Mechanical equipment electric spark monitoring method based on voiceprint technology

Abstract

A mechanical equipment electric spark monitoring method based on the voiceprint technology comprises the following steps that S1, for each area, terminal equipment collects voiceprint signals, vibration signals and temperature in a preset time period through a sound pickup, a vibration sensor and an infrared temperature measurement camera; s2, the terminal equipment determines wave beams and temperatures in a preset time period according to the voiceprint signals, the vibration signals and the temperatures corresponding to all the areas; s3, the terminal equipment determines an oscillogram in a preset time period according to the wave beam and the temperature; s4, the terminal device determines a sound pressure level, a peak value, a waveform factor and a vibration frequency in a preset time period according to the oscillogram; and S5, when the terminal equipment finds that the voiceprint and the vibration are abnormal, an infrared temperature measurement camera is turned on to carefully observe a specific part, and temperature analysis is carried out on the area. The invention provides a mechanical equipment electric spark monitoring method based on the voiceprint technology. Real-time online detection of electric sparks of mechanical equipment is achieved through the voiceprint recognition technology.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation equipment monitoring, and in particular to a method for monitoring electric sparks of mechanical equipment based on voiceprint technology. Background Art

[0002] The collector ring and carbon brushes are arranged in the slip ring chamber on the upper part of the rotor. The excitation current flows from the stationary excitation device into the brush holder through the lead wire, and then is transmitted to the slip ring on the collector ring (corresponding to the upper and lower slip rings respectively) through multiple contact brushes. The current continues to pass through the lead wire to the rotor field winding (pole coil).

[0003] Common faults in the slip ring room include the swing of the main shaft, which causes periodic friction sounds. The vibration swing measurement system can be used to assist in judging whether there are any abnormal vibrations. Friction and spark sounds, poor contact of the carbon brushes, causing friction noises in many places, obvious sparking, and noisy sounds in the slip ring room. It is difficult to judge the spark sounds by human ears, and the on-site inspection can only be assisted by human inspections. The main shaft naturally replenishes air, and even the sound of water return. The VGS and ALSTOM units, one through negative pressure and the other through output, naturally replenish the main shaft. The sound of air replenishment can be collected, and it can also help us understand the pressure pulsation in the volute. If water returns through the main shaft check valve, it can also be discovered in time to reduce the harm to electrical equipment. There is no auxiliary judgment method. The pressure pulsation can be combined with the audio at the volute to judge. Once the above faults occur, it will affect the safety and stability of the hydro-turbine generator set during operation.

[0004] Currently, the power generation industry does not have a method for detecting sparks in mechanical equipment using voiceprint technology. The method for detecting sparks in most generator sets still relies solely on operators to conduct regular on-site inspections and make judgments using their eyes and ears. Operators can carefully observe and analyze the phenomena at the locations where abnormal noises occur. Operators can also use infrared temperature measuring instruments to measure the temperature of carbon brushes and slip rings during the inspection to determine if any abnormalities occur. If the temperature exceeds a set value, the unit is operating abnormally.

[0005] The existing technology cannot realize real-time online monitoring of electric sparks in the slip ring chamber. It only relies on the operating personnel to observe with the naked eyes and ears and measure with instruments during regular on-site inspections. It is also troublesome to carry infrared temperature measuring instruments during the inspection. Therefore, if electric sparks are generated before or after the inspection, they cannot be discovered in time, and it is impossible to quickly know whether the operating status of the unit is normal. Once more serious electric sparks are generated during operation, it will directly affect the normal operation of the unit and may cause more serious consequences such as wire damage. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method for monitoring electric sparks in mechanical equipment based on voiceprint technology, which realizes real-time online detection of electric sparks in mechanical equipment through voiceprint recognition technology, and fills the gap in the electrical industry in real-time online detection of electric sparks in mechanical equipment.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for monitoring electric sparks of mechanical equipment based on voiceprint technology, comprising the following steps: S1. For each area, the terminal device collects voiceprint signals, vibration signals, and temperature within a preset time period through microphones, vibration sensors, and infrared temperature measurement cameras; S2. The terminal device determines the beam and temperature within a preset time period according to the voiceprint signal, vibration signal, and temperature corresponding to each area; S3, the terminal device determines the waveform diagram within a preset time period according to the beam and temperature; S4. The terminal device determines the sound pressure level, peak value, waveform factor, and vibration frequency within a preset time period according to the waveform diagram; S5. When the terminal device finds abnormal voiceprint or vibration, it turns on the infrared temperature measurement camera to carefully observe the specific parts and perform temperature analysis on the area; S6. The terminal device monitors the electric sparks of the electrical equipment within a preset time period based on the sound pressure level, vibration frequency, and temperature image.

[0008] Preferably, in step S1, four infrared temperature measuring cameras are provided and installed on the diagonal side of the main shaft of the slip ring chamber; the microphone adopts a bone conduction microphone, and multiple microphones are provided and installed on the carbon brush handle of the slip ring chamber to better collect sound; two vibration sensors are provided and installed on the bottom of the slip ring chamber.

[0009] Preferably, the bone conduction microphone, infrared temperature measurement camera and vibration sensor in step S1 transmit the detected voiceprint detection data, infrared temperature measurement data, image recognition data and vibration detection data to the edge computing gateway, which is connected to the server through a switch, and the server is respectively connected to the data platform and the supercomputer, and the supercomputer is also connected to the data platform. The edge gateway has a preliminary data screening function to transmit the voiceprint detection data, infrared temperature measurement data and image data separately for subsequent analysis.

[0010] Preferably, a voiceprint model is established during voiceprint signal recognition. First, the collected voiceprint is optimized by using the CMFMC3.0 voiceprint algorithm. CMFMC3.0 performs weighted dimensionality reduction optimization based on MFCC feature vectors and applies a vector quantization algorithm to recognize sound signals. Specifically, the following steps are included: (1) MFCC extraction, including: 1) Obtaining original sound data as continuous sound samples; 2) Pre-emphasis processing; 3) Framing; 4) Add window; 5) FFT transformation; 6) Triangular bandpass filtering; 7) Calculate the logarithmic energy of each filter bank output; 8) DCT transform; 9) Calculate logarithmic energy; 10) Extraction of dynamic differential parameters; (2) Calculation of UBM The Universal Background Model (UBM) first collects a large amount of speech irrelevant to the target, trains a UBM, and then uses a small amount of target speech data to adjust the parameters of the UBM through an adaptive algorithm to obtain the target model parameters.

[0011] (3) Linear Probability Discriminant Analysis (PLDA) Assume that the training data speech consists of the speech of I speakers, each of whom has J different speech segments, and define the jth speech of the i-th speaker as X ij ; According to factor analysis, define X ij The generative model is: ; Among them, μ represents the mean of all training data; F is the identity space, which contains information that can be used to represent various speakers; hi is the identity of a specific speaker (or the position of the speaker in the identity space); G is the error space, which contains information that can be used to represent different voice changes of the same speaker; w ij represents the position in G space; ϵ ij is the final residual noise term, used to represent something that has not yet been explained. This term is zero-mean Gaussian with variance Σ. The goal of PLDA model training is to input a bunch of data X ij , the output can best represent the parameters θ = [μ, F, G, Σ] of the data set; in PLDA, we calculate whether two speech sounds are generated by the feature hi in the speaker space, or the likelihood of being generated by hi, regardless of the difference in the intra-class space. The scoring formula is given below: ; η1 and η2 are the I-Vector vectors of the two voices. The assumption that the two voices come from the same space is Hs, and the assumption that they come from different spaces is Hd. Among them, p(η1,η2|hs) is the likelihood function of the two voices coming from the same space; p(η1|hd), p(η2|hd) are the likelihood functions of η1 and η2 coming from different spaces respectively; by calculating the logarithm, the similarity of the two voices can be measured; The higher the ratio, the higher the score, and the greater the possibility that the two voices belong to the same speaker; The lower the ratio, the lower the score, and the less likely it is that the two voices belong to the same speaker.

[0012] Preferably, in step 5, the abnormality determination is specifically as follows: When an abnormality is found in the voiceprint score of the real-time analysis device, it is compared with the known fault type model in the database. If it is a known fault, it is directly output as a known fault. If it does not match the known fault model library, an unknown abnormality alarm is output.

[0013] The present invention provides a method for monitoring sparks in mechanical equipment based on voiceprint technology, which has the following beneficial effects: 1. Through the installation of various types of acquisition sensors, multi-modal online real-time monitoring of electric sparks of machine equipment can be realized to fill the gap in the industry; 2. By collecting audio data through bone conduction microphones, temperature image data through infrared thermometers, and vibration information through vibration detection devices, it is possible to identify all locations and causes of electric sparks in mechanical equipment, greatly improving the efficiency of problem discovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a flow chart of the method of the present invention; Figure 2 This is the installation position diagram of the infrared temperature measuring camera of the present invention; Figure 3 This is a data transmission flow chart of the present invention; Figure 4 A schematic diagram of training UBM of the present invention; Figure 5 This is a fault judgment flow chart of the present invention. DETAILED DESCRIPTION

[0015] like Figure 1 As shown, a method for monitoring sparks in mechanical equipment based on voiceprint technology includes the following steps: S1. For each area, the terminal device collects voiceprint signals, vibration signals, and temperature within a preset time period through microphones, vibration sensors, and infrared temperature measurement cameras; S2. The terminal device determines the beam and temperature within a preset time period according to the voiceprint signal, vibration signal, and temperature corresponding to each area; S3, the terminal device determines the waveform diagram within a preset time period according to the beam and temperature; S4. The terminal device determines the sound pressure level, peak value, waveform factor, and vibration frequency within a preset time period according to the waveform diagram; S5. When the terminal device finds abnormal voiceprint or vibration, it turns on the infrared temperature measurement camera to carefully observe the specific parts and perform temperature analysis on the area; S6. The terminal device monitors the electric sparks of the electrical equipment within a preset time period based on the sound pressure level, vibration frequency, and temperature image.

[0016] Preferably, Figure 2 As shown, in the step S1, four infrared temperature measuring cameras are provided and installed on the diagonal side of the main shaft of the slip ring chamber; the pickup adopts a bone conduction pickup, and multiple of them are provided and installed on the carbon brush handle of the slip ring chamber to better collect sound; two vibration sensors are provided and installed on the bottom of the slip ring chamber.

[0017] Preferably, Figure 3 As shown, the bone conduction microphone, infrared temperature measurement camera and vibration sensor in step S1 transmit the detected voiceprint detection data, infrared temperature measurement data, image recognition data and vibration detection data to the edge computing gateway, which is connected to the server through a switch, and the server is respectively connected to the data platform and the supercomputer, and the supercomputer is also connected to the data platform. The edge gateway has the function of preliminary data screening to transmit the voiceprint detection data, infrared temperature measurement data and image data separately for subsequent analysis.

[0018] Preferably, a voiceprint model is established during voiceprint signal recognition. First, the collected voiceprint is optimized by using the CMFMC3.0 voiceprint algorithm. CMFMC3.0 performs weighted dimensionality reduction optimization based on MFCC feature vectors and applies a vector quantization algorithm to recognize sound signals. Specifically, the following steps are included: (1) MFCC extraction, including: 1) Obtaining original sound data as continuous sound samples; 2) Pre-emphasis processing: Pre-emphasis can compensate for the loss of high-frequency part to a certain extent and protect the integrity of the sound channel information.

[0019] 3) Framing: After framing, processing each frame of the signal is equivalent to processing a continuous signal with fixed features, which can reduce the impact of non-steady-state time variations.

[0020] 4) Windowing: After dividing the signal into frames, we apply a window function (such as a Hamming window) to each frame. There are several reasons to apply a window function to the window frame, especially to counteract the assumption of the FFT that the data is infinite and to reduce spectral leakage.

[0021] 5) FFT transformation: FFT is the fast Fourier transform, which transforms the waveform data into the frequency domain through Fourier transform. Then the Mel coefficient filter is performed. The Mel scale is the cepstrum parameter extracted from the frequency domain, which describes the nonlinear characteristics of the human ear frequency. Its relationship with the frequency can be approximately expressed by the following formula: Mel(f)=2585×log(1+f / 700); 6) Triangular bandpass filtering: The triangular bandpass filter smoothes the spectrum and eliminates the effect of harmonics, highlighting the resonance peaks of the original speech. The spectrum has an envelope and a fine structure, which correspond to timbre and pitch respectively. For speech recognition, timbre is the main useful information, and pitch is generally useless. By integrating within each triangle, the fine structure can be eliminated, and only the timbre information is retained. Therefore, the pitch or pitch of a speech will not be presented in the MFCC parameters. In other words, the speech recognition system characterized by MFCC will not be affected by the different pitches of the input speech.

[0022] 7) Calculate the logarithmic energy of each filter bank output; since human perception is proportional to the logarithm of the frequency, log simulation is used.

[0023] 8) DCT transformation; a DCT can separate the above logarithmic energy into discrete cosine transform and find the L-order Mel-scale Cepstrum parameters. L-order refers to the order of MFCC coefficients, usually 12-16. Here M is the number of triangular filters. This may be a problem in some machine learning algorithms. Therefore, we can apply discrete cosine transform (DCT) to decorrelate the filter bank coefficients and produce a compressed representation of the filter bank. Typically, for automatic speech recognition (ASR), the resulting cepstrum coefficients 2-13 are retained and the rest are discarded.

[0024] 9) Calculate logarithmic energy; logarithmic energy is the volume (i.e. energy) of a frame, which is also an important feature of speech and is very easy to calculate. Therefore, the logarithmic energy of a frame is usually added (definition: the sum of the squares of the signals in a frame, then the logarithmic value with base 10, and then multiplied by 10) so that the basic speech features of each frame have one more dimension, including a logarithmic energy and the remaining cepstrum parameters.

[0025] 10) Extraction of dynamic differential parameters: The standard MFCC cepstrum parameters only reflect the static characteristics of speech parameters. The dynamic characteristics of speech can be described by the differential spectrum of these static features. Experiments have shown that combining dynamic and static features can effectively improve the recognition performance of the system. The calculation of differential parameters can be done using the following formula: ; From the above steps, we can see that MFCC is actually composed of: N-dimensional MFCC parameters (N / 3 MFCC coefficients + N / 3 first-order difference parameters + N / 3 second-order difference parameters) + frame energy (this item can be replaced according to needs).

[0026] (2) Calculation of UBM The Universal Background Model (UBM) first collects a large amount of speech irrelevant to the target, trains a UBM, and then uses a small amount of target speech data to adjust the parameters of the UBM through an adaptive algorithm to obtain the target model parameters. In general, the only parameter adjusted by the adaptive algorithm is the mean. This approach can reduce the amount of data and parameters in actual use, and facilitates rapid training convergence and decoding calculations on mobile terminals.

[0027] The training of UBM is actually the training of Gaussian mixture model (GMM). Suppose we use a large amount of background speech data to train a UBM. The next focus is how to adapt the parameters of UBM to obtain the GMM parameters of the target speaker. The schematic diagram is as follows Figure 4 shown.

[0028] (3) Linear Probability Discriminant Analysis (PLDA) Probabilistic Linear Discriminant Analysis (PLDA) is also a channel compensation algorithm, also known as the probabilistic form of LDA algorithm. In the field of voiceprint recognition, it is assumed that the training data speech consists of the speech of I speakers, where each speaker has J different segments of his own speech, and the jth speech of the i-th person is defined as X. ij ; According to factor analysis, define X ij The generative model is: ; Among them, μ represents the mean of all training data; F is the identity space, which contains information that can be used to represent various speakers; hi is the identity of a specific speaker (or the position of the speaker in the identity space); G is the error space, which contains information that can be used to represent different voice changes of the same speaker; w ij represents the position in G space; ϵ ijis the final residual noise term, used to represent something that has not yet been explained. This term is zero-mean Gaussian with variance Σ. The goal of PLDA model training is to input a bunch of data X ij , the output can best represent the parameters θ = [μ, F, G, Σ] of the data set; in PLDA, we calculate whether two speech sounds are generated by the feature hi in the speaker space, or the likelihood of being generated by hi, regardless of the difference in the intra-class space. The scoring formula is given below: ; η1 and η2 are the I-Vector vectors of the two voices. The assumption that the two voices come from the same space is Hs, and the assumption that they come from different spaces is Hd. Among them, p(η1, η2 | hs) is the likelihood function of the two voices coming from the same space; p(η1|hd) and p(η2|hd) are the likelihood functions of η1 and η2 coming from different spaces respectively. By calculating the logarithm, the similarity of the two voices can be measured; The higher the ratio, the higher the score, and the greater the possibility that the two voices belong to the same speaker; The lower the ratio, the lower the score, and the less likely it is that the two voices belong to the same speaker.

[0029] Preferably, in step 5, the abnormality determination is specifically as follows: After obtaining the soundprint model of the spark of mechanical equipment, the parameters transmitted by the bone conduction pickup, thermal infrared thermometer and vibration sensor installed in the slip ring room are used to determine whether sparks are generated, output the results and take corresponding measures.

[0030] When an abnormality is found in the voiceprint score of the real-time analysis device, it is compared with the known fault type model in the database. If it is a known fault, it is directly output as a known fault. If it does not match the known fault model library, an unknown abnormality alarm is output.

[0031] Once an unknown abnormal alarm is reported, the surveillance camera will carefully observe and analyze the alarm point, and at the same time arrange for the operation personnel to go to the site to confirm the situation and make corresponding treatment. After the treatment is completed, they will come back to report and then improve the fault model database. Next time, the same fault will be reported as a known fault, such as Figure 5 shown.

[0032] Generally, infrared temperature cameras are activated on a scheduled basis but not for real-time monitoring, because the image data they collect is too large and places high demands on the database. We set it up so that when the voiceprint or vibration signal is abnormal, the infrared temperature camera is activated to conduct a detailed analysis of the area, thus solving the problem of real-time monitoring of the electric spark situation of power equipment.

[0033] The present invention integrates the data of multiple modules into the same system by installing bone conduction pickups, cameras with thermal infrared temperature measurement functions, vibration sensors, etc. at appropriate locations, and realizes real-time online monitoring of the sparks of mechanical equipment through system algorithms. By adjusting the algorithm and giving it a large amount of data parameters to allow it to learn and update in depth, it realizes comprehensive detection and judgment of the sparks of mechanical equipment, greatly improving the speed of discovering the sparks of mechanical equipment.

[0034] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for monitoring sparks in mechanical equipment based on voiceprint technology, characterized in that: The following steps are involved: S1. For each area, the terminal device collects voiceprint signals, vibration signals, and temperature within a preset time period through microphones, vibration sensors, and infrared temperature measurement cameras; S2. The terminal device determines the beam and temperature within a preset time period according to the voiceprint signal, vibration signal, and temperature corresponding to each area; S3, the terminal device determines the waveform diagram within a preset time period according to the beam and temperature; S4. The terminal device determines the sound pressure level, peak value, waveform factor, and vibration frequency within a preset time period according to the waveform diagram; S5. When the terminal device finds abnormal voiceprint or vibration, it turns on the infrared temperature measurement camera to carefully observe the specific parts and perform temperature analysis on the area; S6. The terminal device monitors the electric sparks of the electrical equipment within a preset time period based on the sound pressure level, vibration frequency, and temperature image.

2. According to claim 1, a method for monitoring sparks in mechanical equipment based on voiceprint technology is characterized in that: In step S1, four infrared temperature measurement cameras are provided and installed on the diagonal side of the main shaft of the slip ring chamber; the pickup adopts a bone conduction pickup, and there are multiple ones installed on the carbon brush handle of the slip ring chamber; there are two vibration sensors, which are installed on the bottom of the slip ring chamber.

3. According to claim 2, a method for monitoring sparks in mechanical equipment based on voiceprint technology is characterized in that: The bone conduction microphone, infrared temperature measurement camera and vibration sensor in step S1 transmit the detected voiceprint detection data, infrared temperature measurement data, image recognition data and vibration detection data to the edge computing gateway. The edge computing gateway is connected to the server through a switch, and the server is respectively connected to the data platform and the supercomputer. The supercomputer is also connected to the data platform.

4. According to claim 1, a method for monitoring sparks in mechanical equipment based on voiceprint technology is characterized in that: When recognizing voiceprint signals, a voiceprint model is established. First, the collected voiceprint is optimized through the CMFMC3.0 voiceprint algorithm. CMFMC3.0 performs weighted dimensionality reduction optimization based on MFCC feature vectors and applies a vector quantization algorithm to recognize sound signals. The specific steps include the following: (1) MFCC extraction, including: 1) Obtaining original sound data as continuous sound samples; 2) Pre-emphasis processing; 3) Framing; 4) Add window; 5) FFT transformation; 6) Triangular bandpass filtering; 7) Calculate the logarithmic energy of each filter bank output; 8) DCT transform; 9) Calculate logarithmic energy; 10) Extraction of dynamic differential parameters; (2) Calculation of UBM Universal background model, that is, first collect a large amount of speech irrelevant to the target, train a UBM, and then use a small amount of target speech data to adjust the parameters of the UBM through an adaptive algorithm to obtain the target model parameters; (3) Linear Probability Discriminant Analysis (PLDA) Assume that the training data speech consists of the speech of I speakers, each of whom has J different speech segments, and define the jth speech of the i-th speaker as X ij ; According to factor analysis, define X ij The generative model is: ； Among them, μ represents the mean of all training data; F is the identity space, which contains information that can be used to represent various speakers; hi is the identity of a specific speaker; G is the error space, which contains information that can be used to represent different voice changes of the same speaker; w ij represents the position in G space; ϵ ij is the final residual noise term, used to represent what has not been explained; the scoring formula is given below: ； η1 and η2 are the I-Vector vectors of the two voices. The assumption that the two voices come from the same space is Hs, and the assumption that they come from different spaces is Hd. Among them, p(η1,η2|hs) is the likelihood function of the two voices coming from the same space; p(η1|hd), p(η2|hd) are the likelihood functions of η1 and η2 coming from different spaces respectively; by calculating the logarithm, the similarity of the two voices can be measured; The higher the ratio, the higher the score, and the greater the possibility that the two voices belong to the same speaker; The lower the ratio, the lower the score, and the less likely it is that the two voices belong to the same speaker.

5. According to claim 1, a method for monitoring sparks in mechanical equipment based on voiceprint technology is characterized in that: In step 5, the abnormality judgment is specifically as follows: When an abnormality is found in the voiceprint score of the real-time analysis device, it is compared with the known fault type model in the database. If it is a known fault, it is directly output as a known fault. If it does not match the known fault model library, an unknown abnormality alarm is output.

Images