Research Article

Fault Diagnosis of Reciprocating Compressor Based on Convolutional Neural Networks with Multisource Raw Vibration Signals

Table 1

The network structure of MSRCNN.

Layer (type)Output ShapeParam #

input_1 (InputLayer)(None, 1024, 3, 1)0
conv1 (Conv2D)(None, 1024, 3, 8)80
bn1 (BatchNormalization)(None, 1024, 3, 8)32
activation_1 (Activation relu)(None, 1024, 3, 8)0
max_pool1 (MaxPooling2D)(None, 512, 3, 8)0
conv2 (Conv2D)(None, 512, 3, 16)3216
bn2 (BatchNormalization)(None, 512, 3, 16)64
activation_2 (Activation relu)(None, 512, 3, 16)0
max_pool2 (MaxPooling2D)(None, 256, 3, 16)0
conv3 (Conv2D)(None, 256, 3, 32)4640
bn3 (BatchNormalization)(None, 256, 3, 32)128
activation_3 (Activation relu)(None, 256, 3, 32)0
max_pool3 (MaxPooling2D)(None, 128, 3, 32)0
conv4 (Conv2D)(None, 128, 3, 64)2112
bn4 (BatchNormalization)(None, 128, 3, 64)256
activation_4 (Activation relu)(None, 128, 3, 64)0
conv5 (Conv2D)(None, 128, 3, 32)18464
bn5 (BatchNormalization)(None, 128, 3, 32)128
activation_5 (Activation relu)(None, 128, 3, 32)0
max_pool5 (MaxPooling2D)(None, 64, 3, 32)0
conv6 (Conv2D)(None, 64, 3, 16)4624
bn6 (BatchNormalization relu)(None, 64, 3, 16)64
activation_6 (Activation relu)(None, 64, 3, 16)0
max_pool6 (MaxPooling2D)(None, 32, 3, 16)0
flatten_1 (Flatten)(None, 1536)0
fc1 (Dense)(None, 128)196736
fc2 (Dense)(None, 32)4128
fc3 (Dense)(None, 4)132