GAN-1D

using WGAN to generate fault bearing vibration signals

request:

python 3.5+

tensorflow-gpu

numpy scipy os

open cmd and cd to the folder

$ python train.py 

--learning_rate 0.000001 #change the learning rate,default 0.000001

--epoch 2000000 #how much epochs to train,default 2000000

--sample_rate 50000 #how many epochs you want to sample once,default 50000

--train_data x1 #there 9 kinds of signals you can choose,default x1

--train_times 3 #default 3, you can try 4 or 5

your model will be saved at ./checkpoint/ per 100000 epochs

$ python test.py 

test you model, output will be saved at ./output/

It takes about 11 hours to run 2000000 epochs in Titan XP.

signals are from here http://csegroups.case.edu/bearingdatacenter/home

its difficult to train this model, please try different learning rate

if you want to train with your own data, please normalize it