旋转机械振动监测及故障诊断(机械故障诊断及工业工程故障诊断若干例子（第5篇）)

Posted 2023-05-28

篇首语：别裁伪体亲风雅，转益多师是汝师。本文由小常识网(cha138.com)小编为大家整理，主要介绍了旋转机械振动监测及故障诊断(机械故障诊断及工业工程故障诊断若干例子（第5篇）)相关的知识，希望对你有一定的参考价值。

旋转机械振动监测及故障诊断(机械故障诊断及工业工程故障诊断若干例子（第5篇）)

Python环境下基于多传感器数据和周期性采样的滚动轴承故障诊断方法

算法程序运行环境为Python，采用pytorch深度学习模块

ReadData_2.py : Channel fusion and periodic sampling.

model_2_view.py : the code of training and testing, and it provides functions to evaluate the performance changes under different stripe and sample length.

Python环境下基于注意力机制的小样本轴承故障诊断

算法程序代码运行环境为Python，采用Pytorch深度学习模块，执行基于注意力机制的小样本轴承故障诊断，附带参考文献。

Python环境下基于1D-CNN、2D-CNN和LSTM的一维信号分类

算法程序运行环境为Python，采用tensorflow深度学习模块，执行基于1D-CNN、2D-CNN和LSTM的一维信号分类（轴承故障诊断），可迁移至其他一维信号。

所使用的模块如下

import scipy.io import numpy as npfrom sklearn.model_selection import train_test_split, KFoldfrom sklearn.metrics import confusion_matriximport tensorflow as tffrom tensorflow.keras import layers, modelsimport matplotlib.pyplot as pltimport seaborn as sns import pandas as pd

MATLAB环境下基于机器学习的旋转机械故障诊断方法

算法程序运行环境为MATLAB R2018a，从旋转机械振动信号中提取相关特征，然后使用 KNN 算法进行故障聚类。

旋转机械故障包括轴承故障（bearing）、齿轮啮合故障（gearmesh）、不平衡故障（imbalance）、不对中故障（misalignment）、多点故障（multifault）、共振故障（resonance）

首先，观察一下各故障的时域波形及对应的频谱图

 plot_frequency_domain('bearing.mat') plot_frequency_domain('gearmesh.mat') plot_frequency_domain('imbalance.mat') plot_frequency_domain('misalignment.mat') plot_frequency_domain('multifault.mat') plot_frequency_domain('resonance.mat')

然后对信号进行特征提取（只提取轴承故障、齿轮啮合故障、不平衡故障、不对中故障、谐振故障状态的信号）

x_gear=segment_data('bearing.mat');x_normalised_gear=normalize(x_gear);f1_bearing=calculate_f1(x_normalised_gear);f2_bearing=calculate_f2(x_normalised_gear);  f3_bearing=calculate_f3(x_normalised_gear);f4_bearing=calculate_f4(x_normalised_gear);%=================  Gearmesh    ========================%x_gearmesh=segment_data('gearmesh.mat');x_normalised_gearmesh=normalize(x_gearmesh);f1_gearmesh=calculate_f1(x_normalised_gearmesh);f2_gearmesh=calculate_f2(x_normalised_gearmesh);f3_gearmesh=calculate_f3(x_normalised_gearmesh);f4_gearmesh=calculate_f4(x_normalised_gearmesh);%================== Imbalance   ========================%x_imbalance=segment_data('imbalance.mat');x_normalised_imbalance=normalize(x_imbalance);f1_imbalance=calculate_f1(x_normalised_imbalance);f2_imbalance=calculate_f2(x_normalised_imbalance);f3_imbalance=calculate_f3(x_normalised_imbalance);f4_imbalance=calculate_f4(x_normalised_imbalance);%=================  misalignment    ======================%x_misalignment=segment_data('misalignment.mat');x_normalised_misalignment=normalize(x_misalignment);f1_misalignment=calculate_f1(x_normalised_misalignment);f2_misalignment=calculate_f2(x_normalised_misalignment);f3_misalignment=calculate_f3(x_normalised_misalignment);f4_misalignment=calculate_f4(x_normalised_misalignment);%=================  resonance   ==========================%x_resonance=segment_data('resonance.mat');x_normalised_resonance=normalize(x_resonance);f1_resonance=calculate_f1(x_normalised_resonance);f2_resonance=calculate_f2(x_normalised_resonance);f3_resonance=calculate_f3(x_normalised_resonance);f4_resonance=calculate_f4(x_normalised_resonance);bearing_fault=[f1_bearing;f2_bearing;f3_bearing;f4_bearing]';gearmesh_fault=[f1_gearmesh;f2_gearmesh;f3_gearmesh;f4_gearmesh]';misalignment_fault=[f1_misalignment;f2_misalignment;f3_misalignment;f4_misalignment]';imbalance_fault=[f1_imbalance;f2_imbalance;f3_imbalance;f4_imbalance]';resonance_fault=[f1_resonance;f2_resonance;f3_resonance;f4_resonance]';G=[bearing_fault;gearmesh_fault;misalignment_fault;imbalance_fault;resonance_fault];save 'fault_data_merged' G bearing_fault gearmesh_fault misalignment_fault imbalance_fault resonance_fault;

至于提取的啥子特征，以calculate_f1为例，看一下

function [f1] = calculate_f1(x_normalised)[P,f]=pwelch(x_normalised,[],[],[],1000);f1=rms(P);end

即calculate_f1提取的是功率谱密度相关特征

最后进行KNN聚类

load fault_data_merged.mat;c=corrcoef(G); % Calculates a correlation coefficient matrix c of G[v,d] =eig(c); % Find the eigenvectors v and the eigenvalues d of GT=[ v(:,end)';v(:,end-1)']; % Create the transformation matrix T from% the first two principal componentsz=T*G'; % Create a 2-dimensional feature vector za=0:50:250;color='or' 'or' 'sg' 'dc' '^k' '<b';figure();for i=2:6    plot(z(1,a(i-1)+1:a(i)),z(2,a(i-1)+1:a(i)),colori) % Scatter plot of the 2-dimensional features    hold onend title('fault clusters for LAB')

MATLAB环境下基于CNN的轴承故障诊断及特征可视化

算法程序运行环境为MATLAB R2018a，使用 CNN 进行滚动轴承故障诊断，原始数据来自西储大学轴承数据中心，包含3种故障工况（内圈故障，外圈故障和滚动体故障）和1种正常工况。

首先从数据中心下载如下文件

然后建立4个文件夹，并运行pic函数导入各工况图像

folder_name1 = 'set1_inner';folder_name2 = 'set2_normal';folder_name3 = 'set3_outer';folder_name4 = 'set4_roll';pic(2000,200,106,folder_name1);pic(2000,200,98,folder_name2);pic(2000,200,131,folder_name3);pic(2000,200,119,folder_name4);

导入数据

digitDatasetPath = fullfile('.') imds = imageDatastore(digitDatasetPath, ...    'IncludeSubfolders',true,'LabelSource','foldernames');

计算每个标签的图像数量

Count_of_Lable = countEachLabel(imds)

划分训练集

number_of_each_train_test = 150;[imds_of_train_set,imds_of_validation_set]= splitEachLabel(imds,number_of_each_train_test, 'randomize');

定义CNN网络

layers = [    imageInputLayer([64 86 3])        convolution2dLayer(3,8,'Padding','same')    batchNormalizationLayer    reluLayer        maxPooling2dLayer(2,'Stride',2)        convolution2dLayer(3,16,'Padding','same')    batchNormalizationLayer    reluLayer        maxPooling2dLayer(2,'Stride',2)        convolution2dLayer(3,32,'Padding','same')    batchNormalizationLayer    reluLayer        fullyConnectedLayer(4)    softmaxLayer    classificationLayer];

确定训练参数

options = trainingOptions('sgdm', ...    'InitialLearnRate',0.01,...    'LearnRateSchedule','piecewise',...    'LearnRateDropFactor',0.1,...    'LearnRateDropPeriod',10,...    'MaxEpochs',16, ...    'Shuffle','every-epoch',...    'ValidationData',imds_of_validation_set,...    'ValidationFrequency',10,...    'Verbose',false, ...    'Plots','training-progress');

训练与测试

CNN_network = trainNetwork(imds_of_train_set,layers,options);Prediction_of_validation_set = classify(CNN_network,imds_of_validation_set);True_lable_of_validation_set = imds_of_validation_set.Labels;accuracy = sum(Prediction_of_validation_set == True_lable_of_validation_set)/numel(True_lable_of_validation_set)

最后进行deepDream可视化，看看CNN学到了些啥玩意

以上项目的代码https://mbd.pub/o/GeBENHAGEN

此外，接受知乎咨询:哥廷根数学学派

擅长现代信号处理(改进小波分析系列，改进变分模态分解，改进经验小波变换，改进辛几何模态分解等等)，改进机器学习，改进深度学习，机械故障诊断，改进时间序列分析(金融信号，心电信号，振动信号等)