基于EWT-LSTM的工业机器人关节异常检测

蒋沁诚,陶建峰,王洋洋,张宇磊,刘成良

EWT-LSTM based industrial robot joint anomaly detection

Qincheng JIANG,Jianfeng TAO,Yangyang WANG,Yulei ZHANG,Chengliang LIU

表 1 CNN-LSTM-Attention模型的详细信息

Tab.1 Detail of proposed CNN-LSTM-Attention model

层编号 描述 详细信息 输出尺寸 Layer 1 Input Input_shape = (12,6) 6×12 Layer 2 Conv1D filters=128, kernel_size=1, activation=ReLU, padding='same' 128×12 Layer 3 AveragePooling1D pool_size=4 128×3 Layer 4 ZeroPadding1D padding=(1, 0) 128×4 Layer 5 Conv1D filters=128, kernel_size=1, activation=ReLU, padding='same' 128×4 Layer 6 AveragePooling1D pool_size=4 128×1 Layer 7 Dropout 0.2 128×1 Layer 8 Bidirectional Units=32, return_sequences=True 64×1 Layer 9 Dropout 0.2 64×1 Layer 10 Permute (2,1) 1×64 Layer 11 Dense layer_size = 64, activation='softmax' 1×64 Layer 12 Permute (2,1) 64×1 Layer 13 Concatenate [Layer 9, Layer 12] 128×1 Layer 14 Flatten — 128 Layer 15 Dense layer_size = 6, activation='sigmoid' 6