JP2021102429A - Extraction method for propeller blade number characteristic based on radiation noise modulation - Google Patents

Abstract

To provide a method for extracting characteristic amount information of a propeller axis frequency, a blade frequency, and a propeller blade number by a voice sensor for determining a vessel type in a simple operation.SOLUTION: An extraction method includes: (1) a step of collecting radiation noise signals of a vessel and acquiring a modulation spectrum diagram by Fourier transformation; (2) a step S01 of searching for a local peak in a modulation diagram and acquiring a resonance frequency in a local peak position; (3) a step S02 of determining the resonance frequency in a first local peak in the modulation diagram, determining an axis frequency, and determining the number N of the resonance frequency based on the number of local peaks; (4) a step S03 of determining a multiplication relation between the axis frequency and a blade frequency and determining other resonance frequencies; and (5) a step S04 of determining an average spectrum coherent value for each higher harmonic frequency position, a step S05 of acquiring a propeller blade number by assumption method of naive Bayes, and a step of finally determining the blade frequency.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technical area for extracting frequency components of a signal, and more particularly to a method for extracting propeller blade number features based on radiation noise modulation.

Radiation noise of a ship is caused by cavitation noise during rotation of a rotor blade (propeller). The modulation spectrum of the ship's radiation noise includes characteristic information on the number of propeller blades of the ship. By analyzing the relationship between each resonance frequency and amplitude, the shaft frequency of the ship propeller,
The features of the blade frequency and the number of propeller blades can be extracted.

If the feature quantity information of the propeller shaft frequency, k blade frequency, and propeller blade number of a private ship can be obtained, it will contribute to the determination of the ship model number. There are smuggling and stowaways along the sea. If the noise information of the ship is captured by the voice sensor, the modulation spectrum is acquired by the Fourier transform, and the feature quantity information of the shaft frequency, the blade frequency and the number of propeller blades of the propeller can be extracted, the supervisor can judge the type of the ship.

Conventionally, the following identification methods have been used in the world. The modulation spectrum is acquired, and the resonance frequency of the peak of the width value and the peak position is read from the modulation spectrum. Based on the propeller blade number feature identification rule (Table 1), the propeller blade number feature value is extracted from the resonance frequency and amplitude in the modulation spectrum. P (n) indicates the width value of the nth harmonic line spectrum of the axis frequency.

になった場合、表１のすべてのプロペラ翼数識別規則の条件は満たされるため、プロペラ
翼数を識別できない。 However, the configuration of the modulation spectrum is complicated depending on the structure, operating conditions, environment, etc. of the ship, and only a typical case can be identified based on the discrimination rule, and all the modulation spectrum configurations are not applicable. For example

If, the conditions of all the propeller blade number identification rules in Table 1 are satisfied, and the propeller blade number cannot be identified.

Chinese patent application 2019107990217. Regarding the extraction method of rotor characteristics in X,
A multidimensional statistical modeling method for rotor cavitation wake microfeatures has been disclosed, but the extracted features are mainly limited to the geometric parameters of the rotor and the characteristics of the operating conditions, and the characteristics of the shaft frequency, blade frequency, and propeller blade number. Not as good as that.

In 2015, "Research on the identification of the number of ship propeller blades by applying multi-class classification to vector computers" was published by Mr. Daiei Kuni, Mr. Akira, and others. An improved algorithm for applying multiclass classification to a vector computer with an error correction code combination output, which is suitable for experiments that perform propeller blade number classification of a ship based on the entanglement signal identification spectrum of the radiation noise of the target ship, has been proposed. The method utilizes a vector computer-applied inference method and requires a large number of known samples and many feature extractions up to 33 dimensions in the modulation spectrum, which is complicated to operate. In addition, the specific identification method of the shaft frequency and the blade frequency was not submitted in the text.

In response to conventional technical problems, the present invention presents a method for extracting propeller blade number features based on radiation noise modulation. The radiation noise modulation spectrum of each type of private ship can be analyzed, which is convenient for extracting the features of the shaft frequency, blade frequency and the number of propeller blades.

The method for extracting the propeller blade number feature based on this radiation noise modulation includes the following steps.
(1) Collect the radiation noise signal of the ship and acquire the modulation spectrum diagram by Fourier transform.
(2) Search for a local peak in the modulation diagram and acquire the resonance frequency at the local peak position.
(3) The resonance frequency of the first local peak in the modulation diagram is determined, the axis frequency is determined, and the number N of the resonance frequencies is determined based on the number of local peaks.
(4) Determine the multiplication relationship between the axis frequency and the blade frequency, and determine other resonance frequencies.
(5) The average spectral coherent value of each harmonic frequency position is determined, the number of propeller blades is obtained by the naive Bayesian inference method, and finally the blade frequency is determined.
In step 1, the modulation diagram to be acquired has the circulation frequency as the horizontal axis and the average spectral coherent value as the vertical axis.

を探出する。 The findpeaks function of MATLAB software is used to search for the local peak of the average coherent value and the corresponding circulating frequency. Find the wave peak position in the original waveform with the findpeaks toolbox function in MATLAB. First, the first local peak and its resonant frequency

To find out.

ただし、

はｎ番目ローカルピーク位置の循環周波数の値である。 In step (2), when searching for a local peak, the difference in resonance frequency between two adjacent local peaks satisfies the following relationship.

However,

Is the value of the circulation frequency at the nth local peak position.

In step (3), when determining the shaft frequency, the resonance frequency at the first local peak position is set as the shaft frequency. If the resonance frequency value at the first local peak position is smaller than 0.9 Hz, it is removed and the second resonance frequency is used as the axis frequency.

ただし、

はｎ番目共振周波数、

；

は軸周波数である。
ステップ（５）では、共振周波数によって対応の平均コヒーレント値を確定する。共振周
波数

位置の平均コヒーレント値は

とする。

はｎ番目共振周波数位置の幅値を示す。
各共振周波数位置の平均スペクトルコヒーレント値

は

区間内の平均コヒーレント値から平均を求めて得られる。ただし、

はサンプリング点

前の5番目サンプリング点から

後の5番目サンプリング点までの区間を示す。 In step (4), the formula for determining the other resonant frequencies:

However,

Is the nth resonance frequency,

;

Is the axis frequency.
In step (5), the corresponding average coherent value is determined by the resonance frequency. Resonance frequency

The average coherent value of the position is

And.

Indicates the width value of the nth resonance frequency position.
Average spectral coherent value of each resonant frequency position

Is

It is obtained by calculating the average from the average coherent value in the section. However,

Is the sampling point

From the previous 5th sampling point

The section up to the 5th sampling point after that is shown.

ただし、Ｙ＝｛

｝（Ｙはすべての可能のプロペラ翼数の集合、

はその一つのプロペラ翼数、

はプロペラ翼数が３、

はプロペラ翼数が４、

はプロペラ翼数が５、

はプロペラ翼数が６、

はプロペラ翼数が７とする）、Ｘ＝｛

｝（Ｘは分類目標の変調スペクトル、

は該変調スペクトルにおける各共振周波数位置の振幅大小関係とする）、Ｐ（Ｘ）は分類
自身の確率（定常数）、

は各プロペラ翼数の類型

の先験的確率、

は所定Ｘの

プロペラ翼数類型への従属確率、

はプロペラ翼数類型

のＸ発生確率、

はプロペラ翼数類型

の変調スペクトルにおける特徴

発生の確率である。各

を算出すると、

が最大値であれば、Ｘは類型

に従属する。 In step (5), based on the naive Bayesian inference method, the relationship between the acquired resonant frequency value and the corresponding average coherent value determines which type of propeller blade number set the modulation diagram depends on. Determine. Finally, the number of propeller blades is determined.
If the number of samples is sufficient, the naive Bayesian inference method may be used directly and the results are also very certain. When the number of samples is small, analog samples may be generated based on the relationship between the width values at the time of each propeller blade number in the propeller blade number identification rule in Table 1 to classify the modulation diagram of the judgment target. The classification result is the number of propeller blades. Naive Bayes Official:

However, Y = {

} (Y is the set of all possible propeller blade numbers,

Is the number of propeller wings,

Has 3 propeller wings,

Has 4 propeller wings,

Has 5 propeller wings,

Has 6 propeller wings,

Has 7 propeller wings), X = {

} (X is the modulation spectrum of the classification target,

Is the amplitude magnitude relationship of each resonance frequency position in the modulation spectrum), P (X) is the probability of the classification itself (steady number),

Is the type of each propeller wing number

A priori probability,

Is a given X

Probability of dependence on propeller wing number type,

Is a propeller wing number type

X occurrence probability,

Is a propeller wing number type

Features in the modulation spectrum of

The probability of occurrence. each

When you calculate

If is the maximum value, X is a type

Subordinate to.

The naive Bayes inference method is based on Bayes'Theorem, and each characteristic condition is considered to be mutually independent. The joint probability distribution from input to output is learned by the training set sample provided in advance, and the classification target X is input based on the model obtained by the training, and the output that can maximize the posterior probability Y is obtained. It is what you want.
Obtaining the shaft frequency and the number of propeller blades, the blade frequency is the number of propeller blades x the shaft frequency.
Compared with the conventional technique, the present invention has the following effects.
According to the present invention, by adding a limiting condition to the search for local peaks, it is possible to prevent discrimination between two local peaks that are close to each other. By adding a limiting condition to the determination of the axis frequency,
It is possible to avoid erroneous judgment that identifies the under-axis frequency due to the influence of noise. When determining the peak at the resonance frequency position, the average peak value is obtained within one frequency section. When determining the number of propeller blades, use the naive Bayesian inference method. Its predominance is applicable to small samples and can solve problems that cannot be solved by the conventional identification rules described in the prior art. Finally, the axis frequency, blade frequency and propeller blade number features can be extracted from various modulation spectrum configurations.

It is a schematic flowchart figure of the extraction method of the propeller blade number feature based on radiation noise modulation. It is a modulation spectrum of the feature extraction target according to the Example of this invention. It is a figure which shows the peak obtained by the findpeaks function of MATLAB which concerns on Example of this invention. It is a figure which shows the frequency of the shaft frequency obtained by the judgment which concerns on embodiment of this invention. It is a result figure which was finally identified which concerns on Example of this invention.

Hereinafter, the present invention will be described in more detail with reference to the drawings and examples. It should be noted that the action of the examples is to make the invention easier to understand and not to limit the invention.

As shown in FIG. 1, the method for extracting the propeller blade number feature based on the radiation noise modulation includes the following steps.
S01: In this example, radiation noise from a certain merchant ship is used. The number of propeller blades on a commercial ship is 3, the number of revolutions on a commercial ship's propeller is 111 rpm, the blade frequency is 5.55 Hz, and the shaft frequency is 1.85 Hz.
And. The modulation spectrum is acquired by performing a short-time Fourier transform on the radiation noise of a merchant ship. The resonance frequency of the local peak and the local peak position is searched for in the modulation diagram.

式中、

はｎ番目ローカルピーク位置の循環周波数の値である。ピークの探索結果は図３に示され
る。 The modulation spectrum of the feature extraction target is shown in FIG. In that step, Matlab
The findpeaks function inside searches for the local peak of the average coherent value and the corresponding circulating frequency.
The difference in resonance frequency between two adjacent local peaks satisfies the relationship of the following equation.

During the ceremony

Is the value of the circulation frequency at the nth local peak position. The search result of the peak is shown in FIG.

とする。図４に示されるように、本実施例において

とし、１番目ローカルピーク位置の共振周波数が０．９Ｈｚ未満でないという条件を満た
し、１番目位置の共振周波数が軸周波数であり、

となっている。ローカルピークの数により、共振周波数の数Ｎを確定する。本例では、７
つのピークがあるから、Ｎ＝７となる。 S02: The shaft frequency and its resonance frequency are determined. If the resonance frequency value at the first local peak position is smaller than 0.9 Hz, it is removed and the second resonance frequency is used as the axis frequency. The axis frequency is

And. As shown in FIG. 4, in this embodiment

The condition that the resonance frequency of the first local peak position is not less than 0.9 Hz is satisfied, and the resonance frequency of the first position is the shaft frequency.

It has become. The number N of resonance frequencies is determined by the number of local peaks. In this example, 7
Since there are two peaks, N = 7.

（ただし、

はｎ番目の共振周波数、

とする）ので、
逓倍関係に基づいて、軸周波数が既定であれば、翼周波数位置の可能の共振周波数を確定
することができる。 S03: The value of another resonance frequency is determined by the multiplication relationship between the shaft frequency and the blade frequency.
Shaft frequency and blade frequency are multiplying relations:

(However,

Is the nth resonance frequency,

) So
Based on the multiplication relationship, if the axis frequency is the default, the possible resonance frequency of the blade frequency position can be determined.

区間内の平均コヒーレント値の平均を求めて、

位置の平均コヒーレント値

を得られる。ただし

はサンプリング点

前の５番目サンプリング点から

後の５番目サンプリング点までの区間を示す。 S04: The average spectral coherent value of each harmonic frequency position is determined.
Due to the error, when calculating the coherent value for each harmonic frequency position,

Find the average of the average coherent values in the interval,

Average coherent value of position

Can be obtained. However,

Is the sampling point

From the previous 5th sampling point

The section up to the 5th sampling point after that is shown.

ある。ナイーブベイズの推断法に基づくナイーブベイズの公式：

本例では、Ｙ＝｛

｝（Ｙはすべての可能のプロペラ翼数の集合、

はその一つのプロペラ翼数、

はプロペラ翼数が３、

はプロペラ翼数が４、

はプロペラ翼数が５、

はプロペラ翼数が６、

はプロペラ翼数が７である）、Ｘ＝｛

｝（Ｘは分類目標の変調スペクトル、

は該変調スペクトルにおける各共振周波数位置の振幅の大小関係である。またＰ（Ｘ）は
分類自身の確率である（定乗数）。

は各プロペラ翼数類型の先験的確率、即ち

の確率である。

は所定Ｘの

プロペラ翼数類型への従属確率である。

はプロペラ翼数類型

のＸの発生確率である。

はプロペラ翼数類型

の変調スペクトルにおける特徴

発生の確率である。各

を計算すると、

が最大値であれば、Ｘが類型

に従属すると思われる。 S05: Obtain the number of propeller blades based on the naive Bayesian inference method. Solve the classification problem. That is, based on the naive Bayesian inference method, the relationship between the obtained resonance frequency value and the corresponding average coherent value determines which type of set of propeller blades the modulation diagram depends on. Finally, the number of propeller blades is determined. In this example, the relationship between the width values in the obtained modulation diagram is

is there. Naive Bayes formula based on Naive Bayes inference:

In this example, Y = {

} (Y is the set of all possible propeller blade numbers,

Is the number of propeller wings,

Has 3 propeller wings,

Has 4 propeller wings,

Has 5 propeller wings,

Has 6 propeller wings,

Has 7 propeller wings), X = {

} (X is the modulation spectrum of the classification target,

Is the magnitude relationship of the amplitude of each resonance frequency position in the modulation spectrum. Further, P (X) is the probability of the classification itself (constant multiplier).

Is the a priori probability of each propeller wing number type, i.e.

Probability of.

Is a given X

It is the probability of dependence on the propeller wing number type.

Is a propeller wing number type

Is the probability of occurrence of X.

Is a propeller wing number type

Features in the modulation spectrum of

The probability of occurrence. each

When you calculate

If is the maximum value, then X is the type

Seems to be subordinate to.

が既定量であるから、

を算出できる。本例では、得られる変調図における幅値間の関係は

がある。該変調図における振幅間の関係はＸのある特徴

により示される。最後に計算により

が最大値となる。したがって、プロペラ翼数が３であると確定できる。最後に翼周波数は
プロペラ翼数×軸周波数となっており、即ち翼周波数は５．５６５Ｈｚとなる。最後の識
別結果は図５に示される。 A large number of samples of propeller blade number modulation diagrams are known, or according to the samples simulated according to the rules in Table 1.

Is the default amount

Can be calculated. In this example, the relationship between the width values in the resulting modulation diagram is

There is. The relationship between the amplitudes in the modulation diagram is a characteristic of X

Indicated by. Finally by calculation

Is the maximum value. Therefore, it can be determined that the number of propeller blades is 3. Finally, the blade frequency is the number of propeller blades × the axis frequency, that is, the blade frequency is 5.565 Hz. The final identification result is shown in FIG.

The description of the above examples is for understanding the method of the present invention and its spirit. It goes without saying that those skilled in the art may also implement the invention with some modifications and modifications without departing from the principles of the invention, all of which are within the scope of protection of the invention. No.

The present invention relates to a technical area for extracting frequency components of a signal, and more particularly to a method for extracting propeller blade number features based on radiation noise modulation.

Radiation noise of a ship is caused by cavitation noise during rotation of a rotor blade (propeller). The modulation spectrum of the radiant noise of the ship includes the characteristic information of the number of propeller blades of the ship. By analyzing the relationship between each resonance frequency and amplitude, the shaft frequency of the ship propeller,
The features of the blade frequency and the number of propeller blades can be extracted.

If the feature quantity information of the propeller shaft frequency, k blade frequency, and propeller blade number of a private ship can be obtained, it will contribute to the determination of the ship model number. There are smuggling and stowaways along the sea. If the noise information of the ship is captured by the voice sensor, the modulation spectrum is acquired by the Fourier transform, and the feature quantity information of the shaft frequency, the blade frequency and the number of propeller blades of the propeller can be extracted, the supervisor can judge the type of the ship.

Conventionally, the following identification methods have been used in the world. The modulation spectrum is acquired, and the resonance frequency of the peak of the width value and the peak position is read from the modulation spectrum. Based on the propeller blade number feature identification rule (Table 1), the propeller blade number feature value is extracted from the resonance frequency and amplitude in the modulation spectrum. P (n) indicates the width value of the nth harmonic line spectrum of the axis frequency.

になった場合、表１のすべてのプロペラ翼数識別規則の条件は満たされるため、プロペラ
翼数を識別できない。 However, the configuration of the modulation spectrum is complicated depending on the structure, operating conditions, environment, etc. of the ship, and only a typical case can be identified based on the discrimination rule, and all the modulation spectrum configurations are not applicable. For example

If, the conditions of all the propeller blade number identification rules in Table 1 are satisfied, and the propeller blade number cannot be identified.

Chinese patent application 2019107990217. Regarding the extraction method of rotor characteristics in X,
A multidimensional statistical modeling method for rotor cavitation wake microfeatures has been disclosed, but the extracted features are mainly limited to the geometric parameters of the rotor and the characteristics of the operating conditions, and the characteristics of the shaft frequency, blade frequency, and propeller blade number. Not as good as that.

In 2015, "Research on the identification of the number of ship propeller blades by applying multi-class classification to vector computers" was published by Mr. Daiei Kuni, Mr. Akira, and others. An improved algorithm for applying multiclass classification to a vector computer with an error correction code combination output, which is suitable for experiments that perform propeller blade number classification of a ship based on the entanglement signal identification spectrum of the radiation noise of the target ship, has been proposed. The method utilizes a vector computer-applied inference method and requires a large number of known samples and many feature extractions up to 33 dimensions in the modulation spectrum, which is complicated to operate. In addition, the specific identification method of the shaft frequency and the blade frequency was not submitted in the text.

In response to conventional technical problems, the present invention presents a method for extracting propeller blade number features based on radiation noise modulation. The radiation noise modulation spectrum of each type of private ship can be analyzed, which is convenient for extracting the features of the shaft frequency, blade frequency and the number of propeller blades.

The method for extracting the propeller blade number feature based on this radiation noise modulation includes the following steps.
(1) Collect the radiation noise signal of the ship and acquire the modulation spectrum diagram by Fourier transform.
(2) The local peak is searched for in the modulation spectrum diagram, and the resonance frequency at the local peak position is acquired.
(3) The resonance frequency of the first local peak in the modulation spectrum diagram is determined, the axis frequency is determined, and the number N of the resonance frequencies is determined based on the number of local peaks.
(4) Determine the multiplication relationship between the axis frequency and the blade frequency, and determine other resonance frequencies.
(5) The average spectral coherent value of each resonance frequency position is determined, the number of propeller blades is obtained by the naive Bayesian inference method, and finally the blade frequency is determined.
In step 1, the modulation spectrum diagram to be acquired has the circulation frequency as the horizontal axis and the average spectrum coherent value as the vertical axis.

を探出する。 The MATLAB (TM) software findpeaks function of searching for a local peak and a corresponding cyclic frequency of average coherent values. To Sagude wave peak position in the original waveform by findpeaks toolbox function in MATLAB (registered trademark). First, the first local peak and its resonant frequency

To find out.

ただし、

はｎ番目ローカルピーク位置の循環周波数の値である。 In step (2), when searching for a local peak, the difference in resonance frequency between two adjacent local peaks satisfies the following relationship.

However,

Is the value of the circulation frequency at the nth local peak position.

In step (3), when determining the shaft frequency, the resonance frequency at the first local peak position is set as the shaft frequency. If the resonance frequency value at the first local peak position is smaller than 0.9 Hz, it is removed and the second resonance frequency is used as the axis frequency.

ただし、

はｎ番目共振周波数、

；

は軸周波数である。
ステップ（５）では、共振周波数によって対応の平均コヒーレント値を確定する。共振周
波数

位置の平均コヒーレント値は

とする。

はｎ番目共振周波数位置の幅値を示す。
各共振周波数位置の平均スペクトルコヒーレント値

は

区間内の平均コヒーレント値から平均を求めて得られる。ただし、

はサンプリング点

前の5番目サンプリング点から

後の5番目サンプリング点までの区間を示す。 In step (4), the formula for determining the other resonant frequencies:

However,

Is the nth resonance frequency,

;

Is the axis frequency.
In step (5), the corresponding average coherent value is determined by the resonance frequency. Resonance frequency

The average coherent value of the position is

And.

Indicates the width value of the nth resonance frequency position.
Average spectral coherent value of each resonant frequency position

Is

It is obtained by calculating the average from the average coherent value in the section. However,

Is the sampling point

From the previous 5th sampling point

The section up to the 5th sampling point after that is shown.

ただし、Ｙ＝｛

｝（Ｙはすべての可能のプロペラ翼数の集合、

はその一つのプロペラ翼数、

はプロペラ翼数が３、

はプロペラ翼数が４、

はプロペラ翼数が５、

はプロペラ翼数が６、

はプロペラ翼数が７とする）、Ｘ＝｛

｝（Ｘは分類目標の変調スペクトル、

は該変調スペクトルにおける各共振周波数位置の振幅大小関係とする）、Ｐ（Ｘ）は分類
自身の確率（定常数）、

は各プロペラ翼数の類型

の先験的確率、

は所定Ｘの

プロペラ翼数類型への従属確率、

はプロペラ翼数類型

のＸ発生確率、

はプロペラ翼数類型

の変調スペクトルにおける特徴

発生の確率である。各

を算出すると、

が最大値であれば、Ｘは類型

に従属する。 In step (5), based on naive Bayes inferred method, the relationship between the resonance frequency value of the acquired and the average coherent values of corresponding, or the modulation spectrum diagram is dependent on a set of propeller speed wing which type To confirm. Finally, the number of propeller blades is determined.
If the number of samples is sufficient, the naive Bayesian inference method may be used directly and the results are also very certain. When the number of samples is small, analog samples may be generated based on the relationship between the width values at each propeller blade number in the propeller blade number identification rule in Table 1 to classify the modulation spectrum diagram of the judgment target. The classification result is the number of propeller blades. Naive Bayes Official:

However, Y = {

} (Y is the set of all possible propeller blade numbers,

Is the number of propeller wings,

Has 3 propeller wings,

Has 4 propeller wings,

Has 5 propeller wings,

Has 6 propeller wings,

Has 7 propeller wings), X = {

} (X is the modulation spectrum of the classification target,

Is the amplitude magnitude relationship of each resonance frequency position in the modulation spectrum), P (X) is the probability of the classification itself (steady number),

Is the type of each propeller wing number

A priori probability,

Is a given X

Probability of dependence on propeller wing number type,

Is a propeller wing number type

X occurrence probability,

Is a propeller wing number type

Features in the modulation spectrum of

The probability of occurrence. each

When you calculate

If is the maximum value, X is a type

Subordinate to.

The naive Bayes inference method is based on Bayes'Theorem, and each characteristic condition is considered to be mutually independent. The joint probability distribution from input to output is learned by the training set sample provided in advance, and the classification target X is input based on the model obtained by the training, and the output that can maximize the posterior probability Y is obtained. It is what you want.
Obtaining the shaft frequency and the number of propeller blades, the blade frequency is the number of propeller blades x the shaft frequency.
Compared with the conventional technique, the present invention has the following effects.
According to the present invention, by adding a limiting condition to the search for local peaks, it is possible to prevent discrimination between two local peaks that are close to each other. By adding a limiting condition to the determination of the axis frequency,
It is possible to avoid erroneous judgment that identifies the under-axis frequency due to the influence of noise. When determining the peak at the resonance frequency position, the average peak value is obtained within one frequency section. When determining the number of propeller blades, use the naive Bayesian inference method. Its predominance is applicable to small samples and can solve problems that cannot be solved by the conventional identification rules described in the prior art. Finally, the axis frequency, blade frequency and propeller blade number features can be extracted from various modulation spectrum configurations.

It is a schematic flowchart figure of the extraction method of the propeller blade number feature based on radiation noise modulation. It is a modulation spectrum of the feature extraction target according to the Example of this invention. It is a figure which shows the peak obtained by the findpeaks function of MATLAB (registered trademark) which concerns on Example of this invention. It is a figure which shows the frequency of the shaft frequency obtained by the judgment which concerns on embodiment of this invention. It is a result figure which was finally identified which concerns on Example of this invention.

Hereinafter, the present invention will be described in more detail with reference to the drawings and examples. It should be noted that the action of the examples is to make the invention easier to understand and not to limit the invention.

As shown in FIG. 1, the method for extracting the propeller blade number feature based on the radiation noise modulation includes the following steps.
S01: In this example, radiation noise from a certain merchant ship is used. The number of propeller blades on a commercial ship is 3, the number of revolutions on a commercial ship's propeller is 111 rpm, the blade frequency is 5.55 Hz, and the shaft frequency is 1.85 Hz.
And. The modulation spectrum is acquired by performing a short-time Fourier transform on the radiation noise of a merchant ship. The resonance frequency of the local peak and the local peak position is searched for in the modulation spectrum diagram.

式中、

はｎ番目ローカルピーク位置の循環周波数の値である。ピークの探索結果は図３に示され
る。 The modulation spectrum of the feature extraction target is shown in FIG. In that step, Matlab
The findpeaks function inside searches for the local peak of the average coherent value and the corresponding circulating frequency.
The difference in resonance frequency between two adjacent local peaks satisfies the relationship of the following equation.

During the ceremony

Is the value of the circulation frequency at the nth local peak position. The search result of the peak is shown in FIG.

とする。図４に示されるように、本実施例において

とし、１番目ローカルピーク位置の共振周波数が０．９Ｈｚ未満でないという条件を満た
し、１番目位置の共振周波数が軸周波数であり、

となっている。ローカルピークの数により、共振周波数の数Ｎを確定する。本例では、７
つのピークがあるから、Ｎ＝７となる。 S02: The shaft frequency and its resonance frequency are determined. If the resonance frequency value at the first local peak position is smaller than 0.9 Hz, it is removed and the second resonance frequency is used as the axis frequency. The axis frequency is

And. As shown in FIG. 4, in this embodiment

The condition that the resonance frequency of the first local peak position is not less than 0.9 Hz is satisfied, and the resonance frequency of the first position is the shaft frequency.

It has become. The number N of resonance frequencies is determined by the number of local peaks. In this example, 7
Since there are two peaks, N = 7.

（ただし、

はｎ番目の共振周波数、

とする）ので、
逓倍関係に基づいて、軸周波数が既定であれば、翼周波数位置の可能の共振周波数を確定
することができる。 S03: The value of another resonance frequency is determined by the multiplication relationship between the shaft frequency and the blade frequency.
Shaft frequency and blade frequency are multiplying relations:

(However,

Is the nth resonance frequency,

) So
Based on the multiplication relationship, if the axis frequency is the default, the possible resonance frequency of the blade frequency position can be determined.

区間内の平均コヒーレント値の平均を求めて、

位置の平均コヒーレント値

を得られる。ただし

はサンプリング点

前の５番目サンプリング点から

後の５番目サンプリング点までの区間を示す。 S04: The average spectral coherent value of each resonance frequency position is determined.
There is an error, so when calculating the coherent value for each resonant frequency position,

Find the average of the average coherent values in the interval,

Average coherent value of position

Can be obtained. However,

Is the sampling point

From the previous 5th sampling point

The section up to the 5th sampling point after that is shown.

ある。ナイーブベイズの推断法に基づくナイーブベイズの公式：

本例では、Ｙ＝｛

｝（Ｙはすべての可能のプロペラ翼数の集合、

はその一つのプロペラ翼数、

はプロペラ翼数が３、

はプロペラ翼数が４、

はプロペラ翼数が５、

はプロペラ翼数が６、

はプロペラ翼数が７である）、Ｘ＝｛

｝（Ｘは分類目標の変調スペクトル、

は該変調スペクトルにおける各共振周波数位置の振幅の大小関係である。またＰ（Ｘ）は
分類自身の確率である（定乗数）。

は各プロペラ翼数類型の先験的確率、即ち

の確率である。

は所定Ｘの

プロペラ翼数類型への従属確率である。

はプロペラ翼数類型

のＸの発生確率である。

はプロペラ翼数類型

の変調スペクトルにおける特徴

発生の確率である。各

を計算すると、

が最大値であれば、Ｘが類型

に従属すると思われる。 S05: Obtain the number of propeller blades based on the naive Bayesian inference method. Solve the classification problem. That based on naive Bayes inferred method, the relationship between the obtained resonance frequency value and the corresponding average coherent values, to determine whether the modulation spectrum diagram is dependent on which set of types of propeller speed blades. Finally, the number of propeller blades is determined. In this example, the resulting modulation spectrum
The relationship between width values in the figure

is there. Naive Bayes formula based on Naive Bayes inference:

In this example, Y = {

} (Y is the set of all possible propeller blade numbers,

Is the number of propeller wings,

Has 3 propeller wings,

Has 4 propeller wings,

Has 5 propeller wings,

Has 6 propeller wings,

Has 7 propeller wings), X = {

} (X is the modulation spectrum of the classification target,

Is the magnitude relationship of the amplitude of each resonance frequency position in the modulation spectrum. Further, P (X) is the probability of the classification itself (constant multiplier).

Is the a priori probability of each propeller wing number type, i.e.

Probability of.

Is a given X

It is the probability of dependence on the propeller wing number type.

Is a propeller wing number type

Is the probability of occurrence of X.

Is a propeller wing number type

Features in the modulation spectrum of

The probability of occurrence. each

When you calculate

If is the maximum value, then X is the type

Seems to be subordinate to.

が既定量であるから、

を算出できる。本例では、得られる変調スペクトル図における幅値間の関係は

がある。該変調スペクトル図における振幅間の関係はＸのある特徴

により示される。最後に計算により

が最大値となる。したがって、プロペラ翼数が３であると確定できる。最後に翼周波数は
プロペラ翼数×軸周波数となっており、即ち翼周波数は５．５６５Ｈｚとなる。最後の識
別結果は図５に示される。 A large number of samples of the propeller blade number modulation spectrum diagram are known, or according to the samples simulated according to the rules in Table 1.

Is the default amount

Can be calculated. In this example, the relationship between the width values in the resulting modulation spectrum diagram is

There is. Wherein the relationship between the amplitudes in the modulation spectrum diagram with X

Indicated by. Finally by calculation

Is the maximum value. Therefore, it can be determined that the number of propeller blades is 3. Finally, the blade frequency is the number of propeller blades × the axis frequency, that is, the blade frequency is 5.565 Hz. The final identification result is shown in FIG.

The description of the above examples is for understanding the method of the present invention and its spirit. It goes without saying that those skilled in the art may also implement the invention with some modifications and modifications without departing from the principles of the invention, all of which are within the scope of protection of the invention. No.

Claims (4)

A method for extracting propeller blade number characteristics based on radiation noise modulation.
(1) A step of collecting a ship's radiation noise signal and acquiring a modulation spectrum diagram by Fourier transform.
(2) The step of searching for a local peak in the modulation diagram and acquiring the resonance frequency at the local peak position,
(3) A step of determining the resonance frequency of the first local peak in the modulation diagram, determining the axis frequency, and determining the number N of the resonance frequencies based on the number of local peaks.
(4) The step of determining the multiplication relationship between the axis frequency and the blade frequency and determining the other resonance frequencies,
(5) Includes a step of determining the average spectral coherent value of each harmonic frequency position, obtaining the number of propeller blades by the naive Bayesian inference method, and finally determining the blade frequency.
In step (2), when searching for a local peak, the difference in resonance frequency between two adjacent local peaks satisfies the following relationship.

However,

Is the value of the circulation frequency at the nth local peak position,
In step (3), when determining the shaft frequency, the resonance frequency at the first local peak position is set as the shaft frequency, and if the resonance frequency value at the first local peak position is smaller than 0.9 Hz, it is removed. The second resonance frequency is the axis frequency,
A method for extracting propeller blade number characteristics based on radiation noise modulation. In step (4), the formula for determining the other resonance frequencies is

However,

Is the nth resonance frequency,

;

The method for extracting a propeller blade number feature based on the radiation noise modulation according to claim 1, wherein is an axial frequency. Average spectral coherent value of each resonant frequency position

Is

Obtained by calculating the average from the average coherent value within the interval, however

Is the sampling point

From the previous 5th sampling point

The method for extracting a propeller blade number feature based on the radiation noise modulation according to claim 1, wherein the section up to the fifth sampling point is shown later. In step (5), based on the naive Bayesian inference method, the relationship between the acquired resonance frequency value and the corresponding average coherent value determines which type of propeller blade number set the modulation diagram depends on. Confirm, finally confirm the number of propeller wings,
Naive Bayes official is

However, Y = {

} (Y is the set of all possible propeller blade numbers,

Is the number of propeller wings,

Has 3 propeller wings,

Has 4 propeller wings,

Has 5 propeller wings,

Has 6 propeller wings,

Has 7 propeller wings), X = {

} (X is the modulation spectrum of the classification target,

Is the amplitude magnitude relationship of each resonance frequency position in the modulation spectrum), P (X) is the probability of the classification itself (steady number),

Is the type of each propeller wing number

A priori probability,

Is a given X

Probability of dependence on propeller wing number type,

Is a propeller wing number type

X occurrence probability,

Is a propeller wing number type

Features in the modulation spectrum of

Probability of occurrence, each

When you calculate

If is the maximum value, X is a type

The method for extracting a propeller blade number feature based on the radiation noise modulation according to claim 1, wherein the propeller blade number feature is dependent on the above.

Images