1 Introduction
2 Related Work
3 Spectro Temporal Fusion with CLSTM-Autoencoder
3.1 Spectro Temporal Fusion
3.2 CLSTM-AutoEncoder Based Approach
4 Experimental Studies
Model with input | Accuracy (%) |
---|---|
Mel-Spectrogram with SCR-LSTM [12] | 67.54 |
Mel-Spectrogram with CLSTM-AE | 69.13 |
Temporal gram with SCR-LSTM [12] | 55.33 |
Temporal gram with CLSTM-AE | 65.33 |
Mel-Spectro-tempogram with SCR-LSTM [12] | 90.62 |
Mel-Spectro-tempogram with LSTM-AE | 94.54 |
Input | −6 dB | 0 dB | 6 dB |
---|---|---|---|
Fan | 87.23 | 84.02 | 88.11 |
Pump | 83.65 | 81.59 | 79.87 |
Valve | 82.23 | 80.32 | 82.94 |
Slide Rail | 88.79 | 91.84 | 87.69 |
Framework | AUC(%) | |||||
---|---|---|---|---|---|---|
Toy car | Toy conveyor | Fan | Pump | Slide rail | Valve | |
Conv AE [12] | 69.12 | 60.03 | 52.63 | 60.96 | 76.20 | 53.10 |
ANP [23] | 70.1 | 60.1 | 48.0 | 56.9 | 85.4 | 43.5 |
SCR-LSTM [12] | 69.13 | 66.79 | 65.83 | 72.89 | 84.76 | 66.28 |
Semi supervised Auto Encoder [12] | 87.27 | 90.35 | 78.63 | 80.33 | 78.94 | 80.94 |
Classification based ASD [18] | 82.79 | 80.66 | 85.60 | 82.42 | 65.84 | 56.22 |
Dense AE [12] | 80.79 | 76.43 | 72.03 | 73.06 | 87.08 | 72.16 |
GroupMADE AE [18] | 80.51 | 76.03 | 70.10 | 75.68 | 93.29 | 89.68 |
Glow aff. [21] | 80.1 | 61.0 | 49.6 | 65.7 | 87.8 | 77.7 |
MobileNet V2 [18] | 87.66 | 69.71 | 80.19 | 82.53 | 95.27 | 88.65 |
ResNet [18] | 88.69 | 65.04 | 78.87 | 83.50 | 90.49 | 86.24 |
IDCAE [22] | 91.25 | 72.23 | 81.82 | 88.17 | 86.49 | 84.59 |
STgram+CLSTM-AE | 94.54 | 96.12 | 89.96 | 84.80 | 92.94 | 84.23 |