Design method of low-power-consumption voiceprint sensor suitable for mechanical fault detection of high-voltage switch cabinet circuit breaker

Abstract

The invention discloses a low-power-consumption voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection. Comprising the following steps: S1, designing a high-sensitivity sensor structure; s2, micro-power consumption design of the high-sensitivity sensor; s3, evaluating the installation position of the high-sensitivity sensor; s4, a high-sensitivity sensor collects an audio signal; s5, processing the audio by a high-sensitivity sensor; s6, processing the audio signal; according to the high-sensitivity wireless audio sensor, random noise interference and electromagnetic interference in the environment are shielded, and the problems of random noise and electromagnetic interference are solved; the micro-power consumption technology is applied to the wireless audio sensor; carrying out near sound field reconstruction on the circuit breaker by using a discontinuous Galerkin finite element method so as to analyze a change rule of sound pressure in a sound field along with time; judging the size relationship between the energy value and a preset threshold value, and determining whether the equipment connection is in a normal state; and the extracted features are subjected to dimension reduction, so that the analysis accuracy and the calculation efficiency of the audio are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of low-power consumption voiceprint sensors, and in particular relates to a design method of a low-power consumption voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker. Background Art

[0002] The purpose of circuit breaker mechanical fault monitoring is to accurately determine the fault type or study and analyze the development trend of its operating status, so as to carry out reasonable status maintenance. Common monitoring characteristic parameters include circuit breaker opening (closing) coil current, main circuit current and contact stroke, etc. The fault type is generally determined based on threshold range comparison; and when one or several components of the operating mechanism have early defects, its weak dynamic information is difficult to reflect by the characteristic quantities extracted by the commonly used monitoring methods. Therefore, it is necessary to use a variety of monitoring methods and multi-dimensional state characteristic parameter fusion processing to identify the operating status of the circuit breaker.

[0003] Among the many circuit breaker mechanical condition monitoring methods, voiceprint signal analysis, as a non-invasive monitoring method, has gradually been favored by the industry because of its stable and reliable extraction of characteristic parameters. The method of circuit breaker voiceprint vibration monitoring is to obtain the voiceprint signal during its operation through a number of voiceprint sensors, obtain the characteristic information contained therein through signal processing and other means, and then use fault identification methods to locate the fault type or analyze the development trend of the circuit breaker operation status.

[0004] Realizing low power consumption in the true sense is a systematic project, and we must strengthen research and use appropriate methods and technologies at all links and levels. Usually, there are many performance indicators that are contradictory in implementation, such as speed, accuracy and load capacity. Sometimes, in order to obtain higher speed, accuracy and load capacity, it is necessary to further study and weigh indicators such as power consumption and speed according to actual conditions. Low power design is an issue that needs to be considered at the beginning of the design. Using targeted technologies to achieve power reduction will affect other design parameters, such as area and timing, and more importantly, it will have a significant impact on the design methodology. The impact of methodology includes architecture selection, verification, synthesis, testing and implementation stages. In the design of low-power systems, the demonstration and research of methods are necessary. For a macroscopic large application system, the system method, hardware method and software method are summarized; from a microscopic perspective, it can be as small as the composition of each unit circuit. Through many circuit models, control models, mathematical models and optimization of algorithms, the purpose of reducing power consumption can be achieved. For this reason, we propose a low-power voiceprint sensor design method suitable for mechanical fault detection of high-voltage switch cabinet circuit breakers. Summary of the invention

[0005] The purpose of the present invention is to provide a low-power consumption voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection, so as to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker, characterized in that it includes the following steps:

[0007] S1. High-sensitivity sensor structure design: The high-sensitivity sensor is designed based on the principle of stethoscope and air circulation. It will form a closed environment when used. The high-sensitivity sensor includes a cone-shaped sound collector, a mounting ring, an energy converter, an upper pressure piece and an isolator;

[0008] S2. Micro-power design of high-sensitivity sensors: The selected main control microcontroller has a sleep function, and the system is in sleep mode most of the time; low-power chips are used for integrated circuits; unused pins are grounded or high-level; pull-up or pull-down resistors are used as little as possible on IO pins; the power supply of each functional module in the circuit is reasonably divided and managed, and the microcontroller is used for real-time control; and the use of peripheral components is reduced;

[0009] S3. Evaluation of the installation position of high-sensitivity sensors: First, collect the circuit breaker sound signal to form the circuit breaker equivalent sound source; then use the discontinuous Galerkin finite element method to reconstruct the near sound field of the circuit breaker to analyze the temporal variation of the sound pressure in the sound field; finally, construct the Pearson equilibrium similarity coefficient to calculate the spectral similarity between the soundprint at the alternative measurement point and the equivalent sound source, so as to achieve quantitative evaluation of the soundprint monitoring position;

[0010] S4. High-sensitivity sensor collects audio signals: The cone-shaped sound collector fits the device to form a closed measurement environment that prevents the entry of external noise. The closed cavity allows all energy to be concentrated and pushed to the PVDF film.

[0011] S5. High-sensitivity sensor processing of audio: The sampling and compression process of audio signals are combined, and low-speed random sampling of audio signals is performed directly on the audio signal acquisition node at a frequency far below the Nyquist frequency. Low-speed random sampling of audio signals can greatly reduce the energy consumption of node data acquisition and transmission;

[0012] S6. Processing of audio signals: Detect the collected audio signals, then collect and preprocess the audio signals, and finally perform multi-scale feature extraction.

[0013] 2. According to claim 1, a low-power voiceprint sensor design method suitable for mechanical fault detection of high-voltage switch cabinet circuit breakers is characterized in that: the cone-shaped sound collector of the high-sensitivity sensor in S1 uses the gas compression principle to amplify the original audio signal, the sound collector is made of metal copper material, and the cable is a transmission line. When in use, a closed environment is formed, so that the collected audio has the advantages of stability and accuracy; there is a mounting hole on the top of the sound collector, which is used to fix the mounting ring, energy converter and upper pressing plate; the mounting ring and the upper pressing plate are made of polyformaldehyde thermoplastic crystalline polymer, the material has a hard texture and good insulation effect, is conducive to wireless signal transmission, and is easy to process and manufacture; the mounting ring is a circular concave structure as a whole, and the concave part is used to install the energy converter, and the diameter of the energy converter is equal to the diameter of the concave part of the mounting ring; the upper pressing plate is a circular convex structure as a whole. , the diameter of the convex part is equal to the diameter of the concave part of the mounting ring; the energy converter is placed in the concave structure of the mounting ring, and is embedded and squeezed and fixed by the convex part of the upper pressing plate; the connecting seam between the mounting ring and the upper pressing plate is sealed and fixed as a whole in the mounting hole at the top of the cone-shaped structure sound collector; there are corresponding groove structures in the mounting ring and the upper pressing plate for leading out cables, one end of the two cables is respectively connected to the upper and lower surfaces of the energy converter, and the other end is respectively connected to the positive and negative inputs of the amplifier in the hardware processing circuit; the isolator is made of polyformaldehyde thermoplastic crystalline polymer, the bottom is a hollow cylindrical structure, the top is a semi-hollow spherical structure, and is connected to the cone-shaped structure sound collector through threads; the hollow cylindrical structure is used to store the hardware processing circuit, and a vent is provided on the top of the semi-hollow spherical structure to ensure the fluidity of the surrounding air when the energy converter vibrates, and the isolator is connected to the sound collector through threads.

[0014] 3. According to claim 1, a low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection is characterized in that: the installation position evaluation of the high-sensitivity sensor in S3 is generally divided into signal measurement, sound field calculation, and measurement point analysis, wherein the signal measurement part is mainly to provide a sound field reconstruction data basis for the sound field calculation, and the result of the sound field calculation and the collected sound source data will be used together for the evaluation index calculation of the monitoring point;

[0015] The signal measurement is used to measure the distance space parameters between the circuit breaker surface and the switch cabinet; the circuit breaker surface is divided into a number of equivalent source areas, and the equivalent sound source signal is collected by a microphone; the circuit breaker equivalent sound source signal is phase corrected; the equivalent sound source signal is fast Fourier transformed to calculate the frequency domain information of the equivalent sound source;

[0016] The sound field calculation simplifies and constructs a three-dimensional model of the calculation domain based on measured information, and then divides the grid; defines the parameters of the calculation domain and defines the environmental boundary conditions based on the simplified model; uses the discontinuous Galerkin finite element method to calculate the transient change process of the sound field; extracts the sound pressure time domain signal of the candidate monitoring point, and performs fast Fourier transform to obtain its corresponding frequency domain information;

[0017] The measuring point analysis calculates the Pearson correlation matrix of the frequency spectrum between all candidate measuring points and the equivalent sound source; calculates the Pearson equilibrium correlation coefficient according to the calculation result of the Pearson correlation matrix; and selects the candidate monitoring point with the highest Pearson equilibrium correlation coefficient as the optimal monitoring point.

[0018] 4. According to claim 3, a low-power soundprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection is characterized in that: when the signal measurement is performed to collect the equivalent sound source, it is necessary to simultaneously obtain the circuit breaker surface sound signals of all measuring points, and adopt the phase correction method used in the operation deformation vibration mode measurement to correct the equivalent sound source signal; fix a measuring point as a reference signal, record the fixed phase angle relationship between the signals of other measuring points and the reference signal, and after all the collection work is completed, all measuring points are based on the anchor phase of the reference measuring point for phase correction;

[0019] Vibration and sound are essentially mechanical waves. Mechanical waves appear as vibration waves in the circuit breaker medium. The process of mechanical wave transmission to the air is relatively complicated, and additional vibrations will be excited when transmitted to the circuit breaker. As the final stage of mechanical wave transmission, in order to avoid errors in the sound field calculation caused by complex mechanical parameters and internal propagation processes, the near-acoustic field analysis of the circuit breaker is regarded as a transient calculation problem of the sound field with the circuit breaker surface as the vibration source and the air medium as the calculation domain.

[0020] 5. A low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection according to claim 4, characterized in that: the simplification of the calculation domain:

[0021] Simplification of acoustic field boundary: There is a boundary around the circuit breaker, which is simplified according to the hard acoustic field boundary and the hard acoustic field interface with sound-absorbing pores;

[0022] Principle of transient sound field calculation based on DGFEM:

[0023] DGFEM has obvious advantages in transient sound field analysis within a large spatial range. The spatial discretization form of the sound pressure conduction equation is:

[0024] Mp+Cp+Kp=Q(t),

[0025] Where: M is the fluid mass matrix; C is the fluid damping matrix; K is the fluid stiffness matrix; p is the fluid node pressure matrix; Q(t) is the sound source vector at each node; the time domain discretization of the sound wave propagation equation is: 0<… <t n <… <t N ;

[0026] For the discontinuity time t n , its step function is expressed as:

[0027]

[0028] Where:

[0029]

[0030] For any time step In, the sound pressure distribution function p is calculated using the following third-order Hermit interpolation function:

[0031]

[0032] Where p represents the node sound pressure at time t, and the expressions of A(t) and B(t) are:

[0033]

[0034]

[0035] The derivative v of the sound pressure distribution function is linearly interpolated within the time step:

[0036]

[0037] The sound source Q also uses linear interpolation:

[0038]

[0039] Based on the equilibrium equations and boundary conditions, the weak form of the discontinuous Galerkin finite element method for the sound wave propagation problem is obtained as follows:

[0040]

[0041] It is further transformed into a decoupled formula, which is the basic solution formula of the time domain discontinuous Galerkin finite element method:

[0042]

[0043] in,

[0044]

[0045] 6. A low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection according to claim 1, characterized in that: the audio signal detection in S6 includes energy value calculation and the relationship between the energy value and a preset threshold value;

[0046] Calculate the energy value E,

[0047] The audio energy value of the extracted signal to be detected is calculated; when the energy value E of the collected audio signal is calculated, it is as shown in the following formula:

[0048]

[0049] Among them, N represents the total number of frames of the time domain audio data of the collected audio signal; k represents the frame sequence number of one frame in the time domain data of the collected signal; M represents the number of time domain sampling point data of each frame in the time domain data information of the collected signal; the mathematical expression x(i) represents the obtained value of the i-th time domain sampling point data of each frame in the time domain data of the collected signal.

[0050] 7. A low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection according to claim 1, characterized in that: the relationship between the energy value and the preset threshold is:

[0051] If the calculated energy value E is less than the preset energy threshold Eth, the device connection is in an abnormal state, and an alarm signal is generated to alarm;

[0052] If the calculated E≥Eth, first convert the N frames of time domain data information of the collected audio signal into frequency domain data information by corresponding methods for detection; the spectrum peak frequency of each frame in the N frames of frequency domain data information must be extracted, that is, the amplitude of each frame of the frequency domain information data after the conversion contains P frequency domain sampling points. When the frequency of the frequency domain sampling point is 16kHz, P is 1024; at the same time, record the frequency value with the largest amplitude among the P sampling points, and record this point as the spectrum peak frequency of the N frames of frequency domain data information; determine whether the spectrum peak frequency of the frequency domain audio data information remains stable based on the calculation results, and the calculation of the first stability value Dif of the spectrum peak frequency of the frequency domain data information is as follows:

[0053]

[0054] The second stability value St of the spectrum peak frequency is as follows:

[0055]

[0056] Wherein, N represents the total number of frames of the collected audio signal; k represents the frame sequence number of a certain frame of data in the time domain audio of the collected audio signal; index(k) represents the kth frame of frequency domain audio data in the data information of the collected audio signal; index(k+1) represents the spectrum peak frequency of the k+1th frame of data information in the data information of the collected audio signal;

[0057] Compare the calculated Dif with the first stability preset threshold, and the calculated St with the second stability preset threshold in turn; if the calculated Dif value is less than the first stability threshold Difth, and the calculated St value is less than the second stability threshold Stth, then the spectrum peak frequency of the collected audio signal is definitely in a stable state, which means that the device connection is abnormal, and an alarm signal is generated at the same time.

[0058] 8. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 1, characterized in that: the data collection and preprocessing steps in S6 are as follows:

[0059] The preprocessing process mainly includes four steps: pre-emphasis, framing, windowing and speed matching; among them, pre-emphasis processing is mainly to compensate for the loss of high-frequency components; framing is to facilitate local batch processing of audio data; when framing, each frame signal must contain at least one rotation cycle of the component, and the frame shift cannot exceed half of the frame length; due to the non-periodic truncation of the signal, the spectrum will cause serious tailing in the entire frequency band, resulting in very serious errors, called leakage. In order to make each frame signal after framing still meet the continuity and reduce the impact of truncation effect, each frame data needs to be multiplied by a window function, that is, windowing processing; and the Hamming window is used.

[0060] 9. A low-power voiceprint sensor design method suitable for mechanical fault detection of high-voltage switch cabinet circuit breakers according to claim 1, characterized in that: the multi-scale feature extraction in S6 includes performing comprehensive and multi-scale feature extraction on time domain features, frequency domain features, and cepstrum domain features, and then performing dimensionality reduction on the extracted features to facilitate screening out effective features;

[0061] The calculation formula of time domain characteristics is as follows:

[0062]

[0063] Where: y i (n) The signal amplitude of the nth point of the audio in the i-th frame after framing; L represents the frame length; n represents the point number of each frame; sgn represents the sign function;

[0064] Among them, the sgn symbol function is defined as:

[0065]

[0066] Frequency domain features are used to convert time domain signals to frequency domain. They are defined as the signal power within a unit frequency band and intuitively show the functional relationship between signal variance and frequency.

[0067] The cepstral domain feature is the Mel cepstral coefficient feature, which is a perceptual frequency domain cepstral coefficient. Based on the human auditory mechanism, a group of H triangular bandpass filter groups are arranged from low frequency to high frequency in the Mel scale frequency domain according to the size of the critical bandwidth. The number of filter groups and the number of critical bands must be close, generally 22-26.

[0068] After preprocessing the input signal, Fourier transform is performed, and then it passes through the filter group, and the logarithmic energy of each filter group output is calculated, and then the MFCC coefficients can be obtained by discrete cosine transform;

[0069] MFCC feature calculation formula:

[0070]

[0071] Where: S represents the Mel filter energy, kg / s 2 ; The subscript i represents the frame number; h represents the Mel filter number; n represents the spectrum line after discrete cosine transform;

[0072] The invention is used for automatic gain control of voiceprint signal, and mainly consists of two parts: a gain amplifier and a feedback loop. The gain amplifier component uses a certain AGC algorithm to adjust the gain value according to the amplitude and threshold of the feedback loop in the system. That is, the AGC adjusts the automatic control of the signal by the strength of the automatic gain signal in the amplifying circuit. When the signal strength increases, the control of the feedback loop of the AGC system is reduced accordingly according to a certain relationship. On the contrary, when the signal amplitude in the gain amplifier of the AGC system decreases, the gain of the feedback loop will be increased accordingly according to a certain relationship between the two. Through this method, after the amplification and adjustment of the AGC algorithm, it is ensured that the amplitude of the signal output of the communication system can be basically maintained in a constant state.

[0073] 10. A method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 9, characterized in that the automatic gain control step is as follows:

[0074] S601: Determine the dynamic range of the audio signal output: The design is based on the audio signal processing using the AGC algorithm, and needs to be determined according to the actual situation of the audio signal; if in the design process, the expected value in the practical AGC system is a fixed value, then the amplitude of the audio signal output will approach the expected value of the AGC system; in the process of approaching the expected value, the output signal will be continuously adjusted, thereby causing the amplitude of the signal output end to be unstable. In order to ensure the stability of the audio signal output, based on the AGC expected value, a dynamic range of stable audio signal output is designed and determined with the expected value as the center; when the amplitude of the audio signal is within this dynamic range during the AGC algorithm adjustment, it is determined that the output amplitude of the audio signal is stable;

[0075] S602: Determine the gain adjustment speed: In order to determine, ensure and present the normal change trend of the voice signal amplitude, it is necessary to ensure that its gain remains unchanged or changes within a smaller range, thereby ensuring that the amplitude of the audio signal is not distorted during transmission; the interference received during the audio signal transmission process makes the average amplitude of the audio signal relatively small, so it is necessary to increase its gain value to ensure that the overall amplitude of the signal increases, thereby ensuring that the audio signal is not distorted; the application of practical AGC in audio signals, its gain value needs to be slow and fast, and change suddenly and quickly with the slow change of the signal amplitude; when the gain adjustment time of the audio signal is longer than the adjustment time of the audio signal under normal circumstances, the gain value at this time will not change significantly, and the gain adjustment time needs to be set to 4s according to the characteristics of the audio signal output; if the audio signal output is large, the gain at this time needs to be reduced quickly, otherwise it will cause damage to the device; and when the output value of the signal is greater than the upper limit of the expected value, the gain value adjustment time needs to be set to 0.5ms;

[0076] The gain value calculation formula of the practical AGC algorithm is:

[0077]

[0078] Among them, A(n+1) is the gain value after signal transmission adjustment; A(n) is the gain value before adjustment; A in is the signal amplitude value; A0 is the expected amplitude value; 2 -a is a constant; different a values ​​will result in different AGC adjustment speeds. According to the actual requirements of the practical AGC algorithm, when the signal output amplitude is less than the upper limit of the expected value, a=16, the gain adjustment will be slower. If a=3, the gain adjustment speed will be faster.

[0079] S603: Determine the dynamic range of the input signal: During the audio signal input process, if the input value range is set too large, the signal will generate greater noise during transmission, affecting the input and output quality of the signal; on the contrary, if the set range of the audio input signal is smaller, the signal will be ignored due to its small size, thereby causing distortion of the audio input signal; according to the input time of the audio signal, a signal strength of 36dB is determined as noise; using the AGC algorithm, the input dynamic range of the audio signal is determined, and the noise can be judged whether it occurs in the gap of the audio signal transmission according to the existence time of the noise. If the existence time of the noise is relatively short, it can be regarded as the gap of the audio signal; usually, the judgment time of the noise is set to 5s for the best; according to the existence time of the noise and the actual situation of the signal input, the range of about 6dB between the minimum value of the audio signal and the noise threshold is determined as the dynamic range of the audio signal input; within this range, the interference intensity of the noise on the audio signal is relatively high, so the gain value needs to be fixed;

[0080] S604: Limit the gain value: To avoid excessive gain and burnout of signal transmission components or equipment, the gain value needs to be controlled within a range. If the value calculated by the AGC is greater than the upper limit of the range, the gain value takes the maximum value, otherwise it takes the minimum value. According to the actual transmission of the audio signal, the gain range is -3-30dB.

[0081] Compared with the prior art, the present invention has the following beneficial effects:

[0082] The cone-structured sound collector of the high-sensitivity wireless audio sensor of the present invention utilizes the principle of gas compression to amplify the original audio signal. The sound collector is made of metal copper material, and the cable is a transmission line. When in use, a closed environment is formed to shield random noise interference and electromagnetic interference in the environment, thereby overcoming the problems of random noise and electromagnetic interference, and improving the signal-to-noise ratio of the original collected signal, so that the collected audio has the advantages of stability and accuracy; the polyvinylidene fluoride piezoelectric film is the core of the high-sensitivity audio signal sensor, which converts the sound signal into an electrical signal; the material is light and thin, soft, and chemically stable, with strong toughness, and is easy to be processed into various shapes, and the production is simple and cost-effective. Low cost; PVDF material has a wide response frequency range of 0-400MHz. The low frequency limit of traditional microphones is 20Hz. The use of PVDF material can obtain more low-frequency information; in addition, its acoustic impedance is similar to that of human tissue, and it is often used in the signal acquisition structure of precision instruments and equipment; when the surrounding air vibrates, the PVDF piezoelectric film vibrates accordingly, and an electrical signal is generated between the upper and lower surfaces of the film; when the static film voltage is used as a reference, the upper and lower surfaces of the film will generate charges of equal magnitude and opposite polarity, and the signal will be sent to the input end of the differential amplifier; due to the symmetry of the original signal, the common-mode noise in the two signals is offset after the differential mode operation;

[0083] The power consumption of highly sensitive wireless audio sensors is reduced through a variety of hardware, and micro-power consumption technology is applied to wireless audio sensors, that is, the sampling and compression processes of audio signals are merged, and low-speed random sampling of audio signals is performed directly on the audio signal acquisition node at a frequency far below the Nyquist frequency, and signal reconstruction is performed on the user side with strong computing and storage capabilities; under the premise of ensuring the quality of signal reconstruction, low-speed random sampling of audio signals can greatly reduce the energy consumption of node data acquisition and transmission, thereby effectively solving the network congestion and high energy consumption problems caused by large data volumes in wireless audio sensor networks;

[0084] The near-sound field of the circuit breaker is reconstructed using the discontinuous Galerkin finite element method to analyze the temporal variation of the sound pressure in the sound field. Finally, the Pearson equilibrium similarity coefficient is constructed to calculate the spectral similarity between the soundprint at the alternative measurement point and the equivalent sound source, thus achieving quantitative evaluation of the soundprint monitoring position.

[0085] By calculating the energy value and judging the relationship between the energy value and the preset threshold, it is determined whether the device connection is in a normal state; and before analyzing the audio signal data, it is necessary to first ensure that valid data is collected; unified standards must be achieved during the collection process to reduce errors caused by various factors; audio feature extraction is extremely important. According to the characteristics of the sound signal, comprehensive and multi-scale feature extraction is performed from the time domain, frequency domain, cepstrum domain, etc., and a comprehensive analysis is performed, and then the extracted features are reduced in dimension to facilitate the screening of effective features and improve the analysis accuracy and calculation efficiency of the audio. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is a schematic diagram of the steps of the present invention;

[0087] Figure 2 This is a structural principle diagram of a highly sensitive sensor of the present invention;

[0088] Figure 3 The circuit breaker acoustic field reconstruction and sensor arrangement evaluation process of the present invention;

[0089] Figure 4 It is a schematic diagram of the detection process of the audio signal acquisition device of the present invention;

[0090] Figure 5 It is a schematic diagram of a triangular filter group from dense to sparse in the Mel scale frequency domain of the present invention. DETAILED DESCRIPTION

[0091] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0092] See also Figure 1-Figure 5 The present invention provides a suitable technical solution: a low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection, comprising the following steps:

[0093] S1. High-sensitivity sensor structure design: The high-sensitivity sensor is designed based on the principle of stethoscope and air circulation. It will form a closed environment when used. The high-sensitivity sensor includes a cone-shaped sound collector, a mounting ring, an energy converter, an upper pressure piece and an isolator;

[0094] S2. Micro-power design of high-sensitivity sensors: The selected main control microcontroller has a sleep function, and the system is in sleep mode most of the time; low-power chips are used for integrated circuits; unused pins are grounded or high-level; pull-up or pull-down resistors are used as little as possible on IO pins; the power supply of each functional module in the circuit is reasonably divided and managed, and the microcontroller is used for real-time control; and the use of peripheral components is reduced;

[0095] S3. Evaluation of the installation position of high-sensitivity sensors: First, collect the circuit breaker sound signal to form the circuit breaker equivalent sound source; then use the discontinuous Galerkin finite element method to reconstruct the near sound field of the circuit breaker to analyze the temporal variation of the sound pressure in the sound field; finally, construct the Pearson equilibrium similarity coefficient to calculate the spectral similarity between the soundprint at the alternative measurement point and the equivalent sound source, so as to achieve quantitative evaluation of the soundprint monitoring position;

[0096] S4. High-sensitivity sensor collects audio signals: The cone-shaped sound collector fits the device to form a closed measurement environment that prevents the entry of external noise. The closed cavity allows all energy to be concentrated and pushed to the PVDF film.

[0097] S5. High-sensitivity sensor processing of audio: The sampling and compression process of audio signals are combined, and low-speed random sampling of audio signals is performed directly on the audio signal acquisition node at a frequency far below the Nyquist frequency. Low-speed random sampling of audio signals can greatly reduce the energy consumption of node data acquisition and transmission;

[0098] S6. Processing of audio signals: Detect the collected audio signals, then collect and preprocess the audio signals, and finally perform multi-scale feature extraction.

[0099] The closed measurement environment will prevent the entry of external noise and greatly reduce the impact of external random noise. Due to the closed cavity, all energy is concentrated and pushed to the PVDF film. The signal collected by the relatively large collection surface acts on the relatively small film, which itself has physical amplification. The energy loss in a closed environment is very small, making it easier to collect and detect smaller audio signals. In a complex electromagnetic environment, power frequency interference (50Hz) is the most significant. The design of the whole body copper material is to shield this interference as much as possible and create a small electromagnetic interference-free environment. In order to reduce the antenna effect of the signal transmission line, the wires connected on both sides of the PVDF film are all coaxial cables. In this embodiment, preferably, the cone-shaped sound collector of the high-sensitivity sensor in S1 uses the gas compression principle to amplify the original audio signal. The sound collector is made of metal copper material, and the cable is a transmission line. When used, a closed environment is formed, so that the collected audio has the advantages of stability and accuracy; there is a mounting hole on the top of the sound collector to fix the mounting ring, energy converter and upper pressing plate; the mounting ring and the upper pressing plate are made of polyformaldehyde thermoplastic crystalline polymer, the material has a hard texture and good insulation effect, is conducive to wireless signal transmission, and is easy to process and manufacture; the mounting ring is a circular concave structure as a whole, and the concave part is used to install the energy converter, and the diameter of the energy converter is equal to the diameter of the concave part of the mounting ring; the upper pressing plate is a circular convex structure as a whole, The diameter of the convex part is equal to the diameter of the concave part of the mounting ring; the energy converter is placed in the concave structure of the mounting ring and is fixed by embedding and squeezing the convex part of the upper pressing plate; the connecting seam between the mounting ring and the upper pressing plate is sealed and fixed as a whole in the mounting hole at the top of the cone-shaped structure sound collector; there are corresponding groove structures in the mounting ring and the upper pressing plate for leading out cables, one end of the two cables is respectively connected to the upper and lower surfaces of the energy converter, and the other end is respectively connected to the positive and negative inputs of the amplifier in the hardware processing circuit; the isolator is made of polyformaldehyde thermoplastic crystalline polymer, with a hollow cylindrical structure at the bottom and a semi-hollow spherical structure at the top, which is connected to the cone-shaped structure sound collector through threads; the hollow cylindrical structure is used to store the hardware processing circuit, and a vent is provided on the top of the semi-hollow spherical structure to ensure the fluidity of the surrounding air when the energy converter vibrates, and the isolator is connected to the sound collector through threads.

[0100] In order to explore the distribution law of the near-acoustic field of the circuit breaker and optimize the position of the soundprint monitoring point, the circuit breaker sound signal is first collected to form an equivalent sound source of the circuit breaker; then the discontinuous Galerkin finite element method is used to reconstruct the near-acoustic field of the circuit breaker to analyze the law of sound pressure variation in the sound field over time; finally, the Pearson equilibrium similarity coefficient is constructed to calculate the spectral similarity between the soundprint and the equivalent sound source at the alternative measuring point, so as to achieve quantitative evaluation of the soundprint monitoring position. This method can evaluate the perception ability of the sensor to the overall soundprint state of the circuit breaker when it is in different positions, and has a certain reference value for the deployment of the circuit breaker online monitoring device. In this embodiment, preferably, the installation position evaluation of the high-sensitivity sensor in S3 is divided into signal measurement, sound field calculation, and measurement point analysis. The signal measurement part mainly provides the sound field reconstruction data basis for the sound field calculation, and the result of the sound field calculation and the collected sound source data will be used together for the calculation of the evaluation index of the monitoring point;

[0101] The signal measurement is used to measure the distance space parameters between the circuit breaker surface and the switch cabinet; the circuit breaker surface is divided into a number of equivalent source areas, and the equivalent sound source signal is collected by a microphone; the circuit breaker equivalent sound source signal is phase corrected; the equivalent sound source signal is fast Fourier transformed to calculate the frequency domain information of the equivalent sound source;

[0102] The sound field calculation simplifies and constructs a three-dimensional model of the calculation domain based on measured information, and then divides the grid; defines the parameters of the calculation domain and defines the environmental boundary conditions based on the simplified model; uses the discontinuous Galerkin finite element method to calculate the transient change process of the sound field; extracts the sound pressure time domain signal of the candidate monitoring point, and performs fast Fourier transform to obtain its corresponding frequency domain information;

[0103] The measuring point analysis calculates the Pearson correlation matrix of the frequency spectrum between all candidate measuring points and the equivalent sound source; calculates the Pearson equilibrium correlation coefficient according to the calculation result of the Pearson correlation matrix; and selects the candidate monitoring point with the highest Pearson equilibrium correlation coefficient as the optimal monitoring point.

[0104] It is necessary to obtain the circuit breaker surface acoustic signals of all measuring points at the same time. However, due to the limitation of the number of microphones, only signals of limited measuring points can be collected at the same time. In order to avoid the error caused by complex mechanical parameters and internal propagation process to the sound field calculation, in this embodiment, preferably, when the signal measurement is performing the acquisition of the equivalent sound source, the circuit breaker surface acoustic signals of all measuring points need to be obtained at the same time, and the equivalent sound source signal is corrected by the phase correction method used in the operation deformation vibration mode measurement; a measuring point is fixed as the reference signal, and the fixed phase angle relationship between the signals of other measuring points and the reference signal is recorded. After all the acquisition work is completed, all measuring points are subjected to phase correction based on the anchor phase of the reference measuring point;

[0105] Vibration and sound are essentially mechanical waves. Mechanical waves appear as vibration waves in the circuit breaker medium. The process of mechanical wave transmission to the air is relatively complicated, and additional vibrations will be excited when transmitted to the circuit breaker. As the final stage of mechanical wave transmission, in order to avoid errors in the sound field calculation caused by complex mechanical parameters and internal propagation processes, the near-acoustic field analysis of the circuit breaker is regarded as a transient calculation problem of the sound field with the circuit breaker surface as the vibration source and the air medium as the calculation domain.

[0106] The vibration characteristics at different positions on the circuit breaker surface are different. The circuit breaker surface cannot be simply regarded as a whole, but its spatial differences need to be fully considered. According to the equivalent source concept, the circuit breaker surface is divided into several sub-sound sources with independent amplitudes and phases. The sound pressure signal at the center point of the sub-sound source area measured experimentally is used as the sound signal on the entire surface of the sub-sound source. The combined effect of all sub-sound sources can be equivalent to simulating the sound source of the air domain sound field. Theoretically, when the equivalent source is small enough, the sound source state generated on the transformer surface can be completely restored. However, in the actual sound signal acquisition process, only a limited number of point source signals can be sampled. Therefore, it is necessary to simplify the entire circuit breaker surface into a limited number of sub-sound source surfaces according to actual conditions. In this embodiment, preferably, the simplification of the calculation domain is:

[0107] Simplification of acoustic field boundary: There is a boundary around the circuit breaker, which is simplified according to the hard acoustic field boundary and the hard acoustic field interface with sound-absorbing pores;

[0108] Principle of transient sound field calculation based on DGFEM:

[0109] DGFEM has obvious advantages in transient sound field analysis within a large spatial range. The spatial discretization form of the sound pressure conduction equation is:

[0110] Mp+Cp+Kp=Q(t),

[0111] Where: M is the fluid mass matrix; C is the fluid damping matrix; K is the fluid stiffness matrix; p is the fluid node pressure matrix; Q(t) is the sound source vector at each node; the time domain discretization of the sound wave propagation equation is: 0<… <t n <… <t N ;

[0112] For the discontinuity time t n , its step function is expressed as:

[0113]

[0114] Where:

[0115]

[0116] For any time step In, the sound pressure distribution function p is calculated using the following third-order Hermit interpolation function:

[0117]

[0118] Where p represents the node sound pressure at time t, and the expressions of A(t) and B(t) are:

[0119]

[0120]

[0121] The derivative v of the sound pressure distribution function is linearly interpolated within the time step:

[0122]

[0123] The sound source Q also uses linear interpolation:

[0124]

[0125] Based on the equilibrium equations and boundary conditions, the weak form of the discontinuous Galerkin finite element method for the sound wave propagation problem is obtained as follows:

[0126]

[0127] It is further transformed into a decoupled formula, which is the basic solution formula of the time domain discontinuous Galerkin finite element method:

[0128]

[0129] in,

[0130]

[0131] In order to calculate the magnitude of the audio energy value in the signal to be detected, in this embodiment, preferably, the detection of the audio signal in S6 includes energy value calculation and the magnitude relationship between the energy value and a preset threshold value;

[0132] Calculate the energy value E,

[0133] The audio energy value of the extracted signal to be detected is calculated; when the energy value E of the collected audio signal is calculated, it is as shown in the following formula:

[0134]

[0135] Among them, N represents the total number of frames of the time domain audio data of the collected audio signal; k represents the frame sequence number of one frame in the time domain data of the collected signal; M represents the number of time domain sampling point data of each frame in the time domain data information of the collected signal; the mathematical expression x(i) represents the obtained value of the i-th time domain sampling point data of each frame in the time domain data of the collected signal.

[0136] In order to detect whether the device connection is abnormal and generate an alarm signal, in this embodiment, preferably, the relationship between the energy value and the preset threshold is:

[0137] If the calculated energy value E is less than the preset energy threshold Eth, the device connection is in an abnormal state, and an alarm signal is generated to alarm;

[0138] If the calculated E≥Eth, first convert the N frames of time domain data information of the collected audio signal into frequency domain data information by corresponding methods for detection; the spectrum peak frequency of each frame in the N frames of frequency domain data information must be extracted, that is, the amplitude of each frame of the frequency domain information data after the conversion contains P frequency domain sampling points. When the frequency of the frequency domain sampling point is 16kHz, P is 1024; at the same time, record the frequency value with the largest amplitude among the P sampling points, and record this point as the spectrum peak frequency of the N frames of frequency domain data information; determine whether the spectrum peak frequency of the frequency domain audio data information remains stable based on the calculation results, and the calculation of the first stability value Dif of the spectrum peak frequency of the frequency domain data information is as follows:

[0139]

[0140] The second stability value St of the spectrum peak frequency is as follows:

[0141]

[0142] Wherein, N represents the total number of frames of the collected audio signal; k represents the frame sequence number of a certain frame of data in the time domain audio of the collected audio signal; index(k) represents the kth frame of frequency domain audio data in the data information of the collected audio signal; index(k+1) represents the spectrum peak frequency of the k+1th frame of data information in the data information of the collected audio signal;

[0143] Compare the calculated Dif with the first stability preset threshold, and the calculated St with the second stability preset threshold in turn; if the calculated Dif value is less than the first stability threshold Difth, and the calculated St value is less than the second stability threshold Stth, then the spectrum peak frequency of the collected audio signal is definitely in a stable state, which means that the device connection is abnormal, and an alarm signal is generated at the same time.

[0144] In order to compensate for the loss of high-frequency components; the framing is to more conveniently perform local batch processing on the audio data. In this embodiment, preferably, the data collection and preprocessing steps in S6 are as follows:

[0145] The preprocessing process mainly includes four steps: pre-emphasis, framing, windowing and speed matching; among them, pre-emphasis processing is mainly to compensate for the loss of high-frequency components; framing is to facilitate local batch processing of audio data; when framing, each frame signal must contain at least one rotation cycle of the component, and the frame shift cannot exceed half of the frame length; due to the non-periodic truncation of the signal, the spectrum will cause serious tailing in the entire frequency band, resulting in very serious errors, called leakage. In order to make each frame signal after framing still meet the continuity and reduce the impact of truncation effect, each frame data needs to be multiplied by a window function, that is, windowing processing; and the Hamming window is used.

[0146] In the field of audio recognition, it is extremely important to extract audio features. The extracted features are comprehensively analyzed, and then the extracted features are reduced in dimension to facilitate screening out effective features. In this embodiment, preferably, the multi-scale feature extraction in S6 includes time domain features, frequency domain features, and cepstrum domain features to extract all aspects and multi-scale features, comprehensively analyze them, and then reduce the extracted features in dimension to facilitate screening out effective features;

[0147] The calculation formula of time domain characteristics is as follows:

[0148]

[0149] Where: y i (n) The signal amplitude of the nth point of the audio in the i-th frame after framing; L represents the frame length; n represents the point number of each frame; sgn represents the sign function;

[0150] Among them, the sgn symbol function is defined as:

[0151]

[0152] Frequency domain features are used to convert time domain signals to frequency domain. They are defined as the signal power within a unit frequency band and intuitively show the functional relationship between signal variance and frequency.

[0153] The cepstral domain feature is the Mel cepstral coefficient feature, which is a perceptual frequency domain cepstral coefficient. Based on the human auditory mechanism, a group of H triangular bandpass filter groups are arranged from low frequency to high frequency in the Mel scale frequency domain according to the size of the critical bandwidth. The number of filter groups and the number of critical bands must be close, generally 22-26.

[0154] After preprocessing the input signal, Fourier transform is performed, and then it passes through the filter group, and the logarithmic energy of each filter group output is calculated, and then the MFCC coefficients can be obtained by discrete cosine transform;

[0155] MFCC feature calculation formula:

[0156]

[0157] Where: S represents the Mel filter energy, kg / s 2 ; The subscript i represents the frame number; h represents the Mel filter number; n represents the spectrum line after discrete cosine transform;

[0158] The invention is used for automatic gain control of voiceprint signal, and mainly consists of two parts: a gain amplifier and a feedback loop. The gain amplifier component uses a certain AGC algorithm to adjust the gain value according to the amplitude and threshold of the feedback loop in the system. That is, the AGC adjusts the automatic control of the signal by the strength of the automatic gain signal in the amplifying circuit. When the signal strength increases, the control of the feedback loop of the AGC system is reduced accordingly according to a certain relationship. On the contrary, when the signal amplitude in the gain amplifier of the AGC system decreases, the gain of the feedback loop will be increased accordingly according to a certain relationship between the two. Through this method, after the amplification and adjustment of the AGC algorithm, it is ensured that the amplitude of the signal output of the communication system can be basically maintained in a constant state.

[0159] In order to effectively reduce the interference when the audio signal is output and ensure the stability of the signal, in this embodiment, preferably, the steps of the automatic gain control are as follows:

[0160] S601: Determine the dynamic range of the audio signal output: The design is based on the audio signal processing using the AGC algorithm, and needs to be determined according to the actual situation of the audio signal; if in the design process, the expected value in the practical AGC system is a fixed value, then the amplitude of the audio signal output will approach the expected value of the AGC system; in the process of approaching the expected value, the output signal will be continuously adjusted, thereby causing the amplitude of the signal output end to be unstable. In order to ensure the stability of the audio signal output, based on the AGC expected value, a dynamic range of stable audio signal output is designed and determined with the expected value as the center; when the amplitude of the audio signal is within this dynamic range during the AGC algorithm adjustment, it is determined that the output amplitude of the audio signal is stable;

[0161] S602: Determine the gain adjustment speed: In order to determine, ensure and present the normal change trend of the voice signal amplitude, it is necessary to ensure that its gain remains unchanged or changes within a smaller range, thereby ensuring that the amplitude of the audio signal is not distorted during transmission; the interference received during the audio signal transmission process makes the average amplitude of the audio signal relatively small, so it is necessary to increase its gain value to ensure that the overall amplitude of the signal increases, thereby ensuring that the audio signal is not distorted; the application of practical AGC in audio signals, its gain value needs to be slow and fast, and change suddenly and quickly with the slow change of the signal amplitude; when the gain adjustment time of the audio signal is longer than the adjustment time of the audio signal under normal circumstances, the gain value at this time will not change significantly, and the gain adjustment time needs to be set to 4s according to the characteristics of the audio signal output; if the audio signal output is large, the gain at this time needs to be reduced quickly, otherwise it will cause damage to the device; and when the output value of the signal is greater than the upper limit of the expected value, the gain value adjustment time needs to be set to 0.5ms;

[0162] The gain value calculation formula of the practical AGC algorithm is:

[0163]

[0164] Among them, A(n+1) is the gain value after signal transmission adjustment; A(n) is the gain value before adjustment; A in is the signal amplitude value; A0 is the expected amplitude value; 2 -a is a constant; different a values ​​will result in different AGC adjustment speeds. According to the actual requirements of the practical AGC algorithm, when the signal output amplitude is less than the upper limit of the expected value, a=16, the gain adjustment will be slower. If a=3, the gain adjustment speed will be faster.

[0165] S603: Determine the dynamic range of the input signal: During the audio signal input process, if the input value range is set too large, the signal will generate greater noise during transmission, affecting the input and output quality of the signal; on the contrary, if the set range of the audio input signal is smaller, the signal will be ignored due to its small size, thereby causing distortion of the audio input signal; according to the input time of the audio signal, a signal strength of 36dB is determined as noise; using the AGC algorithm, the input dynamic range of the audio signal is determined, and the noise can be judged whether it occurs in the gap of the audio signal transmission according to the existence time of the noise. If the existence time of the noise is relatively short, it can be regarded as the gap of the audio signal; usually, the judgment time of the noise is set to 5s for the best; according to the existence time of the noise and the actual situation of the signal input, the range of about 6dB between the minimum value of the audio signal and the noise threshold is determined as the dynamic range of the audio signal input; within this range, the interference intensity of the noise on the audio signal is relatively high, so the gain value needs to be fixed;

[0166] S604: Limit the gain value: To avoid excessive gain and burnout of signal transmission components or equipment, the gain value needs to be controlled within a range. If the value calculated by the AGC is greater than the upper limit of the range, the gain value takes the maximum value, otherwise it takes the minimum value. According to the actual transmission of the audio signal, the gain range is -3-30dB.

[0167] The working principle and use process of the present invention:

[0168] Step 1: High-sensitivity sensor structure design: The high-sensitivity sensor is designed based on the principle of stethoscope and air circulation. It will form a closed environment when used. The high-sensitivity sensor includes a cone-shaped sound collector, a mounting ring, an energy converter, an upper pressure piece and an isolator;

[0169] Step 2: Micro-power design of high-sensitivity sensor: The selected main control microcontroller has a sleep function, and the system is in sleep state most of the time; low-power chips are used for integrated circuits; unused pins are grounded or high-level; pull-up or pull-down resistors are used as little as possible on IO pins; the power supply of each functional module in the circuit is reasonably divided and managed, and the microcontroller is used for real-time control; and the use of peripheral components is reduced;

[0170] Step 3: Evaluation of the installation position of the high-sensitivity sensor: First, collect the circuit breaker sound signal to form the circuit breaker equivalent sound source; then use the discontinuous Galerkin finite element method to reconstruct the near sound field of the circuit breaker to analyze the time-varying law of the sound pressure in the sound field; finally, construct the Pearson equilibrium similarity coefficient to calculate the spectral similarity between the soundprint at the alternative measuring point and the equivalent sound source, so as to achieve quantitative evaluation of the soundprint monitoring position;

[0171] Step 4: High-sensitivity sensor collects audio signals: The cone-shaped sound collector is attached to the device to form a closed measurement environment that prevents the entry of external noise. The closed cavity allows all energy to be concentrated and pushed to the PVDF film.

[0172] Step 5: Processing of audio by high-sensitivity sensors: The sampling and compression processes of audio signals are combined, and low-speed random sampling of audio signals is performed directly on the audio signal acquisition node at a frequency far below the Nyquist frequency. Low-speed random sampling of audio signals can greatly reduce the energy consumption of node data acquisition and transmission.

[0173] Step 6: Processing of audio signals: Detect the collected audio signals, then collect and preprocess the audio signals, and finally perform multi-scale feature extraction.

[0174] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection, characterized in that: The following steps are included: S1. High-sensitivity sensor structure design: The high-sensitivity sensor is designed based on the principle of stethoscope and air circulation. It will form a closed environment when used. The high-sensitivity sensor includes a cone-shaped sound collector, a mounting ring, an energy converter, an upper pressure piece and an isolator; S2. Micro-power design of high-sensitivity sensors: The selected main control microcontroller has a sleep function, and the system is in sleep mode most of the time; low-power chips are used for integrated circuits; unused pins are grounded or high-level; pull-up or pull-down resistors are used as little as possible on IO pins; the power supply of each functional module in the circuit is reasonably divided and managed, and the microcontroller is used for real-time control; and the use of peripheral components is reduced; S3. Evaluation of the installation position of high-sensitivity sensors: First, collect the circuit breaker sound signal to form the circuit breaker equivalent sound source; then use the discontinuous Galerkin finite element method to reconstruct the near sound field of the circuit breaker to analyze the temporal variation of the sound pressure in the sound field; finally, construct the Pearson equilibrium similarity coefficient to calculate the spectral similarity between the soundprint at the alternative measurement point and the equivalent sound source, so as to achieve quantitative evaluation of the soundprint monitoring position; S4. High-sensitivity sensor collects audio signals: The cone-shaped sound collector fits the device to form a closed measurement environment that prevents the entry of external noise. The closed cavity allows all energy to be concentrated and pushed to the PVDF film. S5. High-sensitivity sensor processing of audio: The sampling and compression process of audio signals are combined, and low-speed random sampling of audio signals is performed directly on the audio signal acquisition node at a frequency far below the Nyquist frequency. Low-speed random sampling of audio signals can greatly reduce the energy consumption of node data acquisition and transmission; S6. Processing of audio signals: Detect the collected audio signals, then collect and preprocess the audio signals, and finally perform multi-scale feature extraction.

2. According to claim 1, a low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection is characterized by: The cone-structured sound collector of the high-sensitivity sensor in S1 uses the principle of gas compression to amplify the original audio signal. The sound collector is made of metal copper material, and the cable is the transmission line. When in use, a closed environment is formed, so that the collected audio has the advantages of stability and accuracy; there is a mounting hole on the top of the sound collector, which is used to fix the mounting ring, energy converter and upper pressing plate; the mounting ring and the upper pressing plate are made of polyformaldehyde thermoplastic crystalline polymer, the material has a hard texture and good insulation effect, is conducive to wireless signal transmission, and is easy to process and manufacture; the mounting ring is a circular concave structure as a whole, and the concave part is used to install the energy converter, and the diameter of the energy converter is equal to the diameter of the concave part of the mounting ring; the upper pressing plate is a circular convex structure as a whole, and the diameter of the convex part is equal to the diameter of the concave part of the mounting ring; the energy converter is converted into The converter is placed in the concave structure of the mounting ring and is fixed by embedding and squeezing the convex part of the upper pressing plate; the connecting seam between the mounting ring and the upper pressing plate is sealed and fixed as a whole in the mounting hole at the top of the cone-shaped sound collector; there are corresponding groove structures in the mounting ring and the upper pressing plate for leading out cables, one end of the two cables is respectively connected to the upper and lower surfaces of the energy converter, and the other end is respectively connected to the positive and negative inputs of the amplifier in the hardware processing circuit; the isolator is made of polyformaldehyde thermoplastic crystalline polymer, with a hollow cylindrical structure at the bottom and a semi-hollow spherical structure at the top, which is connected to the cone-shaped sound collector through threads; the hollow cylindrical structure is used to store the hardware processing circuit, and a vent is provided on the top of the semi-hollow spherical structure to ensure the fluidity of the surrounding air when the energy converter vibrates, and the isolator is connected to the sound collector through threads.

3. According to claim 1, a low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection is characterized in that: The evaluation of the installation position of the high-sensitivity sensor in S3 is generally divided into signal measurement, sound field calculation, and measurement point analysis, wherein the signal measurement part mainly provides a sound field reconstruction data basis for the sound field calculation, and the result of the sound field calculation and the collected sound source data will be used together to calculate the evaluation index of the monitoring point; The signal measurement is used to measure the distance space parameters between the circuit breaker surface and the switch cabinet; the circuit breaker surface is divided into a number of equivalent source areas, and the equivalent sound source signal is collected by a microphone; the circuit breaker equivalent sound source signal is phase corrected; the equivalent sound source signal is fast Fourier transformed to calculate the frequency domain information of the equivalent sound source; The sound field calculation simplifies and constructs a three-dimensional model of the calculation domain based on measured information, and then divides the grid; defines the parameters of the calculation domain and defines the environmental boundary conditions based on the simplified model; uses the discontinuous Galerkin finite element method to calculate the transient change process of the sound field; extracts the sound pressure time domain signal of the candidate monitoring point, and performs fast Fourier transform to obtain its corresponding frequency domain information; The measuring point analysis calculates the Pearson correlation matrix of the frequency spectrum between all candidate measuring points and the equivalent sound source; calculates the Pearson equilibrium correlation coefficient according to the calculation result of the Pearson correlation matrix; and selects the candidate monitoring point with the highest Pearson equilibrium correlation coefficient as the optimal monitoring point.

4. According to claim 3, a low-power voiceprint sensor design method suitable for high-voltage switch cabinet circuit breaker mechanical fault detection is characterized in that: When the signal measurement is performed to collect the equivalent sound source, it is necessary to simultaneously obtain the circuit breaker surface sound signals of all measuring points, and use the phase correction method used in the operation deformation vibration mode measurement to correct the equivalent sound source signal; fix one measuring point as the reference signal, record the fixed phase angle relationship between the signals of other measuring points and the reference signal, and after all the collection work is completed, all measuring points are based on the anchor phase of the reference measuring point for phase correction; Vibration and sound are essentially mechanical waves. Mechanical waves appear as vibration waves in the circuit breaker medium. The process of mechanical wave transmission to the air is relatively complicated, and additional vibrations will be excited when transmitted to the circuit breaker. As the final stage of mechanical wave transmission, in order to avoid errors in the sound field calculation caused by complex mechanical parameters and internal propagation processes, the near-acoustic field analysis of the circuit breaker is regarded as a transient calculation problem of the sound field with the circuit breaker surface as the vibration source and the air medium as the calculation domain.

5. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 4 is characterized in that: Simplification of the computational domain: Simplification of acoustic field boundary: There is a boundary around the circuit breaker, which is simplified according to the hard acoustic field boundary and the hard acoustic field interface with sound-absorbing pores; Principle of transient sound field calculation based on DGFEM: DGFEM has obvious advantages in transient sound field analysis within a large spatial range. The spatial discretization form of the sound pressure conduction equation is: Mp+Cp+Kp=Q(t), Where: M is the fluid mass matrix; C is the fluid damping matrix; K is the fluid stiffness matrix; p is the fluid node pressure matrix; Q(t) is the sound source vector at each node; the time domain discretization of the sound wave propagation equation is: 0<… <t n <… <t N ; For the discontinuity time t n , its step function is expressed as: Where: For any time step In, the sound pressure distribution function p is calculated using the following third-order Hermit interpolation function: Where p represents the node sound pressure at time t, and the expressions of A(t) and B(t) are: The derivative v of the sound pressure distribution function is linearly interpolated within the time step: The sound source Q also uses linear interpolation: Based on the equilibrium equations and boundary conditions, the weak form of the discontinuous Galerkin finite element method for the sound wave propagation problem is obtained as follows: It is further transformed into a decoupled formula, which is the basic solution formula of the time domain discontinuous Galerkin finite element method: in, 6. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 1, characterized in that: The audio signal detection in S6 includes energy value calculation and the relationship between the energy value and a preset threshold value; Calculate the energy value E, The audio energy value of the extracted signal to be detected is calculated; when the energy value E of the collected audio signal is calculated, it is as shown in the following formula: Among them, N represents the total number of frames of the time domain audio data of the collected audio signal; k represents the frame sequence number of one frame in the time domain data of the collected signal; M represents the number of time domain sampling point data of each frame in the time domain data information of the collected signal; the mathematical expression x(i) represents the obtained value of the i-th time domain sampling point data of each frame in the time domain data of the collected signal.

7. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 1, characterized in that: The relationship between the energy value and the preset threshold is: If the calculated energy value E is less than the preset energy threshold Eth, the device connection is in an abnormal state, and an alarm signal is generated to alarm; If the calculated E≥Eth, first convert the N frames of time domain data information of the collected audio signal into frequency domain data information by corresponding methods for detection; the spectrum peak frequency of each frame in the N frames of frequency domain data information must be extracted, that is, the amplitude of each frame of the frequency domain information data after the conversion contains P frequency domain sampling points. When the frequency of the frequency domain sampling point is 16kHz, P is 1024; at the same time, record the frequency value with the largest amplitude among the P sampling points, and record this point as the spectrum peak frequency of the N frames of frequency domain data information; determine whether the spectrum peak frequency of the frequency domain audio data information remains stable based on the calculation results, and the calculation of the first stability value Dif of the spectrum peak frequency of the frequency domain data information is as follows: The second stability value St of the spectrum peak frequency is as follows: Wherein, N represents the total number of frames of the collected audio signal; k represents the frame sequence number of a certain frame of data in the time domain audio of the collected audio signal; index(k) represents the kth frame of frequency domain audio data in the data information of the collected audio signal; index(k+1) represents the spectrum peak frequency of the k+1th frame of data information in the data information of the collected audio signal; Compare the calculated Dif with the first stability preset threshold, and the calculated St with the second stability preset threshold in turn; if the calculated Dif value is less than the first stability threshold Difth, and the calculated St value is less than the second stability threshold Stth, then the spectrum peak frequency of the collected audio signal is definitely in a stable state, which means that the device connection is abnormal, and an alarm signal is generated at the same time.

8. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 1, characterized in that: The data collection and preprocessing steps in S6 are as follows: The preprocessing process mainly includes four steps: pre-emphasis, framing, windowing and speed matching; among them, pre-emphasis processing is mainly to compensate for the loss of high-frequency components; framing is to facilitate local batch processing of audio data; when framing, each frame signal must contain at least one rotation cycle of the component, and the frame shift cannot exceed half of the frame length; due to the non-periodic truncation of the signal, the spectrum will cause serious tailing in the entire frequency band, resulting in very serious errors, called leakage. In order to make each frame signal after framing still meet the continuity and reduce the impact of truncation effect, each frame data needs to be multiplied by a window function, that is, windowing processing; and the Hamming window is used.

9. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 1, characterized in that: The multi-scale feature extraction in S6 includes extracting all-round and multi-scale features from time domain features, frequency domain features, and cepstrum domain features, performing comprehensive analysis, and then reducing the dimension of the extracted features to facilitate screening out effective features; The calculation formula of time domain characteristics is as follows: Where: y i (n) The signal amplitude of the nth point of the audio in the i-th frame after framing; L represents the frame length; n represents the point number of each frame; sgn represents the sign function; Among them, the sgn symbol function is defined as: Frequency domain features are used to convert time domain signals to frequency domain. They are defined as the signal power within a unit frequency band and intuitively show the functional relationship between signal variance and frequency. The cepstral domain feature is the Mel cepstral coefficient feature, which is a perceptual frequency domain cepstral coefficient. Based on the human auditory mechanism, a group of H triangular bandpass filter groups are arranged from low frequency to high frequency in the Mel scale frequency domain according to the size of the critical bandwidth. The number of filter groups and the number of critical bands must be close, generally 22-26. After preprocessing the input signal, Fourier transform is performed, and then it passes through the filter group, and the logarithmic energy of each filter group output is calculated, and then the MFCC coefficients can be obtained by discrete cosine transform; MFCC feature calculation formula: Where: S represents the Mel filter energy, kg / s 2 ; The subscript i represents the frame number; h represents the Mel filter number; n represents the spectrum line after discrete cosine transform; The invention is used for automatic gain control of voiceprint signal, and mainly consists of two parts: a gain amplifier and a feedback loop. The gain amplifier component uses a certain AGC algorithm to adjust the gain value according to the amplitude and threshold of the feedback loop in the system. That is, the AGC adjusts the automatic control of the signal by the strength of the automatic gain signal in the amplifying circuit. When the signal strength increases, the control of the feedback loop of the AGC system is reduced accordingly according to a certain relationship. On the contrary, when the signal amplitude in the gain amplifier of the AGC system decreases, the gain of the feedback loop will be increased accordingly according to a certain relationship between the two. Through this method, after the amplification and adjustment of the AGC algorithm, it is ensured that the amplitude of the signal output of the communication system can be basically maintained in a constant state.

10. The method for designing a low-power voiceprint sensor suitable for mechanical fault detection of a high-voltage switch cabinet circuit breaker according to claim 9, characterized in that: The steps of the automatic gain control are as follows: S601: Determine the dynamic range of the audio signal output: The design is based on the audio signal processing using the AGC algorithm, and needs to be determined according to the actual situation of the audio signal; If the expected value in the practical AGC system is a fixed value during the design process, the amplitude of the audio signal output will approach the expected value of the AGC system; in the process of approaching the expected value, the output signal will be constantly adjusted, which will cause instability in the amplitude of the signal output. In order to ensure the stability of the audio signal output, a dynamic range of stable audio signal output is designed and determined based on the AGC expected value and centered on the expected value; when the amplitude of the audio signal is within this dynamic range during the AGC algorithm adjustment, it is determined that the output amplitude of the audio signal is stable; S602: Determine the gain adjustment speed: In order to determine, ensure and present the normal change trend of the voice signal amplitude, it is necessary to ensure that its gain remains unchanged or changes within a smaller range, thereby ensuring that the amplitude of the audio signal is not distorted during transmission; the interference received during the audio signal transmission process makes the average amplitude of the audio signal relatively small, so it is necessary to increase its gain value to ensure that the overall amplitude of the signal increases, thereby ensuring that the audio signal is not distorted; the application of practical AGC in audio signals, its gain value needs to be slow and fast, and change suddenly and quickly with the slow change of the signal amplitude; when the gain adjustment time of the audio signal is longer than the adjustment time of the audio signal under normal circumstances, the gain value at this time will not change significantly, and the gain adjustment time needs to be set to 4s according to the characteristics of the audio signal output; if the audio signal output is large, the gain at this time needs to be reduced quickly, otherwise it will cause damage to the device; and when the output value of the signal is greater than the upper limit of the expected value, the gain value adjustment time needs to be set to 0.5ms; The gain value calculation formula of the practical AGC algorithm is: Among them, A(n+1) is the gain value after signal transmission adjustment; A(n) is the gain value before adjustment; A in is the signal amplitude value; A0 is the expected amplitude value; 2 -a is a constant; different a values ​​will result in different AGC adjustment speeds. According to the actual requirements of the practical AGC algorithm, when the signal output amplitude is less than the upper limit of the expected value, a=16, the gain adjustment will be slower. If a=3, the gain adjustment speed will be faster. S603: Determine the dynamic range of the input signal: During the audio signal input process, if the input value range is set too large, the signal will generate greater noise during transmission, affecting the input and output quality of the signal; on the contrary, if the set range of the audio input signal is smaller, the signal will be ignored due to its small size, thereby causing distortion of the audio input signal; according to the input time of the audio signal, a signal strength of 36dB is determined as noise; using the AGC algorithm, the input dynamic range of the audio signal is determined, and the noise can be judged whether it occurs in the gap of the audio signal transmission according to the existence time of the noise. If the existence time of the noise is relatively short, it can be regarded as the gap of the audio signal; usually, the judgment time of the noise is set to 5s for the best; according to the existence time of the noise and the actual situation of the signal input, the range of about 6dB between the minimum value of the audio signal and the noise threshold is determined as the dynamic range of the audio signal input; within this range, the interference intensity of the noise on the audio signal is relatively high, so the gain value needs to be fixed; S604: Limit the gain value: To avoid excessive gain and burnout of signal transmission components or equipment, the gain value needs to be controlled within a range. If the value calculated by the AGC is greater than the upper limit of the range, the gain value takes the maximum value, otherwise it takes the minimum value. According to the actual transmission of the audio signal, the gain range is -3-30dB.

Images