def dataset_prepare_SU(self, sample_num, sample_len):
"""
here 20 testing-case-level sub-sets will be generated and grepped into two datasets, including:
1. SU_1D_Bearing_DataSet.npz
2. SU_1D_Gear_Dataset.npz
the 2 datasets will be prepared for one kind of transferring scenario:
working condition transferring
"""
logging.info(
"generating new dataset with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
filenames, filelabels = self.filename_SU()
os.chdir(self.path_npz)
data, labels = [[], []], [[], []]
for k, files in enumerate(filenames):
for l, file in enumerate(files):
signal = np.load(file)["arr"]
if len(signal.shape) != 2:
raise ValueError(
"excepted signal has 2 dims, but got %d dim(s)" %
len(signal.shape))
if np.shape(signal)[0] != 8:
raise ValueError(
"excepted signal's channels' number is 8 , but got %d"
% np.shape(signal)[0])
samples = ut.signal_split_as_samples(signal, sample_len,
sample_num)
data[k].append(samples)
sample_labels = np.ones((sample_num, 1)) * filelabels[k][l]
if not k:
sample_labels = ut.onehot_encoding(sample_labels, 5)
" fix me "
# considering the combination of inner and outer fault
# if "comb" in file:
# sample_labels[:, 3] = 1
else:
sample_labels = ut.onehot_encoding(sample_labels, 5)
labels[k].append(sample_labels)
data_bearing, data_gear = np.array(data[0]), np.array(data[1])
labels_bearing, labels_gear = np.array(labels[0]), np.array(labels[1])
os.chdir(self.path_cache)
np.savez("SU_1D_Bearing_DataSet.npz",
data=data_bearing,
labels=labels_bearing)
np.savez("SU_1D_Gear_DataSet.npz", data=data_gear, labels=labels_gear)
os.chdir(self.path)
logging.info("new dataset has been stored")
return 0
def dataset_prepare_CWRU(self, sample_num, sample_len):
"""
this function first use [data_clean_CWRU] to read signals from .text file
then samples examples (.shape = sample_num, channels, sample_len) from each signal
40 (actually 48, last 8 is repeated sampled from normal cases) sub-sets will be generate
and then be packaged into one dataset.
this dataset will be used for 2 transferring scenarios, including:
working condition transferring
fault extent transferring
"""
logging.info(
"generating new dataset with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
signals, filelabels_wc, filelabels_fs = self.data_clean_CWRU()
data, labels_wc, labels_fs = [], [], []
for k, signal in enumerate(signals):
data.append(
ut.signal_split_as_samples(signal, sample_len, sample_num))
sample_labels_wc = np.ones((sample_num, 1)) * filelabels_wc[k]
sample_labels_wc = ut.onehot_encoding(sample_labels_wc,
class_num=10)
labels_wc.append(sample_labels_wc)
sample_labels_fs = np.ones((sample_num, 1)) * filelabels_fs[k]
sample_labels_fs = ut.onehot_encoding(sample_labels_fs,
class_num=4)
labels_fs.append(sample_labels_fs)
for i in range(2):
for signal in signals[-4:]:
data.append(
ut.signal_split_as_samples(signal, sample_len, sample_num))
sample_labels_wc = np.zeros((sample_num, 1))
sample_labels_wc = ut.onehot_encoding(sample_labels_wc,
class_num=10)
labels_wc.append(sample_labels_wc)
sample_labels_fs = np.zeros((sample_num, 1))
sample_labels_fs = ut.onehot_encoding(sample_labels_fs,
class_num=4)
labels_fs.append(sample_labels_fs)
data, labels_wc, labels_fs = np.array(data), np.array(
labels_wc), np.array(labels_fs)
os.chdir(self.path_cache)
np.savez("CWRU_1D_signal.npz",
data=data,
labels_wc=labels_wc,
labels_fs=labels_fs)
os.chdir(self.path)
logging.info("new dataset has been stored")
return 0
def normal_set_prepare_CWRU(self, sample_num, sample_len):
logging.info(
"generating normal set with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
os.chdir(self.path_txt)
files, filelabesl_wc, filelabels_fs = self.filename_CWRU()
signals = []
for file in files:
if "normal" not in file:
continue
try:
with open(file, "r") as fr:
signal = fr.readlines()
signal = np.array([float(item.strip()) for item in signal])
except IOError:
raise IOError("some wrong with the file %s" % file)
signals.append(signal.reshape(1, -1))
os.chdir(self.path)
data, labels_wc, labels_fs = [], [], []
for k, signal in enumerate(signals):
data.append(
ut.signal_split_as_samples(signal, sample_len, sample_num))
sample_labels_wc = np.ones((sample_num, 1)) * 0
sample_labels_wc = ut.onehot_encoding(sample_labels_wc,
class_num=10)
labels_wc.append(sample_labels_wc)
sample_labels_fs = np.ones((sample_num, 1)) * 0
sample_labels_fs = ut.onehot_encoding(sample_labels_fs,
class_num=4)
labels_fs.append(sample_labels_fs)
data, labels_wc, labels_fs = np.array(data), np.array(
labels_wc), np.array(labels_fs)
os.chdir(self.path_cache)
np.savez("CWRU_1D_normal_signal.npz",
data=data,
labels_wc=labels_wc,
labels_fs=labels_fs)
os.chdir(self.path)
logging.info("new normal set has been stored")
return 0
def dataset_prepare_MFPT(self, sample_num, sample_len):
"""
this function first use [data_clean_MFPT] to read signals from .txt file
then samples examples (.shape = sample_num, channels, sample_len) from each signal
15 (actually 18, last 3 is repeated sampled from normal cases) sub-sets will be generate
and then be packaged into one dataset.
this dataset will be used for 1 transferring scenario:
working condition transferring
"""
logging.info(
"generating new dataset with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
signals, filelabels = self.data_clean_MFPT()
data, labels, = [], []
for k, signal in enumerate(signals):
data.append(
ut.signal_split_as_samples(signal, sample_len, sample_num))
sample_labels_wc = np.ones((sample_num, 1)) * filelabels[k]
sample_labels_wc = ut.onehot_encoding(sample_labels_wc,
class_num=3)
labels.append(sample_labels_wc)
for signal in signals[-3:]:
data.append(
ut.signal_split_as_samples(signal, sample_len, sample_num))
sample_labels_wc = np.zeros((sample_num, 1))
sample_labels_wc = ut.onehot_encoding(sample_labels_wc,
class_num=3)
labels.append(sample_labels_wc)
data, labels = np.array(data), np.array(labels)
os.chdir(self.path_cache)
np.savez("MFPT_1D_Bearing_DataSet.npz", data=data, labels=labels)
os.chdir(self.path)
logging.info("new dataset has been stored")
return 0
def normal_set_prepare_SU(self, sample_num, sample_len):
logging.info(
"generating normal set with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
filenames, filelabels = self.filename_SU()
os.chdir(self.path_npz)
data, labels = [[], []], [[], []]
for k, files in enumerate(filenames):
for l, file in enumerate(files):
if "health" not in file and "Health" not in file:
continue
signal = np.load(file)["arr"]
if len(signal.shape) != 2:
raise ValueError(
"excepted signal has 2 dims, but got %d dim(s)" %
len(signal.shape))
if np.shape(signal)[0] != 8:
raise ValueError(
"excepted signal's channels' number is 8 , but got %d"
% np.shape(signal)[0])
samples = ut.signal_split_as_samples(signal, sample_len,
sample_num)
data[k].append(samples)
sample_labels = np.ones((sample_num, 1)) * 0
sample_labels = ut.onehot_encoding(sample_labels, 5)
labels[k].append(sample_labels)
data_bearing, data_gear = np.array(data[0]), np.array(data[1])
labels_bearing, labels_gear = np.array(labels[0]), np.array(labels[1])
os.chdir(self.path_cache)
np.savez("SU_1D_Healthy_Bearing_DataSet.npz",
data=data_bearing,
labels=labels_bearing)
np.savez("SU_1D_Healthy_Gear_DataSet.npz",
data=data_gear,
labels=labels_gear)
os.chdir(self.path)
logging.info("new normal set has been stored")
return 0
def normal_set_prepare_MFPT(self, sample_num, sample_len):
logging.info(
"generating normal set with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
os.chdir(self.path_txt)
files, filelabels = self.filename_MFPT()
signals = []
for i in range(
2
): # to match six different working conditions, we have 3 normal bearings, sampling twice
for file in files:
if "normal" not in file:
continue
try:
with open(file, "r") as fr:
signal = fr.readlines()
signal = np.array(
[float(item.strip()) for item in signal])
except IOError:
raise IOError("some wrong with the file %s" % file)
signal = signal[::2]
signals.append(signal.reshape(1, -1))
os.chdir(self.path)
data, labels, = [], []
for k, signal in enumerate(signals):
data.append(
ut.signal_split_as_samples(signal, sample_len, sample_num))
sample_labels_wc = np.ones((sample_num, 1)) * 0
sample_labels_wc = ut.onehot_encoding(sample_labels_wc,
class_num=3)
labels.append(sample_labels_wc)
data, labels = np.array(data), np.array(labels)
os.chdir(self.path_cache)
np.savez("MFPT_1D_Normal_Bearing_DataSet.npz",
data=data,
labels=labels)
os.chdir(self.path)
logging.info("new normal aset has been stored")
return 0
def dataset_prepare_PU(self, sample_num, sample_len):
"""
here 32 testing-case-level sub-sets will be generated and grepped into one dataset
this dataset will be prepared for two kind of transferring scenarios, including:
working condition transferring
artificial-to-real transferring
"""
logging.info(
"generating new dataset with each case has (sample_num = %d, sample_len = %d)"
% (sample_num, sample_len))
# get filenames and the labels corresponding to each file
filenames, filelabels = self.filename_PU()
# read each file, sampling examples from original vibration signals
os.chdir(self.path_npz)
sample_num = int(sample_num / 20)
data, labels = [], []
for k, file in enumerate(filenames):
signals = np.load(file)["arr"]
if len(signals.shape) != 2:
raise ValueError(
"excepted signals have 2 dims, but got %d dim(s)" %
len(signals.shape))
if np.shape(signals)[0] != 80:
raise ValueError(
"excepted signals' number is 80 , but got %d" %
np.shape(signals)[0])
num, length = signals.shape
signals = signals.reshape(num, 1, length)
samples = []
for i in range(num):
samples.append(
ut.signal_split_as_samples(signals[i], sample_len,
sample_num))
data.append(
np.array(samples).reshape(4, 20 * sample_num, 1, sample_len))
sample_labels = np.ones((80 * sample_num, 1)) * filelabels[k]
sample_labels = ut.onehot_encoding(sample_labels, 3)
sample_labels = sample_labels.reshape(4, 20 * sample_num, -1)
# considering the combination of inner and outer fault
if "Inner_Outer" in file:
sample_labels[:, :, 1] = 0.8
sample_labels[:, :, 2] = 0.2
if "Outer_Inner" in file:
sample_labels[:, :, 1] = 0.2
sample_labels[:, :, 2] = 0.8
labels.append(sample_labels)
data, labels = np.array(data), np.array(labels)
os.chdir(self.path_cache)
np.savez("PU_1D_signal.npz", data=data, labels=labels)
os.chdir(self.path)
logging.info("new dataset has been stored")
return 0