNN方法提供3个卷积通道,每个通道差异化的核尺寸原理有效拓宽网络,实现了对原始时域数据的多尺度特征提取。同时,CBAM对提取的特征进一步赋予权重,增强了模型对不同类型故障的区分度。采用凯斯西储大学(Case Western Reserve University,CWRU)轴承故障数据和行星齿轮箱实验台故障数据分别进行实验验证,与传统深度卷积模型相比,验证集准确率提高7.76%。实验结果表明,该方法的诊断精度高,泛化性能好。
基金Item Sponsored by Fundamental Research Funds for the Central Universities of China(FRF-TP-15-062A3)
文摘Fatigue performance of hot-rolled ribbed-steel bar with the yield strength of 500 MPa （HRB500） was stud- ied with bend-rotating fatigue test at a stress ratio of R = -1. It is determined by staircase method that its fatigue strength for 107 cycles is 451 MPa, which is higher than that of common carbon structural steel. This should be at- tributed to the fine-grain strengthening resulting from the high content of alloy element V and Thermo-Mechanical Control Process （TMCP）. The S-N curve function is also obtained by nonlinear regression with three parameters power function. The fatigue fractures of the specimen were further analyzed with Scanning Electron Microscopy （SEM） and Energy Disperse Spectroscopy （EDS） to study the fracture mechanism. Taking into account microstruc- ture, hardness and cleanliness of the material, it implies that the fatigue fractures of HRB500 rebar all arise from surface substrates in which many brittle inclusions are contained, and that the fatigue crack propagation is principally based on the mechanism of quasi-cleavage fracture, because of the intracrystalline hard spots leading to stress con- centration and thus to the cracks. Moreover, the transient breaking area exhibits microvoid coalescence of ductile fracture due to the existing abundant inclusions.
