基于深度学习的轴承寿命预测实践，开发CNN、融合LSTM/GRU/ATTENTION

关于轴承相关的项目之前做的大都是故障识别诊断类型的，少有涉及回归预测的，周末的时候宅家发现一个轴承寿命加速实验的数据集就想着拿来做一下寿命预测。

首先看下数据集如下：

直接百度即可搜到，这里就不再赘述了。

Learning_set为训练集

Test_set为测试集

我这里为了简单处理直接使用Learning_set作为总数据集，随机划分指定比例作为测试集。

当然了你也可以选择分别读取加载两部分的数据分别作为训练集和测试集都可以的。

每个目录下都是一堆csv文件，样例如下：

样例数据内容如下：

9,11,19,1.1879e+05,0.059,-0.3729,11,19,1.1883e+05,0.603,-0.0859,11,19,1.1887e+05,0.613,0.1129,11,19,1.1891e+05,0.465,0.1859,11,19,1.1894e+05,-0.216,-0.2569,11,19,1.1898e+05,-0.806,0.1779,11,19,1.1902e+05,-0.653,0.1139,11,19,1.1906e+05,-0.007,0.3989,11,19,1.191e+05,0.888,0.1459,11,19,1.1914e+05,1.037,-0.5429,11,19,1.1918e+05,0.299,-0.2019,11,19,1.1922e+05,-0.552,-0.0229,11,19,1.1926e+05,-1.237,0.2649,11,19,1.193e+05,-1.059,0.1559,11,19,1.1934e+05,-0.269,0.1639,11,19,1.1937e+05,0.662,0.2699,11,19,1.1941e+05,0.949,0.0789,11,19,1.1945e+05,0.403,-0.0659,11,19,1.1949e+05,-0.279,-0.4119,11,19,1.1953e+05,-0.856,0.0339,11,19,1.1957e+05,-0.736,0.2019,11,19,1.1961e+05,0.098,0.3269,11,19,1.1965e+05,0.718,-0.1839,11,19,1.1969e+05,0.61,-0.0389,11,19,1.1973e+05,0.201,0.0929,11,19,1.1976e+05,-0.3,0.019,11,19,1.198e+05,-0.378,0.4479,11,19,1.1984e+05,0.149,-0.1899,11,19,1.1988e+05,0.499,-0.4219,11,19,1.1992e+05,0.325,0.0249,11,19,1.1996e+05,-0.265,0.499,11,19,1.2e+05,-0.708,0.4879,11,19,1.2004e+05,-0.443,0.1579,11,19,1.2008e+05,-0.042,-0.4379,11,19,1.2012e+05,0.238,-0.0259,11,19,1.2016e+05,0.46,0.1939,11,19,1.202e+05,0.192,0.0369,11,19,1.2023e+05,-0.093,0.1189,11,19,1.2027e+05,-0.344,0.1489,11,19,1.2031e+05,-0.174,0.1179,11,19,1.2035e+05,-0.029,-0.0269,11,19,1.2039e+05,0.026,0.469

关于数据处理可以直接使用官方提供的代码：

%% 批量读取IEEE PHM 2012轴承全寿命数据clcclear allclose all%% 文件夹路径file_path ='Learning_set/';%% 全寿命振动信号csv_acc_path_list = dir(strcat(file_path,'acc*.csv'));csv_acc_num = length(csv_acc_path_list);%获取文件总数量if csv_acc_num > 0 %有满足条件的文件for j = 1:csv_acc_num %逐一读取文件csv_acc_name = csv_acc_path_list(j).name;% 文件名csv_acc =csvread(strcat(file_path,csv_acc_name));csv_acc_data(:,:,j)=csv_acc;fprintf('%d %d %s\n',csv_acc_num,j,strcat(file_path,csv_acc_name));% 显示正在处理的文件名endend% 合并矩阵 时间*通道channel=6; %信号的通道数csv_acc_data_change=permute(csv_acc_data,[2 1 3]);csv_acc_data=reshape(csv_acc_data_change,channel,prod(size(csv_acc_data))/channel)';%% 全寿命温度信号csv_temp_path_list = dir(strcat(file_path,'temp*.csv'));%获取该文件夹中所有csv格式的文件csv_temp_num = length(csv_temp_path_list);%获取文件总数量delimiter = ',';formatSpec = '%s%s%s%s%s%s%[^\n\r]';if csv_temp_num > 0 %有满足条件的文件for j = 1:csv_temp_num %逐一读取文件csv_temp_name = csv_temp_path_list(j).name;% 文件名csv_temp_fileID = fopen(strcat(file_path,csv_temp_name),'r');csv_temp = textscan(csv_temp_fileID, formatSpec, 'Delimiter', delimiter);for i=1:size(csv_temp{1,1},1)csv_temp_data(i,:,j)=str2num(csv_temp{1,1}{i,1})';endfprintf('%d %d %s\n',csv_temp_num,j,strcat(file_path,csv_temp_name));% 显示正在处理的文件名fclose(csv_temp_fileID);endend% 合并矩阵 时间*通道channel=5; %信号的通道数csv_temp_data_change=permute(csv_temp_data,[2 1 3]);csv_temp_data=reshape(csv_temp_data_change,channel,prod(size(csv_temp_data))/channel)';%% 全寿命振动信号和温度信号的时域图clearvars -except csv_acc_data csv_temp_datafigure;subplot 211;plot(csv_acc_data(:,5));title('水平振动信号'); subplot 212;plot(csv_acc_data(:,6));title('竖直振动信号');figure;plot(csv_temp_data(:,5));title('温度信号')

这里我不是很懂数据背景所以就不多解释了，有懂行的可以交流一下。

为了直观分析数据这里对其进行了可视化如下：