• Infrared and Laser Engineering
  • Vol. 53, Issue 11, 20240294 (2024)
Nianchao LIU1, Qin LI2,3,*, Xiaoting ZHAO1, and Sheng LIANG1
Author Affiliations
  • 1School of Physical Science and Engineering, Beijing Jiaotong University, Beijing 100044, China
  • 2Hebei Key Laboratory of Seismic Disaster Instrument and Monitoring Technology, Langfang 065000, China
  • 3The Third Research Institute of CETC, Beijing 100015, China
  • show less
    DOI: 10.3788/IRLA20240294 Cite this Article
    Nianchao LIU, Qin LI, Xiaoting ZHAO, Sheng LIANG. Cluster-based recognition method for Φ-OTDR system's vibration signals[J]. Infrared and Laser Engineering, 2024, 53(11): 20240294 Copy Citation Text show less
    Schematic diagram of Φ-OTDR
    Fig. 1. Schematic diagram of Φ-OTDR
    Flowchart of identification task
    Fig. 2. Flowchart of identification task
    Flowchart of secondary data cropping
    Fig. 3. Flowchart of secondary data cropping
    (a) Original data graph of manual percussion; (b) Original data graph of machine excavation; (c) Cropped data graph of manual percussion; (d) Cropped data graph of machine excavation; (e) Envelope value graph of manual percussion; (f) Envelope value graph of machine excavation
    Fig. 4. (a) Original data graph of manual percussion; (b) Original data graph of machine excavation; (c) Cropped data graph of manual percussion; (d) Cropped data graph of machine excavation; (e) Envelope value graph of manual percussion; (f) Envelope value graph of machine excavation
    Correlation matrix of raw data
    Fig. 5. Correlation matrix of raw data
    (a) Three-dimensional projection of the original data after dimensionality reduction; (b) Three-dimensional plot of the effect of agglomerative clustering
    Fig. 6. (a) Three-dimensional projection of the original data after dimensionality reduction; (b) Three-dimensional plot of the effect of agglomerative clustering
    Line charts of normalized time-domain characteristics of wind noise. (a) Line chart with 15 characteristic values; (b) Bc and Bi feature values; (c) Bmax, Brm, Ben, Bvar, Bstd, Bff, Bmeans, and Bav feature values
    Fig. 7. Line charts of normalized time-domain characteristics of wind noise. (a) Line chart with 15 characteristic values; (b) Bc and Bi feature values; (c) Bmax, Brm, Ben, Bvar, Bstd, Bff, Bmeans, and Bav feature values
    Line charts of normalized time-domain characteristics for manual knocking. (a) Line chart with 15 characteristic values; (b) Ben, Bff, Bmax, Bav, Bmeans, Bstd, and Brm feature values
    Fig. 8. Line charts of normalized time-domain characteristics for manual knocking. (a) Line chart with 15 characteristic values; (b) Ben, Bff, Bmax, Bav, Bmeans, Bstd, and Brm feature values
    Line charts of normalized time-domain characteristics for machine excavation. (a) Line chart with 15 characteristic values; (b) Feature values of Bmeans and Bav; (c) Feature values of Ben, Bstd, Bff, Brm, and Bmax
    Fig. 9. Line charts of normalized time-domain characteristics for machine excavation. (a) Line chart with 15 characteristic values; (b) Feature values of Bmeans and Bav; (c) Feature values of Ben, Bstd, Bff, Brm, and Bmax
    Correlation matrix of clipped and normalized data
    Fig. 10. Correlation matrix of clipped and normalized data
    Elbow diagrams for agglomerative clustering of cropped normalized data
    Fig. 11. Elbow diagrams for agglomerative clustering of cropped normalized data
    Under K=3 condition (a) 3D projection of cropped normalized data; (b) 3D projection of agglomerative clustering of cropped normalized data
    Fig. 12. Under K=3 condition (a) 3D projection of cropped normalized data; (b) 3D projection of agglomerative clustering of cropped normalized data
    Evaluation coefficients for agglomerative clustering under K =3 condition
    Fig. 13. Evaluation coefficients for agglomerative clustering under K =3 condition
    Confusion matrix for agglomerative clustering under K= 3 condition
    Fig. 14. Confusion matrix for agglomerative clustering under K= 3 condition
    Under the condition of K=2 and truth labels of 0 and 1 (a) 3D projection of cropped normalized data; (b) 3D projection of agglomerative clustering of cropped normalized data
    Fig. 15. Under the condition of K=2 and truth labels of 0 and 1 (a) 3D projection of cropped normalized data; (b) 3D projection of agglomerative clustering of cropped normalized data
    Confusion matrix for agglomerative clustering under the condition of K=4
    Fig. 16. Confusion matrix for agglomerative clustering under the condition of K=4
    (a) Plot of rainfall data; (b) Plot of windblown data; (c) Plot of direct tapping data; (d) Plot of indirect tapping data
    Fig. 17. (a) Plot of rainfall data; (b) Plot of windblown data; (c) Plot of direct tapping data; (d) Plot of indirect tapping data
    Cluster evaluation coefficients of this study and previous studies
    Fig. 18. Cluster evaluation coefficients of this study and previous studies
    No.Time domain featureCharacterization meaning
    1Maximum valueRepresents the peak amplitude of the optical fiber vibration signal
    2Mean valueRepresents the average amplitude of the optical fiber vibration signal
    3VarianceRepresents the power of the optical fiber vibration signal
    4SkewnessRepresents the symmetry of the optical fiber vibration signal
    5Waveform factorRepresents the shape of the optical fiber vibration signal
    6Crest factorRepresents the degree of peak protrusion of the optical fiber vibration signal
    7Impulse factorRepresents the impulse characteristics of the optical fiber vibration signal
    Table 1. Characterization of the significance of the seven time-domain features
    IndicatorsCategoriesDescription
    HomogeneityExternal evaluationMeasures the similarity of data points within a cluster
    CompletenessExternal evaluationMeasures the allocation degree of data points with the same characteristics
    V-measureExternal evaluationComprehensively considers homogeneity and completeness
    Adjustedrand indexExternal evaluationMeasures the consistency between clustering results and true labels in case of randomness
    Adjusted mutual informationExternal evaluationMeasures the correlation between clustering results and true labels
    Silhouette coefficientInternal evaluationMeasures the suitability of the distribution of data points within each cluster
    Table 2. Cluster assessment indicators
    ParameterValue
    Center wavelength of laser/nm1550.92
    Laser linewidth/kHz5
    Modulator modulation pulse width/ns100
    Modulator pulse interval/ms0.02
    Acquisition card sampling rate/106 samples·s−190
    System spatial resolution/m10
    Experimental optical cable length/m2 000
    Table 3. Parameters of devices in experiment
    No.SymbolTime domain characteristics
    1BmaxMaximum value
    2BminMinimum value
    3BmeansAverage value
    4BvarVariance
    5BavStandard deviation
    6BffPeak-to-peak value
    7BavRectified average value
    8BkuKurtosis
    9BskSkewness
    10BrmRoot mean square of the mean
    11BsWaveform factor
    12BcCrest factor
    13BiImpulse factor
    14BlMargin factor
    15BenSignal entropy
    Table 4. Time domain features corresponding to symbols
    EventNumber of input samplesNumber of cluster samplesTotal accuracy
    Wind noise50450488.68%
    Manual knock548543
    Machine excavation591596
    Table 5. Number of input samples and number of clusters for various events
    K setting valueInput sample sizeNumber of samples correctly clusteredAccuracy
    216431643100%
    31643145388.68%
    41643125076.26%
    Table 6. Number of input samples, number of samples correctly clustered and accuracy under different K conditions
    K setting valueNumber of correctly clustered samples for wind noiseNumber of correctly clustered samples for manual knockNumber of correctly clustered samples for excavation
    25041139
    3504477434
    4504438295
    Table 7. Number of samples for three correct clusters under different K conditions
    Nianchao LIU, Qin LI, Xiaoting ZHAO, Sheng LIANG. Cluster-based recognition method for Φ-OTDR system's vibration signals[J]. Infrared and Laser Engineering, 2024, 53(11): 20240294
    Download Citation