• Acta Photonica Sinica
  • Vol. 50, Issue 2, 44 (2021)
Yuzhao MA1、2, Ruisong WANG1, and Xinglong XIONG1、2、*
Author Affiliations
  • 1College of Electronic Information and Automation, Civil Aviation University of China, Tianjin300300, China
  • 2Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin300300, China
  • show less
    DOI: 10.3788/gzxb20215002.0206003 Cite this Article
    Yuzhao MA, Ruisong WANG, Xinglong XIONG. Fiber-optic Vibration Signal Recognition Based on BLCD Decomposition and ACO-DBN Network[J]. Acta Photonica Sinica, 2021, 50(2): 44 Copy Citation Text show less
    The overall process of the proposed method
    Fig. 1. The overall process of the proposed method
    Mach-Zehnder system sensing schematic diagram
    Fig. 2. Mach-Zehnder system sensing schematic diagram
    BLCD-IF screening process
    Fig. 3. BLCD-IF screening process
    Deep belief network structure
    Fig. 4. Deep belief network structure
    Measured climbing signal envelope
    Fig. 5. Measured climbing signal envelope
    Comparison of decomposition results of different decomposition methods
    Fig. 6. Comparison of decomposition results of different decomposition methods
    Four typical intrusion signals
    Fig. 7. Four typical intrusion signals
    Comparison of four characteristic parameters
    Fig. 8. Comparison of four characteristic parameters
    Comparison of convergence iteration times of different methods
    Fig. 9. Comparison of convergence iteration times of different methods
    Linewidth of a laserPowerSensor moduleSampling rate of the collectorCollected number
    2 kHz25 mWStandard single mode fiber5×106sample/s100 000
    Table 1. Experimental instruments and related parameters
    Decomposition methodCorrelation coefficientRoot mean square error
    ISC1ISC2ISC3ISC1ISC2ISC3
    LCD0.751 50.513 30.509 50.041 70.041 00.046 4
    BLCD0.763 00.577 50.522 20.040 10.040 40.044 7
    Table 2. Comparison of results of different decomposition methods
    Filtering methodSignalSignal to noise/ dBRoot mean quare error
    Add larger ISCRain21.590.002 3
    Climb7.3720.019 5
    Knock6.9550.032 5
    Wind8.1320.028 3
    IFRain27.09 30.001 2
    Climb12.11 60.011 3
    Knock17.29 80.009 9
    Wind9.7320.007 4
    Table 3. Comparison of the effects of two filtering ways
    FeatureRainClimbKnockWindRecognition rateRecognition time
    DirectΦ40%100%90%46.67%69.17%2.64 s
    Fisher93.33%100%100%83.33%94.17%0.83 s
    Table 4. Recognition results before and after feature dimensionality reduction
    MethodRainClimbKnockWindRecognition rateRecognition time
    BP76.67%(23/30)96.67%(29/30)17.86%(5/30)100%(30/30)73.48%1.499 s
    ACO-BP90%(27/30)86.67%(26/30)80%(24/30)93.33%(28/30)87.50%1.362 s
    DBN76.67%(23/30)100%(30/30)93.33%(28/30)83.33%(25/30)83.33%1.523 s
    ACO-DBN93.33%(28/30)100%(30/30)93.33%(28/30)96.67%(29/30)95.83%0.715 s
    Table 5. Comparison of recognition results of different methods
    Methodknock at one pointknock at two pointsRecognition rateRecognition time
    BP63.33%(19/30)50%(15/30)56.67%1.214 s
    ACO-BP80%(24/30)73.33%(22/30)76.67%1.327 s
    DBN90%(27/30)86.67%(26/30)88.34%1.163 s
    ACO-DBN96.67%(29/30)90%(27/30)93.33%0.815 s
    Table 6. Comparison of multi-point vibration identification results
    Yuzhao MA, Ruisong WANG, Xinglong XIONG. Fiber-optic Vibration Signal Recognition Based on BLCD Decomposition and ACO-DBN Network[J]. Acta Photonica Sinica, 2021, 50(2): 44
    Download Citation