A Fault Prediction and Cause Identification Approach in Complex Industrial Processes Based on Deep Learning

Table 4

Comparison fault detection result by F1-score.

States	PCA	Typical LSTM	Optimized LSTM	SVM using a linear kernel	Auto encoder	The proposed approach in this paper
Fault1	1	0.09	0.68	0.87	0.98	1
Fault2	0.79	0.12	0.78	0.88	0.85	0.89
Fault3	0.34	0.03	0.45	0.79	0.91	0.94
Fault4	0.99	0.04	0.75	0.9	0.89	0.99
Fault5	0.56	0.2	0.89	0.9	0.93	0.94
Fault6	0.99	0.34	1	0.95	0.85	1
Fault7	1	0.19	0.89	0.92	0.80	1
Fault8	0.97	0.22	0.71	0.63	0.73	0.99
Fault9	0.78	0.01	0.67	0.76	0.76	0.81
Fault10	0.66	0.28	0.77	0.89	0.79	0.99
Fault11	0.71	0.12	0.83	0.9	0.77	0.88
Fault12	0.99	0.31	0.56	0.75	0.93	0.99
Fault13	0.87	0.22	0.89	0.82	0.91	0.89
Fault14	0.98	0.41	0.99	0.88	0.78	0.99
Fault15	0.26	0.01	0	0.21	0.01	0.22
Fault16	0.24	0.12	0	0.14	0.33	0.31
Fault17	0.99	0.24	0.2	0.79	0.89	0.97
Fault18	0.78	0.31	0.88	0.66	0.95	0.89
Fault19	0.88	0.17	0.91	0.86	0.64	0.97
Fault20	0.82	0.28	0.64	0.78	0.83	0.85