Page 91 - 摩擦学学报2025年第8期

第 8 期 王大刚, 等: 钢丝绳摩擦损伤定量识别研究 1189

Pool3 Inception 1
Input_layer maxPooling2dL...
image input layer
Conv1
convolution 2dL... Inception_1b-... Inception_1c-...
convolution 2dL...
convolution 2dL...
Batch_norm1 Maxpool
batch normaliza... Inception_1b-r... Inception_1c-r... Inception_1d-...
relu layer relu layer max pooling 2dL... max pooling 2dL...
Relu1 Inception_1a... Inception_1b... Inception_1c... Relu_1
relu layer convolution 2dL... convolution 2dL... convolution 2dL... Inception_1d-... relu layer
convolution 2dL...
Pool1 Inception_1a-r Inception_1d-r... Batchnorm
max pooling 2dL... relu layer Inception_1b-r Inception_1c-r relu layer batch normaliza...
relu layer
relu layer
Conv2 Gapool
convolution 2dL... global average...
Inception_1-o...
depth concaten...
Relu2 Dropout
relu layer dropout layer
Inception 2
Pool2 Fc
max pooling 2dL fully connected...
Pool3-3×3_s2
max pooling 2dL...
Conv2 Regressionout...
convolution 2dL... regression layer
Inception 3
Relu3
relu layer
Inception 4
Batch_norm2
batch normaliza...
Inception 5

Fig. 14 IncepRegCNN neural network structure
图 14 IncepRegCNN神经网络结构

表 11 Inception模块1网络层参数
Table 11 Inception Module 1 Network layer parameters
Network layer name Convolution/pooling size Number of filters Stride Fill Activate
Maximum pooling 3 [3, 3] − [2, 2] [0, 1, 0, 1] 28(S)×28(S)×192(C)
Convolution 1a [1, 1] 64 [1, 1] [0, 0, 0, 0] 28(S)×28(S)×64(C)
[1, 1] 96 [1, 1] [0, 0, 0, 0] 28(S)×28(S)×96(C)
Convolution 1b 1
[3, 3] 128 [1, 1] [1, 1, 1, 1] 28(S)×28(S)×128(C)
Convolution 1b 2
[1, 1] 16 [1, 1] [0, 0, 0, 0] 28(S)×28(S)×16(C)
Convolution 1c 1
[5, 5] 32 [1, 1] [2, 2, 2, 2] 28(S)×28(S)×32(C)
Convolution 1c 2
Maximum pooling 1d [3, 3] − [1, 1] [1, 1, 1, 1] 28(S)×28(S)×192(C)
Convolution 1d [1, 1] 32 [1, 1] [0, 0, 0, 0] 28(S)×28(S)×32(C)
Deep series 1 − − − − 28(S)×28(S)×256(C)

4 结论

表 12 不同深度串联层网络参数
Table 12 Network parameters of series a. 漏磁损伤信号部分特征值间存在较强的线性
layers at different depths
相关性，均值与均方根特征值间的线性相关性系数为
Different network layers Activate
Deep series 1 28(S)×28(S)×256(C) 0.98，峰值因数与裕度因数之间的线性相关性系数为
Deep series 2 28(S)×28(S)×480(C)
0.89，波形因数与脉冲因数之间线性相关性系数为0.9，
Deep series 3 14(S)×14(S)×512(C)
小波能量与峰峰值的线性相关性系数高于小波能量
Deep series 4 14(S)×14(S)×512(C)
Deep series 5 14(S)×14(S)×512(C)
与其余特征值间的线性相关性系数且为0.8.