Structural Vibration Data Anomaly Detection Based on Multiple Feature Information Using CNN-LSTM Model

Table 1

Description of data patterns.

No.	Data patterns	Feature description	Quantity
1	Normal	The time response is a normal oscillation curve; the envelope has high symmetry; frequency response presents multiple peaks	3600 (20%)
2	Outlier	The time response has obvious outliers, trend, or drift; the envelope has no symmetry; the frequency response has no obvious peak	3600 (20%)
3	Square	The time response is like a square wave; the envelope is close to a horizontal line; the frequency response shows a single obvious peak	3600 (20%)
4	Missing	Most or all of the time responses are missing, most or all of the envelope values are zero, and the frequency response depends on the part that is not missing	3600 (20%)
5	Minor	The time response is a slight vibration; the envelope has no symmetry; the frequency response has no obvious peak	3600 (20%)