A small volume flow safety method and system for a steam turbine based on voiceprint auscultation

Abstract

The application discloses a steam turbine small volume flow safety method and system based on voiceprint auscultation, and belongs to the technical field of steam turbine operation safety monitoring. The method comprises the following steps: collecting a voiceprint signal of a steam turbine under a small volume flow working condition; performing noise reduction and filtering pretreatment on the voiceprint signal; extracting frequency characteristics, amplitude characteristics and spectrum entropy characteristics to form a voiceprint feature vector; comparing and analyzing the feature vector with a pre-stored safety working condition benchmark model to obtain a safety evaluation result; and issuing a warning signal according to the evaluation result. The application further discloses a system for implementing the above method. The application takes the voiceprint signal as a monitoring medium, utilizes the characteristics that sound waves have a high propagation speed and are sensitive to internal state changes, solves the problems that the prior art has a lagging response and is not sensitive to early weak faults, and realizes real-time identification and graded early warning of safety hazards of the steam turbine under the small volume flow working condition.

Description

Technical Field

[0001] This application belongs to the field of steam turbine operation safety monitoring technology, specifically relating to a steam turbine small volume flow safety method and system based on voiceprint auscultation. Background Technology

[0002] When a steam turbine operates under low-volume flow conditions, the inlet steam flow rate is far lower than the design value. Steam cannot evenly fill the flow path, causing severe instability and disturbances in the flow field. These abnormal operating conditions can easily lead to backflow eddies at the blade tips, intensified vibration at the blade roots due to excitation forces, and uneven thermal expansion of moving and stationary components causing gap changes and rubbing. Furthermore, the increased wet steam content increases the risk of water erosion. If these hidden dangers are not identified and addressed in a timely manner, they can cause blade fatigue fracture, or even lead to complete unit damage, unplanned shutdowns, or even major safety accidents.

[0003] Currently, turbine operation safety monitoring mainly relies on three traditional methods: vibration, temperature, and pressure. Vibration monitoring uses eddy current or velocity sensors at the bearing housing to collect signals, reflecting rotor balance and bearing operating status; temperature monitoring uses thermocouples to detect the temperature of bearing bushes and exhaust steam, indirectly determining the equipment's thermal condition; pressure monitoring uses pressure transmitters to collect pressure at various levels and assess blockages and leaks in the flow path. However, these conventional monitoring methods have inherent shortcomings that are difficult to avoid.

[0004] Traditional monitoring methods suffer from three major drawbacks: First, they suffer from slow response times, with high thermal inertia in temperature and pressure limitations imposed by the working fluid, making them unsuitable for real-time early warning of rapidly evolving faults under low-volume flow conditions. Second, they have low sensitivity to early, subtle faults, with initial characteristic signals such as micro-cracks in blades and minor abrasions easily masked by background noise, and thermodynamic parameters only able to identify more obvious faults in the later stages. Third, sensor deployment is limited, requiring installation only on the exterior of the bearing housing, preventing direct monitoring of internal core components such as blades and shrouds, resulting in a significant loss of crucial internal operational information. Therefore, achieving rapid and sensitive identification of early safety hazards under low-volume flow conditions in steam turbines has become a pressing technical challenge for the industry. Summary of the Invention

[0005] This application addresses the technical problems in existing small-volume flow safety monitoring technologies for steam turbines, such as delayed response to temperature and pressure parameters, inability to meet real-time early warning requirements, insufficient sensitivity to early and minor faults such as initial blade cracks and slight rubbing, making it difficult to issue early warnings at the fault initiation stage, and the limitation of vibration sensor installation location, which prevents direct monitoring of changes in the state of internal core components such as blades and shrouds. The application provides a method and system for small-volume flow safety monitoring of steam turbines based on acoustic auscultation.

[0006] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides a safety method for small-volume flow rates in steam turbines based on voiceprint auscultation, comprising the following steps: Acoustic fingerprint signals of steam turbines under low volumetric flow conditions were collected. The acquired voiceprint signal is preprocessed, including noise reduction and filtering. Extracting voiceprint feature vectors from preprocessed voiceprint signals; The extracted acoustic signature feature vector is compared and analyzed with the pre-stored benchmark acoustic signature feature model of the steam turbine under safe operating conditions with small volume flow rate to obtain the safety assessment result. Based on the safety assessment results, corresponding early warning signals will be issued.

[0007] Furthermore, the specific steps for collecting acoustic signature signals of the steam turbine under low volumetric flow conditions include: Multiple acoustic sensors are evenly distributed on the surface of the turbine cylinder block, and the acoustic signature signal is acquired through a multi-sensor synchronous acquisition method. The noise reduction process employs a wavelet threshold noise reduction algorithm, which specifically includes: performing wavelet decomposition on the collected voiceprint signal, selecting a threshold function and threshold to process the wavelet coefficients, and then obtaining the noise-reduced signal through inverse wavelet transform. The filtering process employs an adaptive filter to eliminate interference signals at specific frequencies.

[0008] Furthermore, the specific method for extracting the frequency features is as follows: extracting frequency features from the preprocessed voiceprint signal through Fast Fourier Transform; The specific method for extracting the amplitude feature is as follows: extract the peak amplitude of the preprocessed voiceprint signal as the amplitude feature; The specific method for extracting the spectral entropy features is as follows: the spectral entropy features are calculated based on the power spectral density of the preprocessed acoustic signature signal.

[0009] Furthermore, the comparison analysis employs similarity calculation methods, including Euclidean distance, cosine similarity, or correlation coefficient. When the calculated similarity is lower than a set threshold, it is determined that there is a security risk.

[0010] Furthermore, the warning signal includes multi-level warnings, with different warning levels corresponding to the severity of safety hazards in the safety assessment results, and different warning levels corresponding to different warning methods and handling strategies; The multi-level early warning system consists of three levels. When the similarity is between 0.8 and 0.85, a level one early warning is triggered and an optical alarm is activated. When the similarity is between 0.7 and 0.8, a level 2 warning is triggered and an audible and visual alarm is activated. When the similarity is below 0.7, a level 3 warning is triggered and an alarm is triggered.

[0011] Secondly, this application provides a steam turbine small-volume flow safety system based on voiceprint auscultation, used to implement a steam turbine small-volume flow safety method based on voiceprint auscultation, including: The acoustic signature acquisition module is used to acquire acoustic signature signals of the steam turbine under small volume flow conditions; The signal preprocessing module is used to perform noise reduction and filtering on the acquired voiceprint signals; The feature extraction module is used to extract voiceprint feature vectors from the preprocessed voiceprint signal. The voiceprint feature vectors include frequency features, amplitude features, and spectral entropy features. A benchmark model library is used to store benchmark acoustic signature models of steam turbines under safe operating conditions with small volumetric flow rates. The security analysis module is used to compare and analyze the extracted voiceprint feature vector with the benchmark voiceprint feature model to obtain a security assessment result. The early warning module, connected to the security analysis module, is used to issue corresponding early warning signals based on the security assessment results.

[0012] Furthermore, the acoustic signature acquisition module includes multiple acoustic sensors, which are evenly distributed on the surface of the turbine cylinder block to acquire the acoustic signature signal through a multi-sensor synchronous acquisition method.

[0013] Furthermore, the signal preprocessing module employs a wavelet threshold denoising algorithm for noise reduction and an adaptive filter for filtering. The wavelet threshold denoising algorithm specifically includes wavelet decomposition of the voiceprint signal, selection of a threshold function and threshold to process the wavelet coefficients, and then obtaining the denoised signal through inverse wavelet transform. The adaptive filter is used to eliminate interference signals at specific frequencies.

[0014] Furthermore, the feature extraction module extracts the frequency features through fast Fourier transform, extracts the peak amplitude of the signal as the amplitude feature, and calculates the spectral entropy feature based on the power spectral density of the signal; the security analysis module performs comparison analysis using similarity calculation methods, including Euclidean distance, cosine similarity, or correlation coefficient.

[0015] Furthermore, the early warning module issues multi-level early warning signals based on the security assessment results, with different early warning levels corresponding to different early warning methods and processing strategies. The multi-level early warning is a three-level early warning: when the similarity is between 0.8 and 0.85, a level one early warning is triggered and a light alarm is used; when the similarity is between 0.7 and 0.8, a level two early warning is triggered and an audible and visual alarm is used; and when the similarity is below 0.7, a level three early warning is triggered and a prompt alarm is used.

[0016] Compared with the prior art, this application has the following beneficial effects: This application provides a safety method for steam turbines with small volumetric flow rates based on acoustic signature auscultation. Existing steam turbine safety monitoring relies on parameters such as vibration, temperature, and pressure. Temperature parameters have a thermal inertia lag of tens of seconds, which cannot meet the real-time early warning requirements for rapidly evolving faults under small volumetric flow conditions. Vibration signals have weak energy for early faults such as initial blade cracks and minor abrasion, and are easily drowned out by background noise. Vibration sensors can only be installed in limited locations such as bearing housings, and cannot directly monitor internal core components such as blades and shrouds. To address the aforementioned issues, this invention employs acoustic signature signals as the monitoring medium. The speed of sound propagation in metal is approximately 5000 m / s, enabling it to reflect changes in the internal state of the equipment with a millisecond-level response speed. The warning response time can be controlled within 10 seconds, overcoming the defect of response lag. The acoustic signature signals originate directly from the internal flow field of the turbine, blade vibration, and the interaction between moving and stationary components, making them highly sensitive to minute changes and capable of issuing warnings at the initial stage of a fault. The acoustic sensor is non-invasively installed on the outer surface of the cylinder body, requiring no modification to the internal structure of the turbine. It can directly collect acoustic signature signals reflecting changes in the state of internal components from the outside, overcoming the limitations of traditional sensor installation locations.

[0017] Specifically, multiple acoustic sensors are evenly distributed across the turbine cylinder surface. Synchronous acquisition by multiple sensors comprehensively captures acoustic signature information from various locations within the asymmetric and non-uniform flow field under small volumetric flow conditions, avoiding signal blind spots from single sensors. A wavelet threshold denoising algorithm is employed to remove noise while preserving abrupt signal changes. Combined with an adaptive filter, specific frequency noise such as power frequency interference is dynamically eliminated, significantly improving signal quality. Based on this, three complementary features—frequency features, amplitude features, and spectral entropy features—are extracted. Frequency features identify changes in specific frequency components such as blade resonance and flow field pulsation; amplitude features reflect the intensity of fault impacts such as the degree of rubbing and vibration amplitude; and spectral entropy characterizes the degree of system disorder caused by flow field instability. The combination of these three features constitutes a feature vector with strong fault discrimination capabilities.

[0018] Specifically, similarity calculation methods such as Euclidean distance, cosine similarity, or correlation coefficient are used for comparison and analysis. This transforms the complex feature comparison problem into a quantitative numerical comparison. When the similarity is below a set threshold, a safety hazard is identified. The judgment criteria are clear and unambiguous, avoiding the subjectivity of human experience-based judgment. Three levels of early warning are established based on the similarity range: a similarity of 0.8 to 0.85 triggers a Level 1 warning, using a visual alarm to alert operators; a similarity of 0.7 to 0.8 triggers a Level 2 warning, using an audible and visual alarm to require immediate intervention; and a similarity below 0.7 triggers a Level 3 warning, using a warning alarm to suggest immediate shutdown and inspection. This tiered strategy ensures operational safety under small-volume flow conditions while avoiding operational losses caused by overreaction to minor anomalies. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart of a small-volume flow safety method for steam turbines based on voiceprint auscultation, provided as an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of a small-volume flow safety system for steam turbines based on voiceprint auscultation, provided as an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] See Figure 1 This application provides a safety method for small-volume flow rates in steam turbines based on voiceprint auscultation, comprising the following steps: S1, collects acoustic fingerprint signals of the steam turbine under small volume flow conditions; In a more specific embodiment provided in this application, acoustic signature signals are acquired using multiple high-precision acoustic sensors evenly distributed across the surface of the turbine cylinder. As a specific distribution method, eight acoustic sensors are evenly arranged along the circumference of the turbine cylinder, with measurement points set at key locations in the high-pressure, intermediate-pressure, and low-pressure cylinders, and the sampling frequency is set to 50kHz. This multi-sensor synchronous acquisition method allows for the simultaneous acquisition of acoustic signature signals generated inside the turbine from different directions, avoiding the signal blind spots that may exist with a single sensor.

[0024] S2, preprocess the acquired voiceprint signal, the preprocessing including noise reduction and filtering; In a more specific embodiment provided in this application, the noise reduction process employs a wavelet thresholding noise reduction algorithm. First, the acquired original speaker signal is decomposed using wavelet decomposition. A wavelet basis function and the number of decomposition levels are selected to obtain wavelet coefficients at each scale. As a specific implementation, the db4 wavelet can be selected as the wavelet basis function, and the number of decomposition levels is set to 5. Then, a threshold function and a threshold are selected. In this embodiment, a soft threshold function is used, and the threshold is determined based on a heuristic thresholding criterion. Wavelet coefficients with absolute values ​​less than the threshold are set to zero, and wavelet coefficients with values ​​greater than the threshold are contracted. Finally, the signal is reconstructed through inverse wavelet transform to obtain the denoised speaker signal. The wavelet thresholding noise reduction algorithm has good time-frequency localization characteristics, effectively preserving the abrupt changes in the signal while removing noise.

[0025] In this embodiment, an adaptive filter is used for filtering. The adaptive filter can automatically adjust the filter coefficients according to the statistical characteristics of the input signal, track the changes in the frequency of the interference signal in real time, and effectively eliminate interference components of specific frequencies, such as rotor power frequency and its harmonic components, 50Hz power frequency interference and its harmonic components, electromagnetic interference, hydrodynamic noise, etc., while preserving the acoustic signature components that reflect the equipment status to the greatest extent.

[0026] S3, extract the voiceprint feature vector from the preprocessed voiceprint signal; In a more specific embodiment provided in this application, the extracted voiceprint feature vector includes frequency features, amplitude features, and spectral entropy features. These three features describe the voiceprint signal from different dimensions and constitute complementary feature vectors.

[0027] The frequency feature extraction method involves performing a Fast Fourier Transform (FFT) on the preprocessed acoustic signature signal to convert the time-domain signal into a frequency-domain signal, thus obtaining the energy distribution of the signal at different frequency components. As a specific implementation, a 1024-point or 2048-point FFT can be used, and the frequency resolution can be adjusted according to actual needs. The frequency features can reflect key information such as the meshing frequency of rotating components inside the turbine, the natural frequency of the blades, and the excitation frequency of the flow field. When a micro-crack appears on the blade, its natural frequency will shift; when the flow field pulsates, energy concentration at specific frequencies will occur.

[0028] The frequency feature extraction method involves performing a Fast Fourier Transform (FFT) on the preprocessed acoustic signature signal to convert the time-domain signal into a frequency-domain signal, thus obtaining the energy distribution of the signal at different frequency components. As a specific implementation, a 1024-point or 2048-point FFT can be used, and the frequency resolution can be adjusted according to actual needs. The frequency features can reflect key information such as the meshing frequency of rotating components inside the turbine, the natural frequency of the blades, and the excitation frequency of the flow field. When a micro-crack appears on the blade, its natural frequency will shift; when the flow field pulsates, energy concentration at specific frequencies will occur.

[0029] The extraction method for spectral entropy features is as follows: spectral entropy is calculated based on the power spectral density of the preprocessed acoustic signature signal. First, the power spectrum of the signal is estimated to obtain the power values ​​of each frequency component. The power spectrum is then normalized to a probability distribution, and the spectral entropy value is calculated using the Shannon entropy formula. Spectral entropy reflects the degree of disorder or uncertainty in the signal spectrum. When the flow field becomes unstable under small volumetric flow conditions, the spectral components of the acoustic signature signal become chaotic, and the spectral entropy value increases significantly.

[0030] In this embodiment, three features describe the characteristics of the acoustic signature signal from different dimensions, forming a complementary rather than redundant feature combination. A single feature, such as amplitude alone, is insufficient to distinguish between normal operating fluctuations and true pre-fault precursors; frequency alone may be insensitive to broadband random faults. Frequency features identify the occurrence or change of specific frequency components, corresponding to fault modes such as blade resonance and flow field pulsation; amplitude features reflect the intensity of the fault impact, corresponding to the degree of rubbing and vibration amplitude; spectral entropy characterizes the complexity and disorder of the system, and flow field instability under small volume flow conditions leads to a significant increase in spectral entropy. The feature vector formed by combining these three features has stronger fault discrimination capability and robustness, effectively identifying early, weak faults that are difficult to detect using traditional methods.

[0031] S4. The extracted acoustic signature feature vector is compared and analyzed with the pre-stored benchmark acoustic signature feature model of the steam turbine under safe operating conditions with small volume flow rate to obtain the safety assessment result. In a more specific embodiment provided in this application, the baseline voiceprint feature model is pre-stored in a baseline model library. The model is established as follows: under the safe operating condition of a steam turbine with low volumetric flow, multiple sets of voiceprint signals are collected. After preprocessing and feature extraction following the same procedure, a machine learning algorithm is used to train the baseline voiceprint feature model. As a specific implementation, 100 sets of voiceprint data under safe operating conditions are collected and trained using a support vector machine algorithm to obtain a hyperplane or probabilistic model that can characterize the voiceprint feature distribution under this operating condition. Other machine learning algorithms, such as Gaussian mixture models or hidden Markov models, can also be used.

[0032] The comparison analysis employs similarity calculation methods, specifically including Euclidean distance, cosine similarity, or correlation coefficient. As a specific implementation, this embodiment uses cosine similarity, calculating the cosine of the angle between the real-time extracted feature vector and the baseline model feature vector. The similarity value ranges from 0 to 1, with a higher value indicating a closer similarity between the two vectors. When the calculated similarity is below a set threshold, it is considered a potential security risk. As a specific implementation, the threshold is set at 0.85.

[0033] S5, issue a corresponding early warning signal based on the security assessment results.

[0034] In a more specific embodiment provided in this application, the warning signal includes multi-level warnings, with different warning levels corresponding to the severity of safety hazards in the safety assessment results. Different warning levels correspond to different warning methods and handling strategies. Specifically, the multi-level early warning system consists of three levels: When the similarity is between 0.8 and 0.85, it is determined that there is a minor anomaly, triggering a level one warning. The alarm method is to display corresponding text prompts on the monitoring screen to remind the operators to pay attention to changes in the equipment status. When the similarity is between 0.7 and 0.8, it is determined that there is a moderate anomaly, triggering a level 2 warning. An audible and visual alarm is used, that is, a buzzer alarm and a flashing light signal are issued simultaneously, requiring operators to check in time and take appropriate measures. When the similarity is below 0.7, it is determined that there is a serious anomaly, triggering a level 3 warning. The warning method is to send a prompt message to the central control center in addition to the audible and visual alarm, and recommend that the machine be stopped immediately for inspection.

[0035] Different levels of safety hazards require different response strategies. Minor anomalies can be handled by increased monitoring, avoiding immediate shutdown and unnecessary production losses. Moderate anomalies indicate that the fault is developing, requiring operator intervention to check and prepare load reduction or shutdown contingency plans. Severe anomalies indicate that the fault is quite serious, requiring immediate shutdown to prevent equipment damage. This multi-level early warning mechanism ensures safety while avoiding operational losses caused by overreaction.

[0036] Through real-time monitoring and tiered early warning of voiceprint signals, this method can issue alarms at the incipient stage of a fault. Actual field test results show that the fault identification accuracy rate can reach over 90%, and the total response time from signal acquisition to early warning output does not exceed 10 seconds, thus saving valuable time for operators to handle the situation.

[0037] See Figure 2 In another embodiment provided in this application, a small-volume flow safety system for steam turbines based on voiceprint auscultation is provided to implement the method described herein.

[0038] Includes the following modules: The acoustic signature acquisition module is used to collect acoustic signature signals from the steam turbine under low volumetric flow conditions. The module includes multiple acoustic sensors evenly distributed across the surface of the turbine cylinder, acquiring acoustic signature signals through a multi-sensor synchronous acquisition method. In this embodiment, eight piezoelectric acoustic emission sensors are used, with a frequency response range covering 20Hz to 100kHz and a sampling frequency of 50kHz.

[0039] The signal preprocessing module is used to denoise and filter the acquired voiceprint signal. It employs a wavelet threshold denoising algorithm for noise reduction and an adaptive filter for filtering. The wavelet threshold denoising algorithm specifically involves wavelet decomposition of the voiceprint signal, selection of a threshold function and threshold values ​​to process the wavelet coefficients, and then obtaining the denoised signal through inverse wavelet transform. The adaptive filter is used to eliminate 50Hz power frequency interference and its harmonic components.

[0040] The feature extraction module is used to extract voiceprint feature vectors from the preprocessed voiceprint signal. These voiceprint feature vectors include frequency features, amplitude features, and spectral entropy features. The feature extraction module extracts frequency features using a Fast Fourier Transform, extracts the peak amplitude of the signal as the amplitude feature, and calculates the spectral entropy feature based on the signal's power spectral density.

[0041] The benchmark model library is used to store the benchmark acoustic signature feature model of the steam turbine under safe operating conditions with small volume flow. The benchmark acoustic signature feature model is obtained by collecting 100 sets of acoustic signature signals under the safe operating conditions of the steam turbine with small volume flow, and then training them using the support vector machine algorithm after preprocessing and feature extraction.

[0042] The security analysis module is used to compare and analyze the extracted voiceprint feature vector with the benchmark voiceprint feature model to obtain the security assessment result. The security analysis module uses cosine similarity as the similarity calculation method. When the similarity is lower than 0.85, it is determined that there is a security risk.

[0043] The early warning module, connected to the security analysis module, is used to issue corresponding early warning signals based on the security assessment results. The early warning module issues multi-level early warning signals based on the security assessment results, with different early warning levels corresponding to different early warning methods and processing strategies. Specifically, the multi-level early warning is a three-level warning: a level one warning is triggered when the similarity is between 0.8 and 0.85, using a visual alarm; a level two warning is triggered when the similarity is between 0.7 and 0.8, using an audible and visual alarm; and a level three warning is triggered when the similarity is below 0.7, using a prompt alarm.

[0044] The connection relationships and data flow between the modules of this system are as follows: the voiceprint acquisition module transmits the acquired raw voiceprint signal to the signal preprocessing module, the signal preprocessing module transmits the processed signal to the feature extraction module, the feature extraction module transmits the extracted feature vectors in parallel to the security analysis module, the security analysis module calls the benchmark model from the benchmark model library for comparison and analysis and then transmits the evaluation result to the early warning module, and the early warning module issues a corresponding early warning signal based on the evaluation result.

[0045] The working principle of this system has been explained in detail in Example 1 along with the method steps, and will not be repeated here. This system realizes fully automated processing from signal acquisition, preprocessing, feature extraction, comparison and analysis to early warning output. It has the advantages of strong real-time performance, high accuracy, and non-intrusive installation, and can effectively ensure the safe operation of steam turbines under small volume flow conditions.

[0046] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A small volume flow safety method for steam turbines based on acoustic stethoscopy, characterized in that, Includes the following steps: Acoustic fingerprint signals of steam turbines under low volumetric flow conditions were collected. The acquired voiceprint signal is preprocessed, including noise reduction and filtering. Extracting voiceprint feature vectors from preprocessed voiceprint signals; The extracted acoustic signature feature vector is compared and analyzed with the pre-stored benchmark acoustic signature feature model of the steam turbine under safe operating conditions with small volume flow rate to obtain the safety assessment result. Based on the safety assessment results, corresponding early warning signals will be issued.

2. A small volume flow safety method for steam turbines based on acoustic auscultation according to claim 1, characterized in that, The specific steps for collecting acoustic fingerprint signals of a steam turbine under low volumetric flow conditions include: Multiple acoustic sensors are evenly distributed on the surface of the turbine cylinder block, and the acoustic signature signal is acquired through a multi-sensor synchronous acquisition method. The noise reduction process employs a wavelet threshold noise reduction algorithm, which specifically includes: performing wavelet decomposition on the collected voiceprint signal, selecting a threshold function and threshold to process the wavelet coefficients, and then obtaining the noise-reduced signal through inverse wavelet transform. The filtering process employs an adaptive filter to eliminate interference signals at specific frequencies.

3. A small volume flow safety method for steam turbines based on acoustic auscultation according to claim 1, characterized in that, The specific method for extracting the frequency features is as follows: extracting frequency features from the preprocessed voiceprint signal using Fast Fourier Transform; The specific method for extracting the amplitude feature is as follows: extract the peak amplitude of the preprocessed voiceprint signal as the amplitude feature; The specific method for extracting the spectral entropy features is as follows: the spectral entropy features are calculated based on the power spectral density of the preprocessed acoustic signature signal.

4. A small volume flow safety method for steam turbines based on acoustic auscultation according to claim 1, characterized in that, The comparison analysis uses similarity calculation methods, including Euclidean distance, cosine similarity, or correlation coefficient. When the calculated similarity is lower than a set threshold, it is determined that there is a security risk.

5. A small volume flow safety method for steam turbines based on acoustic auscultation according to claim 4, characterized in that, The warning signals include multi-level warnings, with different warning levels corresponding to the severity of safety hazards in the safety assessment results, and different warning levels corresponding to different warning methods and handling strategies; The multi-level early warning system consists of three levels. When the similarity is between 0.8 and 0.85, a level one early warning is triggered and an optical alarm is activated. When the similarity is between 0.7 and 0.8, a level 2 warning is triggered and an audible and visual alarm is activated. When the similarity is below 0.7, a level 3 warning is triggered and an alarm is triggered.

6. A small volume flow safety system for steam turbines based on acoustic stethoscopy, characterized by A method for implementing a small-volume flow safety method for steam turbines based on voiceprint auscultation as described in any one of claims 1-5 includes: The acoustic signature acquisition module is used to acquire acoustic signature signals of the steam turbine under small volume flow conditions; The signal preprocessing module is used to perform noise reduction and filtering on the acquired voiceprint signals; The feature extraction module is used to extract voiceprint feature vectors from the preprocessed voiceprint signal. The voiceprint feature vectors include frequency features, amplitude features, and spectral entropy features. A benchmark model library is used to store benchmark acoustic signature models of steam turbines under safe operating conditions with small volumetric flow rates. The security analysis module is used to compare and analyze the extracted voiceprint feature vector with the benchmark voiceprint feature model to obtain a security assessment result. The early warning module, connected to the security analysis module, is used to issue corresponding early warning signals based on the security assessment results.

7. A small volume flow safety system for steam turbines based on acoustic auscultation according to claim 6, characterized in that The acoustic signature acquisition module includes multiple acoustic sensors, which are evenly distributed on the surface of the turbine cylinder to acquire the acoustic signature signal through a multi-sensor synchronous acquisition method.

8. A small volume flow safety system for steam turbines based on acoustic auscultation according to claim 6, characterized in that The signal preprocessing module uses a wavelet threshold denoising algorithm for noise reduction and an adaptive filter for filtering. The wavelet threshold denoising algorithm specifically includes wavelet decomposition of the voiceprint signal, selection of a threshold function and threshold to process the wavelet coefficients, and then obtaining the denoised signal through inverse wavelet transform. The adaptive filter is used to eliminate interference signals at specific frequencies.

9. A small volume flow safety system for steam turbines based on acoustic auscultation according to claim 6, characterized in that The feature extraction module extracts the frequency features through Fast Fourier Transform, extracts the peak amplitude of the signal as the amplitude feature, and calculates the spectral entropy feature based on the power spectral density of the signal; the security analysis module uses similarity calculation methods for comparison analysis, including Euclidean distance, cosine similarity, or correlation coefficient.

10. A small volume flow safety system for steam turbines based on acoustic auscultation according to claim 6, characterized in that, The early warning module issues multi-level early warning signals based on the security assessment results. Different early warning levels correspond to different early warning methods and processing strategies. The multi-level early warning is a three-level early warning. When the similarity is between 0.8 and 0.85, a level 1 early warning is triggered and a light alarm is used. When the similarity is between 0.7 and 0.8, a level 2 early warning is triggered and an audible and visual alarm is used. When the similarity is below 0.7, a level 3 early warning is triggered and a prompt alarm is used.

Images