CN117746842A - Feature extraction and simulation method for sound signals of autonomous navigation ship tunnel propeller - Google Patents

Abstract

The invention relates to a feature extraction and simulation method for the sound signal of an autonomous navigation ship tunnel thruster. The measurement signal is processed using a digital filter to remove peak signals and the characteristics of the residual broadband noise floor signal are analyzed. By designing a finite impulse response filter and convolving its impulse response with Gaussian distributed white noise, the frequency domain characteristics of the simulated noise and the base noise signal are matched. Calculate the peak signal amplitude and generate a set of sine wave signals with corresponding frequencies as the spike signal. The narrow-band spike signal and the broadband base signal are superimposed and synthesized into simulated tunnel thruster noise. It provides a feasible solution for the research and analysis of tunnel propeller sound signals. By accurately extracting the sound characteristics of tunnel thrusters and improving the predicted sound level, a tunnel thruster sound source database is established to provide valuable information for signal quality analysis, fault diagnosis and improvement of tunnel thruster control strategies for ship autonomous navigation systems. data support.

Description

Feature extraction and simulation method of sound signals from tunnel thrusters of autonomous navigation ships

Technical field

The invention relates to a signal processing and simulation technology, and in particular to a feature extraction and simulation method for the sound signal of an autonomous navigation ship tunnel thruster.

Background technique

Unmanned autonomous ship navigation technology is one of the key technologies for smart ships. Autonomous navigation ships need to navigate and operate in complex marine environments, so they have strict requirements on their maneuverability and reliability. The tunnel thruster is installed in a transverse tunnel under the waterline at the bow or stern of the ship and mainly provides thrust lateral to the hull during navigation. The analysis, modeling and prediction of tunnel thruster sound signals can provide important basis for signal quality analysis, thruster working condition estimation and fault diagnosis, thereby ensuring that the ship's autonomous navigation system can fully and timely grasp and control the status of the tunnel thruster. .

The sound generated by the tunnel propeller during operation can be mainly divided into mechanical noise, hydrodynamic noise, aerodynamic noise and propeller noise. Propeller noise refers to the changes in the surrounding flow field and pressure caused by the rotation of the propeller, resulting in a variety of noises with different mechanisms; mechanical noise is mainly vibration or vibration caused by the periodic changes in gas pressure and motion inertial force of the moving parts of the propeller during operation. Produced by the impact of each other; hydrodynamic noise is the noise produced by water flowing through the surface of the propeller; aerodynamic noise includes excitation vibration noise caused by air intake and exhaust. Different types of noise have different characteristics in the frequency spectrum. The narrow-band peak signals in the noise energy spectrum mainly come from mechanical noise and propeller noise, and most of them are reflected in the form of harmonics in the entire frequency band. Broadband background noise is a collection of other non-narrowband noise, mainly derived from hydrodynamic noise and aerodynamic noise.

Most of the known sound signal processing methods lack the identification, separation and modeling of tunnel thruster noise from different types of sources. The processing algorithms are also relatively complex and require a large amount of computing resources. At the same time, the simulated sound effects are affected by the accuracy of the analysis. It is relatively large, so there are limitations in analyzing and predicting the sound signals of tunnel thrusters in autonomous navigation systems. Therefore, there is an urgent need to study an efficient and fast feature identification and signal simulation method suitable for ship autonomous navigation systems and targeting the sound of tunnel thrusters.

Contents of the invention

In response to the above problems, a feature extraction and simulation method of the tunnel thruster sound signal of an autonomous navigation ship is proposed to provide feature identification, separation, modeling and prediction of the tunnel thruster sound for the autonomous navigation system.

The technical solution of the present invention is: a feature extraction and simulation method of the sound signal of the tunnel thruster of an autonomous navigation ship. High-pass filtering is performed on the collected sound signal of the tunnel thruster, and the power spectral density of the filtered signal is calculated. The obtained power spectral density After the vector is subjected to minimum value filtering, the spike signal representing the narrow-band noise and the broadband base noise are partially separated on the processed signal; the spike signal in the processed signal is identified and extracted, and the frequency and peak signal amplitude corresponding to the spike signal are obtained. It is used to reconstruct the sinusoidal signal to form a narrow-band peak signal; analyze the residual broadband noise signal characteristics of the processed signal, design a digital filter to simulate a frequency response that matches the power spectrum density of the base signal, and then generate white noise and digital The frequency response of the filter is convolved to generate a simulated broadband base noise signal to form a broadband base signal; the narrow-band peak signal and the broadband base signal are superposed and synthesized into simulated tunnel thruster noise, which is used for signal analysis and prediction of autonomous navigation systems.

Further, the method of collecting the sound signal of the tunnel thruster: place the sound sensor near the tunnel thruster at the bottom of the ship, collect and record the sound signals received at this location; change the working conditions of the tunnel thruster, and adjust the position of the sound sensor and direction, collect sound signals at different locations, and record audio signal data; the initial discrete time domain signal obtained is real sound pressure data, and the analog signal is converted into a discrete digital signal through the digital signal converter in the sound sensor to discretely The digital signal is saved for subsequent signal analysis.

Further, the power spectral density of the filtered signal is calculated: select a sound sample of a section of tunnel thruster under stable operating conditions from a complete section of sampled signal, first cut the sample into segments of smaller length, and analyze each segment Perform a weighting calculation, the weighting function is the Hanning window function, and then calculate the discrete Fourier transform of each segment;

The power spectral density of each segment is where c _w is the _window function _correction coefficient _; Average power spectral density over all time; power spectral density at decibel level is calculated as/>

Further, identify and extract spike signals in the processed signal: compare the minimum filtered power spectrum density vector with the original power spectrum vector, perform a difference operation, and set a threshold to determine the spike signal.

Further, design a digital filter method: Use this power spectral density to design an FIR filter so that the frequency characteristics of the FIR digital filter at discrete frequency points are equal to or close to the values of the broadband background noise signal spectrum at these frequency points. And the characteristics at other frequencies are better approximated; according to Nyquist sampling law, the bilateral frequency response Y(k) of the signal is calculated as m(k) is the power spectral density filtered by the average value in the previous paragraph, and its frequency range is from 0 to f _s /2, where f _s is the signal sampling frequency; first calculate the complex common value of Y(k) along f _s /2 Yoked and symmetrical to f _s /2 to f _s , we get H(k) in the frequency range 0 to f _s . At this time, H(k) is conjugately symmetrical along f _s /2, that is, H(Nk)=conj(H(N+k)). Perform discrete inverse Fourier transform on H(k) from 0Hz to f _s to obtain a finite-length digital sequence impulse response h(n), n=0,1,…,2N-1, as follows:

Finally, the discrete points after h(N+1) are translated to the range of k<0, and the final impulse signal response g(k) of the FIR filter is obtained, that is, for k<0, g(k)=h(k+ 2N), since g(n)=g(-n), the phase of this FIR filter is zero.

Further, the white noise generation method: Gaussian distributed random white noise, its average value is 0, the variance is the sampling frequency of the signal, and the power spectral density of the generated white noise is 1.

The beneficial effects of the present invention are: the feature extraction and simulation method of the sound signal of the tunnel thruster of the autonomous navigation ship provided by the present invention provides a feasible solution for the research and analysis of the sound signal of the tunnel thruster. By accurately extracting the sound characteristics of tunnel thrusters and improving the predicted sound level, a tunnel thruster sound source database is established to provide valuable information for signal quality analysis, fault diagnosis and improvement of tunnel thruster control strategies for ship autonomous navigation systems. data support.

Description of drawings

Figure 1 is a schematic diagram of the feature extraction and simulation method of the tunnel thruster noise signal according to the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention and provides detailed implementation modes and specific operating procedures. However, the protection scope of the present invention is not limited to the following embodiments.

Feature extraction and simulation method of sound signals from tunnel thrusters of autonomous navigation ships. The measurement signal is processed using a digital filter to remove peak signals and the characteristics of the residual broadband noise floor signal are analyzed. By designing a finite impulse response filter and convolving its impulse response with Gaussian distributed white noise, the frequency domain characteristics of the simulated noise and the base noise signal are matched. Calculate the peak signal amplitude and generate a set of sine wave signals with corresponding frequencies as the spike signal. The superposed signal of the spike signal and the base signal is used as the simulated predicted tunnel thruster sound signal.

As shown in Figure 1, the embodiment of the present invention provides a feature extraction and simulation method for a tunnel thruster noise signal model. The collected tunnel thruster sound signal is subjected to high-pass filtering, and the filtered signal is subjected to power spectral density calculation. The obtained After the power spectral density vector is subjected to minimum value filtering, the processed signal is partially separated into the spike signal representing the narrow-band noise and the broadband base noise; the spike signal in the processed signal is identified and extracted, and the frequency and spike signal corresponding to the spike signal are obtained. Amplitude is used to reconstruct the sinusoidal signal to form a narrow-band peak signal; analyze the characteristics of the residual broadband background noise signal on the processed signal, and design a digital filter to simulate a frequency response that matches the power spectrum density of the background signal, and then generate white The noise and the frequency response of the digital filter are convolved to generate a simulated wide-band base noise signal to form a wide-band base signal; the narrow-band peak signal and the wide-band base signal are superimposed and synthesized into simulated tunnel thruster noise, which is used for signal analysis and analysis of autonomous navigation systems. predict.

For acoustic signal acquisition, an acoustic sensor is placed near the tunnel thruster on the bottom of the ship. Collect and record the sound signals received at this location. Change the working conditions of the tunnel thruster, adjust the position and direction of the sound sensor, collect sound signals at different locations, and record audio signal data for subsequent analysis of model signals. The initial discrete time domain signal obtained is the real sound pressure data. Through the digital signal converter in the sound sensor, the analog signal can be converted into a discrete digital signal x[n]. The sound signal is saved as a discrete digital signal and used for subsequent analysis of the model signal.

Receive a section of the actual measured tunnel thruster audio signal that will be analyzed, and perform high-pass filtering on the signal. The cutoff frequency is generally less than 20Hz, thereby removing low-frequency signal interference that cannot be heard by human ears in the measurement environment.

The discrete Fourier transform is used to perform power spectral density analysis on the first filtered signal. The specific method is: select a stable and representative sample from a complete section of sampling signal (the tunnel thruster is under stable operating conditions). The method of power spectral density analysis is to first truncate the sample into segments of smaller length, perform a weighted calculation on each segment, the weighting function is the Hanning window function, and then calculate the discrete Fourier transform of each segment .

The power spectral density of each segment is Among them _, c _w is the _window function correction coefficient _; Finally, the all-time average power spectral density of the sample is calculated. The power spectral density in decibels is calculated as/> After calculation, the relationship between signal power spectral density and frequency is obtained.

Separate the spike signal representing narrowband noise and the broadband noise floor portion. The power spectral density of the obtained sample (that is, the power spectral density of the previously calculated decibel level) is regarded as a one-dimensional vector along the frequency domain direction, and minimum value filtering is performed to remove narrow-frequency peaks. Moving minimum filtering uses a sliding window method to calculate the minimum. A window of the specified length is moved sample-by-sample on the vector and the minimum value of the data in the window is independently calculated, thereby removing peaks. Assume that the window length is 2M+1, and the window is moved to the nth point for minimum value filtering, then the filtered value of the nth point is PL _{minimum value filtering} (n) = min{PL _(nM) , PL _{(n-M +1)} ,…,PL _(n) ,…,PL _(n+M-1) ,PL _(n+M) }. The minimum filter order is related to the resolution of the power spectral density. The window length should be ensured to be slightly larger than the bandwidth of the narrow-frequency peak in the spectrum. In actual design, different filter orders need to be tried multiple times to obtain better smoothing effects. .

Compare the power spectrum density vector after minimum filtering with the original power spectrum vector, perform difference calculation, set a threshold to determine the peak signal, and record the size of the frequency difference of the corresponding frequency. Record the frequencies corresponding to a series of peak signals. Generate a sine function of the corresponding frequency and amplitude based on the calculation results, add a random value phase, and the reconstructed sine signal constitutes a narrow-band spike signal model. Since the Hanning window is used when calculating the power spectrum density, the energy of the spike noise is dispersed. Therefore, when calculating the peak noise amplitude, the energy spectrum difference of several discrete points of the frequency response near the peak frequency in the power spectrum density should be accumulated. For example, for a specific frequency f ₁ , assuming that the sum of the difference between the original frequency response and the filtered frequency response of the discrete points near it is A, the expression of the sine function is Where Δ is 1/f _s , i.e. the inverse of the signal sampling frequency; n is the serial number of the discrete signal, n = 0, 1, 2…;/> is the random phase of the trigonometric function, with a random value ranging from 0 to π. For different n in y(n), a phase angle is randomly selected.

After minimum filtering, the power spectral density vector y[n] = PL _{minimum filtering} (n) continues to perform sliding average filtering on the residual broadband base signal, further smoothing the power spectral density curve of the signal. Assuming that the window length is 2K+1 (the typical value of the window length is K=1 or 2), the output of the symmetric moving average filter is where the sum of b _j is 1. For the selection of the one-dimensional vector b, the average or Hanning window distribution can be used. The moving average filter is a low-pass filter that supplements the pre-filter. Its purpose is to reduce the impact of random interference. The order of the moving average filter is usually small, ensuring that the power spectrum is retained while removing sample deviations. Density details. The specific order requires multiple attempts in actual design.

At this time, a broadband background noise signal is obtained. Use this power spectral density to design the FIR filter so that the frequency characteristics of the FIR digital filter at discrete frequency points are equal to or close to the values of the broadband background noise signal spectrum at these frequency points (the broadband background noise signal spectrum, that is, the upper A signal that has been processed by sliding average filtering. This signal is the relationship between power spectral density and frequency, that is, there is a calculated power spectral density value at each discrete frequency point (such as 1Hz, 2Hz, 3Hz...) , and the characteristics at other frequencies are better approximated. Digital filters with finite duration impulse responses have the following major advantages: Have precise linear phase. Always stable. Design methods are often linear. They can be implemented efficiently in hardware. According to the Nyquist sampling law, the bilateral frequency response Y(k) of the signal is calculated as m(k) is the average filtered power spectral density in the previous paragraph, and its frequency range is 0 to f _s /2, where f _s is the signal sampling frequency. To calculate the impulse response of the FIR filter, first calculate the complex conjugate of Y(k) along f _s /2 and symmetry to f _s /2 to f _s to obtain H(k) in the frequency range from 0 to f _s . At this time, H(k) is conjugately symmetrical along f _s /2, that is, H(Nk)=conj(H(N+k)). Perform discrete inverse Fourier transform on H(k) from 0Hz to f _s to obtain a finite-length digital sequence impulse response h(n), n=0,1,…,2N-1, as follows:

Finally, the discrete points after h(N+1) are translated to the range of k<0, and the final impulse signal response g(k) of the FIR filter is obtained. That is, for k<0, g(k)=h(k+2N). Since g(n)=g(-n), this FIR filter has zero phase. After the broadband background noise signal passes through the designed FIR filter, the broadband base noise frequency response is obtained.

In order to generate a simulated broadband background noise signal for a period of time, a period of Gaussian distributed random white noise needs to be generated first, with an average value of 0 and a variance of the sampling frequency of the signal. The power spectral density of the white noise generated at this time is 1. Then the white noise and the frequency response of the FIR filter output are convolved to artificially generate a simulated broadband base noise signal as background noise. Since the frequency response of the impulse response does not change after passing through the white noise, its power spectral density is consistent with the actually measured broadband background noise signal.

The artificially simulated tunnel thruster audio signal is obtained by adding a set of sinusoidal functions representing narrow-band signals and the simulated broadband background noise signal.

After obtaining the simulated tunnel thruster sound signal, you can establish a tunnel thruster sound source database, evaluate the quality of the signal and improve the shipboard operating environment, analyze the relationship between the mechanical structural characteristics and acoustic characteristics of the tunnel thruster, or use the frequency domain relationship To analyze abnormal sounds caused by mechanical failures. This analysis can help detect tunnel thruster failures and take appropriate maintenance measures, thereby improving the reliability and efficiency of tunnel thrusters.

The above-mentioned embodiments only express several implementation modes of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention. It should be noted that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the scope of protection of the patent of the present invention should be determined by the appended claims.

Claims (6)

1. A method for feature extraction and simulation of tunnel thruster sound signals for autonomous navigation ships, which is characterized in that the collected tunnel thruster sound signals are subjected to high-pass filtering, and the power spectral density of the filtered signal is calculated to obtain the power spectral density vector After performing minimum value filtering, the processed signal is partially separated from the peak signal representing narrow-band noise and the broadband base noise; identify and extract the peak signal in the processed signal, obtain the frequency and peak signal amplitude corresponding to the peak signal, and use The reconstructed sinusoidal signal forms a narrow-band spike signal; the processed signal is analyzed for the characteristics of the residual broadband noise signal, and a digital filter is designed to simulate a frequency response that matches the power spectrum density of the base signal, and then generates white noise and digital filtering The frequency response of the transmitter is convolved to generate a simulated broadband base noise signal to form a broadband base signal; the narrow-band peak signal and the broadband base signal are superposed and synthesized into simulated tunnel thruster noise, which is used for signal analysis and prediction of autonomous navigation systems. 2. According to claim 1, the feature extraction and simulation method of the sound signal of the tunnel thruster of an autonomous navigation ship is characterized in that the method of collecting the sound signal of the tunnel thruster is as follows: placing a sound sensor near the tunnel thruster at the bottom of the ship to collect and record the sound signal received at this position; changing the working condition of the tunnel thruster, collecting sound signals at different positions by adjusting the position and direction of the sound sensor, and recording the audio signal data; the obtained initial discrete time domain signal is real sound pressure data, and the analog signal is converted into a discrete digital signal through a digital signal converter in the sound sensor, and saved in the form of a discrete digital signal for subsequent signal analysis. 3. The feature extraction and simulation method of the sound signal of the tunnel thruster of the autonomous navigation ship according to claim 1, characterized in that the filtered signal is subjected to power spectral density calculation: a section of the tunnel thruster is selected from a complete section of the sampled signal and is in a stable state. For sound samples under operating conditions, the samples are first truncated into segments of smaller length, and weighted calculations are performed on each segment. The weighting function is the Hanning window function, and then the discrete Fourier transform of each segment is calculated; The power spectral density of each segment is where c _w is the _window function _correction coefficient _; Average power spectral density over all time; power spectral density at decibel level is calculated as 4. The feature extraction and simulation method of the sound signal of the tunnel thruster of the autonomous navigation ship according to claim 1, characterized in that, to identify and extract the peak signal in the processed signal: the power spectral density vector processed by the minimum value filtering and the original Compare power spectrum vectors, perform difference operations, and set thresholds to determine spike signals. 5. The feature extraction and simulation method of the sound signal of the tunnel thruster of the autonomous navigation ship according to claim 3, characterized in that the digital filter design method: using this power spectral density to design the FIR filter, so that the FIR digital filter The value of the frequency characteristics at discrete frequency points is equal to or close to the value of the broadband background noise signal spectrum at these frequency points, and the characteristics at other frequencies are better approximated; according to the Nyquist sampling law, the bilateral frequency of the signal Response Y(k), its calculation method is m(k) is the power spectral density filtered by the average value in the previous paragraph, and its frequency range is from 0 to f _s /2, where f _s is the signal sampling frequency; first calculate the complex common value of Y(k) along f _s /2 Yoked and symmetrical to f _s /2 to f _s , we get H(k) in the frequency range 0 to f _s . At this time, H(k) is conjugately symmetrical along f _s /2, that is, H(Nk)=conj(H(N+k)). Perform discrete inverse Fourier transform on H(k) from 0Hz to f _s to obtain a finite-length digital sequence impulse response h(n), n=0,1,…,2N-1, as follows: Finally, the discrete points after h(N+1) are translated to the range of k<0, and the final impulse signal response g(k) of the FIR filter is obtained, that is, for k<0, g(k)=h(k+ 2N), since g(n)=g(-n), the phase of this FIR filter is zero. 6. The feature extraction and simulation method of the sound signal of the tunnel thruster of the autonomous navigation ship according to claim 5, characterized in that the white noise generation method: Gaussian distributed random white noise, the average value is 0, and the variance is the sampling frequency of the signal , the power spectral density of the generated white noise is 1.