CN114487952A - A quench detection system and method using acoustic fiber - Google Patents

Abstract

The invention provides a quench detection system using an acoustic fiber, comprising: an acoustic fiber wire, an optical fiber sound listener, and a voiceprint analysis server; wherein one end of the acoustic fiber wire is wound around the outer ring of the superconducting magnet, and the other end is wound around the outer ring of the superconducting magnet. Connect to the optical transceiver port of the fiber optic sound listener; the data communication port of the fiber optic sound listener is connected to the voiceprint analysis server; the optical signal sent by the fiber optic sound listener is reflected by Rayleigh and part of the signal is reflected back to the fiber optic sound listener The optical fiber sound listener restores the reflected light signal to a sound signal, and then transmits it to the voiceprint analysis server, and the voiceprint analysis server performs voiceprint analysis on the sound signal. The invention realizes voiceprint recognition through voiceprint perception and based on MFCC vector, and has the beneficial effects of high detection accuracy, strong anti-interference ability, fast calculation speed and low maintenance cost.

Description

A quench detection system and method using acoustic fiber

technical field

The invention relates to the technical field of quench detection for superconducting magnets, and a method for quenching diagnosis of superconducting magnets by using an acoustic fiber as a sensor and a voiceprint recognition technology.

Background technique

A major breakthrough in controllable thermonuclear fusion energy research is the successful application of superconducting technology to coils that generate strong magnetic fields in tokamak. The fully superconducting tokamak can achieve steady-state operation, and by greatly improving confinement under steady-state operating conditions, lays the engineering technical and physical foundation for future steady-state, advanced fusion reactors. Therefore, future commercial reactors must be fully superconducting to achieve steady-state operation.

The superconducting magnet system is also one of the most important and expensive components of the superconducting fusion device (accounting for 25% of the total cost). Once the superconducting magnet quenches, the process of converting the extremely high electromagnetic heat stored in the superconductor to thermal energy, quenching protection If it is not in time, the magnet will be irreversibly damaged within a few seconds, resulting in the inoperability of the superconducting device, resulting in huge economic losses! Therefore, the quench diagnosis and protection system is the highest level guarantee to ensure the safe operation of large-scale superconducting devices.

When the magnet is quenched, a series of physical quantities such as resistance value, temperature, and pressure will change, and quench diagnosis is to convert these physical quantities into electrical signals as the basis for quenching judgment. The traditional superconducting coil superconducting diagnostic methods mainly include: resistance voltage detection method, temperature rise detection method, pressure detection method, flow detection method, ultrasonic detection and so on.

Among them, the temperature rise detection method determines the quench by detecting the temperature change or temperature change rate of the superconducting coil in real time, such as the superconducting energy storage system SMSE of the Institute of Electrical Engineering of the Chinese Academy of Sciences. high. However, in the face of multiple temperature distribution points, the temperature sensor and subsequent measurement system are required to have high measurement accuracy and stability.

Pressure and flow detection are also the main basis for quenching, but there are shortcomings such as slow sensor response, delay, and different influencing factors. According to the difficulty of the detection circuit construction, the voltage detection method is more commonly used in the quench detection. The quench judgment of the voltage detection method is mainly based on the superconductivity detection signal voltage exceeding the threshold voltage and continuing for the set delay time to determine that the quench occurs.

The most commonly used voltage threshold discrimination methods include the bridge method and the same-winding detection method. Among them, the bridge method is mainly used in superconducting devices such as the HT-7 tokamak device of the Institute of Plasma, Chinese Academy of Sciences, the SMSE of the Chinese Academy of Sciences, and the LHD device of Japan. This quench detection method uses the Wheatstone bridge balance principle to obtain quench detection. The signal has the advantages of being easy to use and requiring less quench detection circuits, but on the one hand, there are blind spots in quench detection, and on the other hand, the anti-electromagnetic interference ability is not strong. The co-winding detection method is mainly used in the EAST device of the Institute of Plasma, Chinese Academy of Sciences. This quench detection method uses the same winding and the superconducting coil to form a quench detection sampling circuit to obtain the reference voltage signal, and at the same time, the sensors are used to measure all energized The current differential signal of the coil is used as the secondary compensation voltage, and the secondary compensation voltage and the reference voltage are accumulated according to a certain compensation coefficient to offset the induced voltage generated by the coupling of the coil itself and other fast alternating coils, and the accumulated voltage is used as the quench. detection amount. This method can realize quench detection in a rapidly changing magnetic field environment, but has the disadvantage that once the same winding is broken, it cannot be repaired.

To sum up, the commonly used quench diagnostic techniques all have certain technical defects, especially in the aspects of quench positioning, quench early warning, and quench auxiliary discrimination. Considering the importance of quench diagnosis to large-scale superconducting devices, more effective diagnostic techniques are needed. as an effective compensation for the prior art.

SUMMARY OF THE INVENTION

In view of the above defects and improvement requirements of the existing methods, the purpose of the present invention is to provide a method for quenching a superconducting magnet by using an acoustic fiber as a sensor and adopting the voiceprint recognition technology. Its advantage is that it only needs to be laid outside the superconducting magnet armor, high voltage resistance, anti-electromagnetic interference, high sensitivity, 20% shortened response time, high accuracy of true quench discrimination, and can be used as a quench positioning and auxiliary quench discrimination method.

In order to solve the above technical problems, the present invention provides a quench detection system using an acoustic fiber, the system includes: an acoustic fiber wire, an optical fiber sound listener, and a voiceprint analysis server; wherein, one end of the acoustic fiber wire is wound On the outer ring of the superconducting magnet, the other end is connected to the optical transceiver port of the fiber optic sound interceptor; the data communication port of the fiber optic sound interceptor is connected to the voiceprint analysis server; the optical signal emitted by the optical fiber sound interceptor is reflected by Rayleigh Some of the signals are reflected back to the fiber optic sound listener, and the fiber optic sound listener restores the reflected light signal to a sound signal, and then transmits it to the voiceprint analysis server, which performs voiceprint analysis on the sound signal.

Further, the optical fiber sound listener includes a light emitting unit, a light receiving unit, an amplifying unit, a demodulation unit, and a communication unit, wherein the light emitting unit emits an optical signal at a certain frequency, and the optical signal travels in the optical fiber wire. When encountering the sound wave generated by the magnet and causing deformation, the optical signal is reflected by Rayleigh and part of the signal is reflected back to the fiber optic sound listener, and the reflected optical signal is obtained after receiving and processing by the light receiving unit and the amplifying unit; The optical signal is restored to a sound signal, and then the communication unit transmits the sound signal and length information to the voiceprint analysis server. Wherein, the communication unit includes an Ethernet communication unit, a WIFI module, a 4G module, a 5G module, and the like. The voiceprint analysis server firstly performs weighted dimensionality reduction optimization on the sound signal based on the MFCC feature vector, and secondly, applies a vector quantization algorithm to identify the optimized sound signal, and finally determines whether a quench occurs.

The present invention also provides a quench detection method utilizing acoustic fiber, applying the above system, and the method includes the following steps:

Step S1, the optical fiber sound listener sends out an optical signal;

Step S2, the optical fiber sound listener receives the reflected light signal and restores it to a sound signal;

Step S3, the voiceprint analysis server performs weighted dimension reduction optimization on the sound signal based on the MFCC feature vector;

Step S4, the voiceprint analysis server uses a vector quantization algorithm to identify the optimized voice signal, and finally determines whether a quench occurs.

The step S3 includes a voiceprint preprocessing step, and the voiceprint preprocessing step includes:

Step 31: Framing the voiceprint signal, and the framing relationship is expressed as M = n - Lb/[L(1-b)], where M is the number of frames, n is the length of the audio signal, L is the frame length, and b is Overlap rate; preferably, take 20ms as the frame length L of one frame, and 40% as the overlap rate b;

Step 32: Obtain the MFCC eigenvectors of the framed signals respectively to form a feature vector group. The obtaining process includes FFT transformation, Mel filtering, logarithmic transformation and discrete cosine transformation.

The step S3 also includes performing dimension reduction and optimization on the preprocessed sound signal: specifically, reducing the dimension of the calculated high-dimensional MFCC vector through the PCA algorithm; the Mel filtering and the PCA algorithm are specifically described in the description of the embodiments.

Further, the dimensionality-reduced and optimized target MFCC feature vectors v, v1, ..., vh, variance contribution rate and cumulative variance contribution rate obtained by analyzing the target signal by the voiceprint analysis server are input into the vector quantization algorithm after training, and finally get Quench judgment result.

The detection system and method provided by the invention can perform superconducting quench detection based on the voiceprint recognition technology, the detection system is relatively independent from the superconducting system, the fault will not occur in the superconducting system, and the maintenance cost is low; The comprehensive judgment of the vector and the neural network can extract the depth information of the fault, with high detection accuracy and strong anti-interference ability; finally, the present invention utilizes dimension reduction optimization in the process of voiceprint analysis, which further reduces the amount of calculation while ensuring the accuracy. , to improve computational efficiency.

Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the description, claims and drawings.

Description of drawings

The accompanying drawings are used to provide a further understanding of the technical solutions of the present invention, and constitute a part of the specification. They are used to explain the technical solutions of the present invention together with the embodiments of the present application, and do not limit the technical solutions of the present invention.

1 is a schematic diagram of a system provided by an embodiment of the present invention;

FIG. 2 is a voiceprint analysis diagram and a current comparison diagram of a timeout period provided by an embodiment of the present invention.

Detailed ways

In order to enable the relevant technical personnel to better understand the present invention and have a clearer understanding of the purpose, technical solutions and advantages of this application, the present invention will be further described below with reference to specific examples and accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention. It should be noted that, the embodiments in the present application and the features in the embodiments may be arbitrarily combined with each other if there is no conflict.

The present invention provides a quench detection system using acoustic optical fiber. As shown in FIG. 1 , the system includes: an acoustic optical fiber wire, an optical fiber sound listener, and a voiceprint analysis server; wherein, one end of the acoustic optical fiber wire is wound On the outer ring of the superconducting magnet, the other end is connected to the optical transceiver port of the fiber optic sound listening instrument; the data communication port of the fiber optic sound listening instrument is connected to the voiceprint analysis server.

The optical fiber sound listener includes a light emitting unit, a light receiving unit, an amplifying unit, a demodulation unit, and a communication unit, wherein the light emitting unit emits an optical signal at a certain frequency, and the optical signal encounters a magnet during the traveling process of the optical fiber wire. The deformation caused by the sound wave generated (it can also be regarded as the modulation of the sound wave to the optical signal), the optical signal is reflected by Rayleigh and part of the signal is reflected back to the optical fiber sound listener, and is received and processed by the light receiving unit and the amplifying unit. After reflecting the light signal, the demodulation unit restores the reflected light signal into a sound signal, and then the communication unit transmits the sound signal and length information to the voiceprint analysis server, and the voiceprint analysis server performs voiceprint analysis on the sound signal.

Preferably, the communication unit includes an Ethernet communication unit, a WIFI module, a 4G module, a 5G module, and the like.

The voiceprint analysis server firstly performs weighted dimensionality reduction optimization on the sound signal based on the MFCC feature vector, and secondly, applies a vector quantization algorithm to identify the optimized sound signal, and finally determines whether a quench occurs.

Further, the present invention also provides a quench detection method using the acoustic optical fiber relying on the above-mentioned quench detection system, the method comprising the following steps:

Step S1, the optical fiber sound listener sends out an optical signal;

Step S2, the optical fiber sound listener receives the reflected light signal and restores it to a sound signal;

Step S3, the voiceprint analysis server performs weighted dimension reduction optimization on the sound signal based on the MFCC feature vector;

Step S4, the voiceprint analysis server uses a vector quantization algorithm to identify the optimized voice signal, and finally determines whether a quench occurs.

Preferably, weighted dimensionality reduction optimization for the sound signal includes voiceprint preprocessing and dimensionality reduction optimization.

1. Voiceprint preprocessing steps: Voiceprint preprocessing includes two sub-steps: framing and windowing.

(1) When dividing the voiceprint signal into frames, in order to ensure the continuity between the signals of two adjacent frames, there is generally overlap between the two frames. The framing relationship can be expressed as M = n - Lb/[L(1-b)],

Where M is the number of frames, n is the length of the audio signal, L is the frame length, and b is the overlap rate.

When using voiceprint to detect the occurrence of quench, preferably, 20ms can be taken as the frame length L of one frame, and 40% is the overlap rate b.

(2) After preprocessing, the signal needs to be subjected to discrete Fourier transform. Specifically, a Hamming window is applied to each framed signal and then a discrete Fourier transform is performed to increase the continuity at both ends of the signal, thereby reducing the Fourier transform. resulting distortion.

Among them, the MFCC coefficients are based on the cepstral coefficients of the Mel frequency domain, where the Mel frequency is the frequency domain transformed according to the auditory perception characteristics of the human ear: B(h)=25951g(1h / 700).

The MFCC eigenvectors are respectively obtained from the above framed signals to form a eigenvector group, and the obtaining process includes FFT transformation, Mel filtering, logarithmic transformation and discrete cosine transformation.

，其计算公式为：

Further, the Mel filtering is implemented by a filter bank composed of several triangular bandpass filters. Let the number of filters be p, the signal can be filtered to obtain p parameters

, its calculation formula is:

N is the number of FFT points, X(k) is the FFT of the framed signal after preprocessing, and H _i (k) is the filter parameter, which can be expressed as:

Where: f[i] is the center frequency of the triangular filter;

When mi is calculated, take the logarithm of it and perform discrete cosine transform, and the calculated result c(i) is the MFCC feature vector of the framed signal.

2. Dimensionality reduction optimization step: When analyzing sound signals, a higher MFCC feature vector dimension can ensure sufficient extraction of sound signal features, but a too high dimension will also consume a lot of computing time and increase computational complexity . In order to improve the computational efficiency, the PCA algorithm is used to reduce the dimension of the calculated high-dimensional MFCC vector, while ensuring the accuracy of the features.

The PCA algorithm is as follows:

(1) There are e eigenvectors to form a matrix G, and the dimension of each eigenvector is h, then G can be expressed as:

，(2) Calculate the correlation matrix R of G, as

,

和对应的特征向量

。According to this, the eigenvalues of the correlation matrix R are calculated

and the corresponding eigenvectors

.

(3) Calculate the variance contribution rate and the cumulative variance contribution rate, respectively:

，

,

The occurrence of quench can be accurately determined by performing vector quantization algorithm prediction through the above feature data and quench-time audio features.

Combined with the voiceprint analysis diagram and the current comparison diagram of the quench time shown in FIG. 2 , it is verified by experiments that the above detection system can accurately determine the occurrence of quenching based on the analysis of the voiceprint.

At the same time, it was also observed through experiments that the current test magnetic field strength will interfere with the sensor, but the interference strength is kept at a low level and will not affect the reception of normal voiceprint signals, while the optical fiber is not disturbed by the magnetic field change at all. This further verifies the anti-interference ability of the present application.

Further, the vector quantization algorithm is obtained by training a neural network, including the following process:

(1) Select the voiceprint training set, perform the above steps 1 and 2 on the training set, obtain the MFCC feature vector, variance contribution rate and cumulative variance contribution rate after dimension reduction and optimization, and calibrate the training set;

(2) Take the MFCC feature vector (v, v1, . The vector quantization algorithm after training is completed;

(3) Input the MFCC feature vector, variance contribution rate and cumulative variance contribution rate of the target MFCC after dimensionality reduction and optimization obtained by analyzing the target signal by the voiceprint analysis server into the vector quantization algorithm, and finally obtain the quench determination result.

The present invention also provides various types of programmable processors (FPGA, ASIC or other integrated circuits), which are used for running programs, wherein the steps in the above embodiments are executed when the programs are running.

The present invention also provides a corresponding computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the memory implements the steps in the above embodiments when the program is executed.

Although the embodiments disclosed in the present invention are as above, the described contents are only the embodiments adopted to facilitate the understanding of the present invention, and are not intended to limit the present invention. Any person skilled in the art to which the present invention belongs, without departing from the spirit and principles disclosed by the present invention, can make any modifications, changes, equivalent replacements, etc. in the form and details of the implementation, which all belong to the present invention. protected range. Therefore, the scope of the patent protection of the present invention shall still be subject to the scope defined by the appended claims.

Claims (10)

1. A quench detection system utilizing an acoustic fiber, the system comprising: the system comprises an acousto-optic fiber wire, an optical fiber sound monitor and a voiceprint analysis server; one end of the acoustic optical fiber wire is wound on the outer ring of the superconducting magnet, and the other end of the acoustic optical fiber wire is connected to an optical transceiving port of the optical fiber sound monitor; the data communication port of the optical fiber sound monitor is connected with the voiceprint analysis server;

the optical signal that the optic fibre sound monitor sent has partial signal reflection to the optic fibre sound monitor through the rayleigh reflection, and optic fibre sound monitor restores the reflected light signal into sound signal, transmits to the voiceprint analysis server afterwards, carries out the voiceprint analysis to sound signal by the voiceprint analysis server.

2. The system of claim 1, wherein: the optical fiber sound monitor comprises a light emitting unit, a light receiving unit, an amplifying unit, a demodulating unit and a communication unit,

the optical signal is deformed when encountering the sound wave generated by the magnet in the process of moving in the optical fiber wire, the optical signal is reflected back to the optical fiber sound monitor by Rayleigh reflection and part of the signal is received by the optical receiving unit and the amplifying unit to obtain a reflected optical signal;

the demodulation unit restores the reflected light signal to a sound signal, and then the communication unit transmits the sound signal and the length information to the voiceprint analysis server.

3. The system of claim 2, wherein: the communication unit comprises an Ethernet communication unit, a WIFI module, a 4G module, a 5G module and the like.

4. The system of any one of claims 1-3, wherein: the voiceprint analysis server firstly performs weighted dimension reduction optimization on the voice signals based on the MFCC characteristic vectors, secondly identifies the optimized voice signals by applying a vector quantization algorithm, and finally judges whether the quench occurs.

5. A quench detection method using an acoustic fiber, using the system according to any of claims 1-4, characterized in that the method comprises the steps of:

step S1, the optical fiber sound monitor sends out optical signals;

step S2, the optical fiber sound monitor receives the reflected light signal and restores the reflected light signal into a sound signal;

s3, the voiceprint analysis server performs weighted dimension reduction optimization on the voice signal based on the MFCC feature vector;

and step S4, the voiceprint analysis server identifies the optimized voice signal by using a vector quantization algorithm, and finally judges whether the quench occurs.

6. The method according to claim 5, wherein the step S3 includes a voiceprint preprocessing step, the voiceprint preprocessing step including:

step 31, framing the voiceprint signal, wherein the framing relation is expressed as M = n-Lb/[ L (1-b) ], where M is the frame number, n is the audio signal length, L is the frame length, and b is the overlap ratio;

and step 32, respectively calculating MFCC characteristic vectors from the framing signals to form a characteristic vector group, wherein the calculation process comprises FFT (fast Fourier transform), Mel (Mel Filter), logarithmic transformation and discrete cosine transformation.

7. The method of claim 6, wherein: the Mel filtering is realized by a filter bank composed of a plurality of triangular band-pass filters, the number of the filters is p, and p parameters can be obtained after the signals are filtered

The calculation formula is as follows:

n is the number of FFT points, X (k) is the FFT of the preprocessed framed signal, H_i(k) Is the filter parameter, expressed as:

wherein: f [ i ] is the center frequency of the triangular filter;

and after mi is obtained through calculation, taking a logarithm of the mi, performing discrete cosine transform, and obtaining a result c (i) which is the MFCC feature vector of the framing signal.

8. The method according to claim 5 or 6, wherein the step S3 further comprises performing dimension reduction optimization on the preprocessed sound signals: specifically, the dimensionality reduction and simplification are carried out on the high-dimensional MFCC vector obtained through calculation through a PCA algorithm;

the PCA algorithm is as follows:

(1) e eigenvectors are set to form a matrix G, the dimension of each eigenvector is h, and G can be expressed as:

(2) calculating a correlation matrix R of G as

；

The eigenvalue of the correlation matrix R is calculated according to the above

And corresponding feature vectors

；

(3) Calculating variance contribution rate and accumulated variance contribution rate, which are respectively:

。

9. the method of claim 8, wherein the dimension-reduced optimized target MFCC feature vectors v, v1, …, vh, the variance contribution rate and the accumulated variance contribution rate obtained by analyzing the target signal by the voiceprint analysis server are input into a trained vector quantization algorithm, and finally the quench determination result is obtained.

10. The method of claim 7, wherein 20ms is taken as a frame length L, and 40% is taken as an overlap ratio b.

Images