A method and system for processing anti-interference of acoustic emission signals of a thermal power plant fan

Abstract

The application discloses a kind of power plant fan acoustic emission signal anti-interference processing method and system based on, the method is by synchronous acquisition multiple fan's acoustic emission original signal and operating state parameter, time synchronization processing is carried out to acoustic emission signal, and candidate acoustic emission event is extracted based on adaptive threshold;Further based on multi-channel time difference of arrival, correlation characteristics and amplitude attenuation characteristics acoustic emission event candidate is carried out fan attribution determination, determines target fan;Combining rotational speed signal, damper opening and start-stop state, the operating process of target fan is divided into working condition stage;Fixed time window feature calculation is carried out to target fan acoustic emission event stream, and baseline compensation is carried out to different working condition stage based on stage dynamic baseline model;Residual signal is normalized, and based on dynamic threshold, fault discrimination is carried out, and the result is output to DCS system.The application can be used for online monitoring and early warning under multiple fan environment.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, and in particular to a method and system for anti-interference processing of acoustic emission signals from thermal power plant fans. Background Technology

[0002] In the operation of thermal power plants, the operation of fans, as key auxiliary equipment, directly affects the safety and stability of the boiler system and the overall unit. In recent years, acoustic emission detection technology has been gradually introduced into the field of fan fault monitoring. By collecting high-frequency elastic wave signals generated during equipment operation, early anomalies such as bearing wear, blade cracks, and rubbing can be identified.

[0003] However, in actual thermal power plant environments, multiple fans (such as induced draft fans, forced draft fans, and primary air fans) typically operate simultaneously. These fans are spatially adjacent to each other, and the acoustic emission signals are prone to superposition, reflection, and structural crosstalk after propagating through the casing, foundation, ductwork, and steel structure platform. This results in the signal collected by a single sensor often containing mixed components from multiple devices. Furthermore, the statistical characteristics of the acoustic emission signals of the fans differ significantly under different operating conditions, such as start-up, shutdown, load changes, and damper adjustments. Traditional methods based on fixed thresholds or single baseline models struggle to ensure consistent detection under different operating conditions.

[0004] Existing methods typically fail to effectively assign acoustic emission signals from multiple sources, making it difficult to distinguish signal origins. Furthermore, the lack of dynamic modeling mechanisms for different operating phases leads to misjudgments or omissions during operating condition transitions or background noise fluctuations. Therefore, there is an urgent need for an anti-interference method capable of accurately attributing acoustic emission signals in multi-fan coupled environments and adaptively processing them based on the characteristics of different operating phases. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0006] Therefore, to solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for anti-interference processing of acoustic emission signals from thermal power plant fans, comprising the following steps: Simultaneously collect raw acoustic emission signals and operating status parameters from multiple wind turbines; The original acoustic emission signal is time-synchronized, and candidate acoustic emission events are extracted based on an adaptive threshold, which is dynamically adjusted according to the envelope statistics within the historical normal operating condition reference window. Based on the time difference of arrival of multiple channels, correlation characteristics and amplitude attenuation characteristics, the wind turbine attribution of candidate acoustic emission events is determined, and the target wind turbine is determined according to the determination results. Based on the operating status parameters of the target wind turbine, the acoustic emission events corresponding to the target wind turbine are divided into operating condition stages, which include the start-up stage, the climbing stage, the steady-state stage, and the shutdown stage. The division criteria include the rate of change of rotational speed, the rate of change of damper opening, and the start-up and shutdown status. The valid acoustic emission events belonging to the target wind turbine are arranged in chronological order to form the target wind turbine acoustic emission event stream, and the acoustic emission characteristic value is calculated by segmenting the target wind turbine acoustic emission event stream into fixed time windows. The acoustic emission characteristic value is the window effective value RMS. A phased dynamic baseline model is established for different operating conditions, and the acoustic emission characteristic value is compensated to obtain the residual signal. The dynamic baseline model adopts a recursive update strategy, and the update conditions include no operating condition switching flag and the absolute value of the residual is less than the preset normal residual threshold. The residual signal is normalized within a stage to obtain a standardized residual signal. The standard deviation used for normalization is calibrated by offline health data and fixed during online operation. The standardized residual signal is judged based on the dynamic threshold, the fault warning result is output, and the linkage control information is sent to the DCS system.

[0007] As a preferred embodiment of the anti-interference processing method for acoustic emission signals of thermal power plant fans described in this invention, the following steps are taken: The raw acoustic emission signals of each fan are collected in real time using acoustic emission sensors. At least two acoustic emission sensors are arranged for each fan, with the acoustic emission sensor closest to the main bearing housing of the fan serving as the reference channel and the others as auxiliary channels. The operating status parameters of each fan are acquired in real time through a DCS system, a local PLC control unit, or a frequency converter control interface. These operating status parameters include at least the speed signal (in rpm), damper opening (in %), and start / stop status (start, stop, run). A unified timestamp is added to the raw acoustic emission signals and operating status parameters, with a timestamp accuracy of not less than 1 millisecond, forming multi-channel unified time series data to ensure strict time alignment of all signals in subsequent steps. The data is then transmitted to the central processing unit via an industrial Ethernet network.

[0008] As a preferred embodiment of the anti-interference processing method for acoustic emission signals of thermal power plant fans described in this invention, the following steps are performed: The original acoustic emission signals of each channel are aligned according to the unified timestamp, and sampling points at the same timestamp are used as data frames at the same time. If there are slight differences in the sampling rates of different channels, linear interpolation is used to resample each channel to a unified time axis. The aligned signals are then preprocessed and envelope extracted sequentially. The preprocessing includes: bandpass filtering using a fourth-order Butterworth filter with a passband frequency range of 50kHz to 500kHz; removing slow baseline drift using a sliding window mid-range filter with a window width of 0.1 seconds; performing a Hilbert transform on the filtered signal to obtain the envelope signal; using a sliding time window with a width of 20 microseconds to detect the energy of the envelope signal; when the average energy within the window exceeds an adaptive threshold and multiple consecutive windows meet this condition, the corresponding waveform segment is marked as a candidate acoustic emission event; the minimum duration is used to remove electromagnetic spikes and sampling glitches.

[0009] As a preferred embodiment of the anti-interference processing method for acoustic emission signals of thermal power plant fans described in this invention, the adaptive threshold is determined based on the envelope statistics within a historical normal operating condition reference window. Specifically, the threshold value is composed of the mean and standard deviation of the envelope signal within the window multiplied by a threshold coefficient. The historical normal operating condition reference window is a data segment that has been in a steady state and has not triggered anomaly markers within the last 30 minutes. The common background events obtained from subsequent fan attribution determination are used to further optimize the threshold parameters: every 10 minutes, the envelope amplitudes of the common background events collected within the last 10 minutes are taken as a batch, their mean and standard deviation are calculated, and they are weighted and fused with the statistics of the current reference window to enhance the adaptability of the threshold to changes in on-site noise. The weight ratio can be adjusted within a preset range according to the degree of on-site noise fluctuation.

[0010] As a preferred embodiment of the anti-interference processing method for acoustic emission signals of thermal power plant fans described in this invention, the arrival time and event amplitude of candidate acoustic emission events are extracted: the arrival time of each channel is determined by the AIC (Akaike Information Criterion) method, and the maximum absolute value of the sampling points within the event window is taken as the event amplitude; Based on the peak value of the normalized cross-correlation between the reference channel and other channels, the maximum normalized cross-correlation coefficient of each channel is calculated. When the coefficient of a channel with respect to the reference channel is lower than the effective channel correlation threshold (e.g., 0.3), the channel is removed from the effective channel set. If the number of effective channels is less than 2 after removal, the event is marked as "awaiting manual review". Based on the set of effective channels, three indicators are calculated: Channel correlation index: the average of the maximum normalized cross-correlation coefficients between the effective channel and the reference channel; Time difference consistency index: the theoretical time difference is calculated based on the equivalent propagation path length (calibrated by hammer test or approximated by Euclidean distance), and compared with the actual arrival time difference; the smaller the deviation, the higher the score; when the deviation exceeds the maximum allowable value, the channel's contribution is zero; Amplitude attenuation matching index: the theoretical amplitude attenuation ratio is calculated based on the exponential attenuation model calibrated on-site, and compared with the actual amplitude attenuation ratio; the smaller the deviation, the higher the score; when the deviation is greater than 1, the channel's contribution is zero; the above three indicators are weighted and fused to obtain the wind turbine attribution score, the weight coefficients are obtained through offline calibration, and satisfy the constraints of summation to 1 and non-negativity; then the maximum wind turbine attribution score is compared with the first threshold (e.g., 0.65) and the second threshold (e.g., 0.3). If the maximum attribution score is greater than or equal to the first threshold, it is classified as a valid acoustic emission event and assigned to the corresponding wind turbine; if the second threshold is less than or equal to the maximum attribution score and less than the first threshold, it is classified as a low-confidence segment (not involved in fault diagnosis, only used to update statistical parameters); if the maximum attribution score is less than the second threshold, it is classified as a common background event (not involved in fault diagnosis, used to update adaptive thresholds).

[0011] As a preferred embodiment of the anti-interference processing method for acoustic emission signals of thermal power plant fans described in this invention, the division of operating conditions includes: calculating the rate of change of the target fan's rotational speed (rpm / s) and the rate of change of the damper opening (% / s) with a fixed analysis interval of 1 second; and dividing the operating conditions according to the following rules in conjunction with the start-stop status: Start-up phase: Start-up status, and speed ≤ 10% of rated speed; Climbing phase: Start-up status, and speed between 10% and 95% of rated speed, with a speed change rate ≥ 5 rpm / s; Steady-state phase: Speed ​​> 95% of rated speed, and |speed change rate| ≤ 1 rpm / s, and |damper opening change rate| ≤ 2% / s; Shutdown phase: Shutdown status, and speed between 10% and 95% of rated speed, with a speed change rate ≤ -5 rpm / s; The above thresholds are preferred examples and can be adjusted according to the actual fan characteristics; When a change in the operating condition is detected, an operating condition switching flag is generated and maintained for a 2-second time window for subsequent control dynamic baseline updates and pseudo-anomaly detection.

[0012] As a preferred embodiment of the anti-interference processing method for acoustic emission signals from thermal power plant fans described in this invention, the staged dynamic baseline model and residual signal normalization include: Independent dynamic baseline models are established for the startup, climb, steady-state and shutdown phases respectively. During 72 hours of continuous, fault-free operation after system installation and commissioning, window-level acoustic emission characteristic values ​​under healthy conditions are collected for each operating stage. After removing outliers, the historical mean and historical standard deviation of each operating stage are calculated as the initial baseline parameters for the corresponding operating stage. When the target wind turbine enters a certain operating condition stage, the historical average value of that operating condition stage is used as the initial value of the dynamic baseline; if the operating condition stage is the first occurrence and the historical healthy sample is insufficient, the median of the acoustic emission characteristic values ​​within the first 10 time windows of that operating condition stage is used as the initial value. Within the current operating condition phase, the dynamic baseline is adaptively updated using a recursive update strategy only when both "no current operating condition switching flag" and "the absolute value of the current residual signal is less than the preset normal residual threshold" are simultaneously met. The preset normal residual threshold is preferably three times the standard deviation of the residual obtained from offline calibration. During the recursive update, the median of the acoustic emission characteristic values ​​within the current time window and the previous four time windows is used as the update input to suppress pulse interference contamination of the baseline. The difference between the current acoustic emission characteristic value and the dynamic baseline is used as the residual signal. In the offline calibration phase, a temporary reference baseline is constructed based on health data, preferably using the sliding median as the temporary reference baseline, and the standard deviation of the temporary residual is calculated accordingly as a fixed normalization parameter within the phase. In the online operation phase, the residual signal is divided by the standard deviation of the temporary residual and a very small positive number (preventing division by zero) is added to obtain a standardized residual signal, making its mean approach 0 and its standard deviation approximately 1 in a healthy state, thereby unifying the discrimination scale across different operating condition phases.

[0013] This invention also provides a system for processing acoustic emission signals from thermal power plant fans to suppress interference, which applies the method for processing acoustic emission signals from thermal power plant fans as described in any of the above claims. The system includes: The data acquisition module is used to collect the raw acoustic emission signals and operating status parameters of multiple wind turbines; The signal preprocessing module is used to perform time synchronization, alignment, filtering, and envelope extraction on the acquired signals, and to extract candidate acoustic emission events based on adaptive thresholds. The attribution and operating condition analysis module is used to determine the wind turbine attribution of candidate acoustic emission events to identify the target wind turbine, and to divide the target wind turbine into operating condition stages, outputting the target wind turbine acoustic emission event stream and common background events; The public background statistics update module is used to update the statistical parameters of the historical normal operating condition reference window based on the public background events, so as to correct the adaptive threshold. The dynamic baseline update module is used to establish a phased dynamic baseline model for different operating conditions and output residual signals based on the target wind turbine acoustic emission event stream. The normalization processing module is used to perform intra-stage normalization processing on the residual signal to obtain a standardized residual signal; The fault identification module is used to perform dynamic threshold identification and anomaly level determination based on the standardized residual signal. The dynamic threshold is set with different coefficients according to different working conditions (start-up / shutdown 3.5, ramp-up 3.0, steady-state 2.5), and the actual fault is judged based on duration, periodicity, repeatability and kurtosis (threshold 5). The linkage early warning module is used to send the anomaly judgment results (including target wind turbine number, operating condition stage, anomaly level, and recommended action) to the DCS system via OPC UA or Modbus TCP to perform alarm, load reduction, or shutdown protection. The storage and management module is used to store historical reference window parameters, dynamic baseline parameters, and calibration parameters (such as sound velocity, attenuation coefficient, weighting coefficient, etc.) for use by other modules.

[0014] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the anti-interference processing method for acoustic emission signals of thermal power plant fans as described in any of the above preferred embodiments.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for anti-interference processing of acoustic emission signals from a thermal power plant fan as described in any of the above preferred embodiments.

[0016] The beneficial effects of this invention are: 1. This invention introduces a wind turbine attribution determination mechanism based on multi-channel arrival time difference, correlation characteristics, and amplitude attenuation characteristics to quantitatively analyze candidate acoustic emission events, thereby distinguishing and attributing acoustic emission signals among multiple wind turbines and significantly improving signal recognition accuracy in complex structure propagation environments.

[0017] 2. This invention divides candidate events into valid acoustic emission events, low-confidence segments, and common background events, and uses common background events to dynamically weight and update the adaptive threshold, enabling the system to adaptively adjust the detection threshold according to changes in environmental noise, thereby enhancing the ability to suppress background interference.

[0018] 3. This invention divides the fan operation process into stages (start-up stage, climbing stage, steady-state stage, and shutdown stage) based on the speed change rate, damper opening, and start-up / stop status, and establishes independent dynamic baseline models for each stage, so that the acoustic emission characteristics under different operating conditions have corresponding reference benchmarks, ensuring the consistency of the detection results at each stage.

[0019] 4. This invention obtains residual signals by differential processing of acoustic emission characteristic values ​​and phased dynamic baselines, and combines residual normalization and dynamic threshold discrimination mechanism to comprehensively judge anomalies, effectively reducing false alarms and missed alarms caused by operating condition fluctuations or noise changes.

[0020] 5. This invention outputs the fault warning results to the DCS system to realize linkage control functions such as alarm prompts, automatic load reduction or shutdown protection, so that this method can be directly applied to the online monitoring and control system of thermal power plants. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the overall workflow of the present invention.

[0022] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention; the phrase "in an embodiment" appearing in different places in this specification does not all refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Example 1 Reference Figure 1 The first embodiment of the present invention provides a method for anti-interference processing of acoustic emission signals from thermal power plant fans, the method comprising the following specific steps: S100: Simultaneously acquires raw acoustic emission signals and operating status parameters from multiple wind turbines; S101. Collect raw acoustic emission signals from multiple wind turbines; The acoustic emission signals of each wind turbine are collected in real time by acoustic emission sensors installed in the bearing housing, casing or foundation of each wind turbine. Furthermore, each wind turbine is equipped with at least two acoustic emission sensors, wherein the acoustic emission sensor located near the main bearing housing is defined as the reference channel of the wind turbine for the wind turbine ownership determination in the subsequent step S300. The sampling frequency of the acoustic emission sensor The sampling frequency in this embodiment is 100kHz to 5MHz. Preferred setting To balance the capture of high-frequency components with data volume control.

[0027] S102. Collect fan operating status parameters; The operating status parameters of each fan are acquired in real time through the DCS system, local PLC control unit, or frequency converter control interface; the operating status parameters include at least: Speed ​​signal The unit is rpm, and k represents the fan number; this parameter is subsequently used to determine whether the fan is accelerating, moving at a constant speed, or decelerating. damper opening The unit is %; this parameter is subsequently used to determine whether the fan load is stable. Start-up and stop status, including at least start, stop, and run; this parameter is subsequently used to determine the status of the fan.

[0028] S103. Attach a unified timestamp and upload it to the central processing unit; The raw acoustic emission signal and the operating status parameters are appended with a unified timestamp to align the data of all channels, and then transmitted to the central processing unit via industrial Ethernet for caching and synchronous calculation. The unified timestamp has an accuracy of no less than 1ms; the central processing unit is an embedded industrial computer or an edge analysis host.

[0029] S200, time synchronization processing and candidate acoustic emission event extraction; S201, Time Synchronization and Preprocessing; The central processing unit aligns the original acoustic emission signals of each channel according to the timestamp index based on the unified timestamp added in step S103, that is, the sampling points under the same timestamp are used as data frames at the same time. If there are slight differences in the sampling rates of different channels, linear interpolation is used to resample each channel to a unified time axis. Preprocessing is performed on the aligned acoustic emission raw signal, including the following steps: A1. Bandpass Filtering: A fourth-order Butterworth filter is used to retain only the effective signal in the 50kHz to 500kHz frequency band. This frequency band is the main energy concentration area of ​​acoustic emission signals generated by wind turbine faults (such as bearing wear and gear damage). At the same time, it can effectively filter out low-frequency structural vibrations generated by the wind turbine itself (such as mechanical resonance below 10kHz) and high-frequency electromagnetic interference introduced by the electrical system (such as noise above 1MHz). By filtering the frequency band, the signal-to-noise ratio of subsequent analysis can be significantly improved, focusing on key fault characteristics. A2. Removing Slow Drift: To address the slow voltage drift phenomenon commonly found in the original signal (e.g., caused by changes in ambient temperature, sensor baseline offset, or circuit temperature drift), a sliding window median filter is used, with a window length of 0.1 seconds. Median filtering replaces the current value by calculating the median of the data points within the sliding window, effectively suppressing slowly changing trend components in the signal while retaining rapidly changing fault characteristics. The choice of a 0.1-second window length is based on the balance between the pulse characteristics of the wind turbine's acoustic emission signal and the drift speed. A window that is too short may not be able to sufficiently smooth the drift, while a window that is too long may obscure the abrupt changes in the true signal. This step eliminates the interference of slow drift on the subsequent amplitude threshold detection algorithm, ensuring the accuracy of fault determination. A3. Envelope Extraction: Perform a Hilbert transform on the filtered signal to obtain its envelope signal. Envelope processing can extract the amplitude variation trend of the original signal into a smooth curve. Its significance lies in the fact that acoustic emission signals are usually transient pulses, and their energy changes are difficult to observe directly in the time domain. Through envelope extraction, high-frequency oscillations can be converted into slowly varying low-frequency signals that reflect the energy envelope of the signal, making the start and end times and intensity changes of the pulse event more intuitive and providing a reliable basis for subsequent event detection and feature analysis.

[0030] S202. Extracting candidate acoustic emission events based on adaptive threshold; In order to obtain the envelope signal Each possible acoustic emission pulse is identified in the sample; this step uses a sliding time window for energy detection. The specific steps are as follows: Setting a width is preferred A sliding time window is used, starting from the signal's initial point. Each time, the window is moved one sampling point to the right, and the average energy of the envelope signal within the window is calculated. If the average energy within a certain window exceeds an adaptive threshold... If multiple consecutive windows satisfy this condition, then the waveform segments covered by these consecutive windows are marked as a candidate acoustic emission event. It should be noted that the width of the sliding time window It is selected based on the duration of a typical acoustic emission pulse, which can capture the transient characteristics of mechanical impacts relatively well; in practical applications, the sampling rate can be adjusted according to... Adjustment; Furthermore, define an adaptive threshold. Let be the threshold value of the k-th channel at the m-th analysis time, and its calculation formula is: ; In the above formula: This represents the mean envelope value of the k-th channel within the historical normal operating condition reference window. Let be the standard deviation of the envelope of the k-th channel within the historical normal operating condition reference window; This is the threshold coefficient, which can range from 2.5 to 4.0. In this embodiment, it is preferably set to... ; In this embodiment, the historical normal operating condition reference window is a data segment that has been in a steady state and has not triggered an anomaly flag within the last 30 minutes, and the reference window is updated every 10 seconds. Furthermore, when the envelope signal Continuously exceeding the adaptive threshold And the duration is not less than When there are 1 sampling point, the waveform segment is marked as a candidate acoustic emission event; If the sampling frequency is Then the time interval for a single sampling point is: ,at this time, The duration corresponding to each sampling point is approximately ; The minimum duration Used to remove electromagnetic spikes and sampling glitch that are too short in duration, in order to retain acoustic emission pulses with mechanical propagation characteristics; It should be noted that the common background events obtained from the classification in the subsequent step S300 are used to update the adaptive threshold parameters in step S202. , Specifically: Every 10 minutes, the envelope amplitudes of all public background events collected in the last 10 minutes are collected as a batch, and their mean and standard deviation are calculated. Then, these two statistics are weighted and fused with the existing mean and standard deviation of the current historical normal operating condition reference window (i.e., the steady-state data of the last 30 minutes) at a weight of 0.2:0.8 to obtain the updated threshold parameters. Furthermore, a weight ratio of 0.2:0.8 is a preferred example value, with the weight of the public background event statistics being 0.2 and the weight of the current reference window statistics being 0.8. This is used to improve the adaptability to changes in on-site noise while ensuring the stability of the threshold. This weight ratio can be adjusted within a preset range according to the degree of fluctuation in on-site noise.

[0031] S203, Generate a set of candidate acoustic emission events; Arrange all waveform segments that meet the above conditions in chronological order to form a candidate acoustic emission event set. Record the occurrence channel, start time, end time, and peak amplitude of each candidate acoustic emission event; The set of candidate acoustic emission events The waveform segment in the waveform will be further combined with the maximum ownership score in the subsequent wind turbine ownership determination step S300. Classify and process them.

[0032] S300, wind turbine ownership determination; S301. Extract the arrival time and amplitude information of candidate acoustic emission events; For each candidate acoustic emission event, the central processing unit extracts its arrival time and event amplitude within the time window corresponding to each channel. The specific steps are as follows: A1. The arrival time of candidate acoustic emission events is determined using the arrival time extraction method based on AIC (Akaike Information Criterion); assuming an event window sample sequence for a certain channel. Where N is the number of sampling points within the window; Define the AIC function as follows: , In the above formula, For the first sample The variance; For the second sample variance; boundary and ,as well as It is not included in the calculation to avoid the logarithmic parameter being zero or negative; it is taken as... Minimum sampling point As the arrival time location of the channel, the arrival time of the channel is... for: .

[0033] A2. Take the maximum absolute value of the sampling points within the event window as the event amplitude of that channel, specifically: ,in, For channel The event amplitude.

[0034] S302, Calculate the fan ownership score; For each wind turbine i, calculate the turbine's affiliation score based on the correlation, time difference consistency, and amplitude attenuation consistency of candidate acoustic emission events within the corresponding channel group of that wind turbine. The calculation formula is as follows: ; In the above formula, This is a channel correlation indicator; As an indicator of time zone consistency; For amplitude attenuation matching index; For the weighting coefficients, satisfying ; Furthermore, channel correlation indicators The calculation steps are as follows: For the i-th wind turbine, its reference channel is pre-defined. This refers to the acoustic emission sensor installed near the main bearing housing of the wind turbine; for the remaining channels k of wind turbine i, the normalized cross-correlation peak value between the aforementioned reference channel and this channel is calculated, as follows: a. Define the normalized cross-correlation function as: ; In the above formula: This is a sample event window for the reference channel; This is a sample of the event window for channel k; , These are the average values ​​within the corresponding event window; The length of the candidate event window; For time delay, the search range is based on the maximum straight-line distance between sensors. With the speed of sound Sure; b. Delay The search scope is defined as: , ; The reference channel represents the maximum Euclidean distance between other channels, expressed in meters. The speed at which sound waves propagate in a steel structure; This is a search margin used to compensate for sensor installation errors, propagation path approximation errors, and arrival time pickup errors. It should be noted that this embodiment takes This value was calibrated through an on-site lead-breaking test, repeated 5 times and the average value was taken. The calibration result was... ; The preferred value range is 0.02ms to 0.1ms, and the preferred value in this embodiment is [value missing]. This time difference margin is sufficient to cover measurement deviations in common industrial settings, while avoiding the problem of reduced time difference discrimination resolution due to an excessively large search range, thus achieving a balance between robustness and accuracy. c. Take the maximum value of the normalized cross-correlation function within the search range as the reference channel and the maximum normalized cross-correlation coefficient of channel k. ; d. Valid Channel Filtering: Define the set of valid channels. A channel that meets all of the following conditions: The channel sensor self-test is normal, with no open circuits or short circuits; the root mean square value of the signal within the last minute is within 0.5 to 2.0 times the historical normal range; the signal strength of this channel and the reference channel are... ; If a certain channel If it is not found, it will be removed from the valid set, and the coupling anomaly log for that acoustic emission sensor will be recorded; if the conditions are met after removal... If so, the candidate acoustic emission event is marked as pending manual review and will not be included in the automatic attribution; e. Channel correlation indicators The calculation formula is: ; In the above formula, ; In a preferred embodiment, when At that time, it was considered that the candidate acoustic emission event had a high channel correlation; Furthermore, time zone consistency indicators The calculation method is as follows: For the i-th wind turbine, calculate the theoretical propagation time difference and the actual propagation time difference between the reference channel and the other effective channels; a. Equivalent propagation path length Defined as the equivalent propagation path length between the reference channel and the kth effective channel; in this embodiment, this length is calibrated through a field hammer test: with the wind turbine stopped, an excitation is applied near the reference channel using a pulse hammer, and the arrival time difference between the reference channel and channel k is recorded. ,but: ; If a hammer test cannot be performed, the equivalent propagation path length can be approximated as the Euclidean distance between the installation positions of the two acoustic emission sensors: ; In the above formula, Let be the coordinate vectors of the installation positions of the acoustic emission sensors corresponding to the reference channel and channel k, respectively, in the same local three-dimensional coordinate system. Specifically, this can be expressed as: , The coordinates can be determined based on the equipment installation diagram, CAD layout diagram, or on-site measurement results. b. Calculate the theoretical propagation time difference and the actual propagation time difference; The formula for calculating the theoretical propagation time difference is as follows: The formula for calculating the actual propagation time difference is: In the above formula, , These are the arrival times of channel k and the reference channel, respectively; c. Before calculating time zone consistency, first determine Is it greater than : like If the channel is not valid, then skip that channel, indicating that there is no effective spatial resolution. in, One sampling period; d. Define the maximum permissible time difference deviation as: ; e. Time zone consistency index The calculation formula is: ; In the above formula, For the k-th channel pair time difference consistency index The contribution value; when the actual time difference is exactly the same as the theoretical time difference. When the deviation between the actual time difference and the theoretical time difference exceeds hour, 0 means no longer applying the time zone consistency index. It makes a positive contribution; In a preferred embodiment, when At that time, it was considered that the candidate acoustic emission event had good time difference consistency with the i-th wind turbine; Furthermore, amplitude attenuation matching index The calculation method is as follows: For the i-th wind turbine, calculate the actual amplitude attenuation ratio and the theoretical amplitude attenuation ratio between the reference channel and the other effective channels; a. The formula for calculating the actual amplitude attenuation ratio is: ; In the above formula, The event amplitude for channel k. For the reference channel's event amplitude, To prevent extremely small positive numbers with a denominator of zero; b. The theoretical amplitude attenuation ratio is calculated using an exponential attenuation fitting model based on field calibration, and its expression is: ; In the above formula, Let be the structural attenuation coefficient corresponding to the i-th wind turbine, in units of . ; It should be noted that, The calibration method is as follows: With the fan off, the same lead-broken source was used to excite the reference channel near it, and at distances of 0.2m, 0.5m, and 1.0m from it. The amplitude of the reference channel was measured. and the amplitude of channel k Calculate the amplitude ratio: and to Perform linear regression to obtain Among them, the regression slope The absolute value corresponds to the structural attenuation coefficient. ; The regression intercept is used to compensate for differences in field coupling and measurement bias. Example values ​​for this embodiment: Suitable for steel structure propagation environments; c. Amplitude attenuation matching index The calculation formula is: When the actual attenuation ratio equals the theoretical attenuation ratio, When the deviation between the actual attenuation ratio and the theoretical attenuation ratio is greater than 1, the amplitude attenuation matching sub-item corresponding to the k-th channel is set to 0, meaning that the amplitude attenuation matching index is no longer adjusted. It makes a positive contribution; In a preferred embodiment, when At that time, it was considered that the candidate acoustic emission event had good consistency with the amplitude attenuation of the i-th wind turbine; Furthermore, weighting coefficients The training samples were obtained through offline calibration. Let the number of training samples used for offline calibration be... , The preferred value range is 30~100, and in this embodiment, the preferred value is... To ensure calibration stability while controlling on-site calibration costs; the specific calibration method is as follows: a. Under healthy conditions, generate standard acoustic emission events at a known wind turbine location using a lead breakage test or pulse hammer, and collect the data. A sample of standard acoustic emission events from known sources; b. For each sample j from a known source ( ): Let the actual source fan number of sample j be... For each wind turbine i ( ,in (For the total number of wind turbines), calculate the channel correlation index of sample j relative to wind turbine i. Time zone consistency index and amplitude attenuation matching index , constitute the feature vector Simultaneously, define the supervisory labels for the sample-fan pair. :like ,but , indicating that sample j originates from wind turbine i; otherwise ; c. From all samples j ( For each fan i ( The training dataset is constructed using the feature vectors and labels of the given data. ,in , ;common One training instance is used to solve for the weight coefficients. ; d. Solve for the weight coefficients using the constrained least squares method, i.e.: The constraints are: , That is, under the condition that the sum of the weight coefficients is 1 and non-negative, minimize the sum of the squared errors between the predicted value and the true label; For example, this embodiment was calibrated to obtain: Channel correlation index The highest weight indicates that the correlation between signals in different sensor channels has the greatest impact on the positioning results. Time zone consistency index Secondly, it demonstrates the crucial role of the time difference in determining the source location; Amplitude decay index The lower weight indicates that although the attenuation law of signal amplitude with propagation distance is important, it is greatly affected by environmental interference, so it is assigned a lower weight. It should be noted that, The value of ensures that the system has a sufficient sample size to guarantee the model's generalization ability to various acoustic emission events, and avoids increasing the workload of on-site calibration (such as the number of lead breakage tests) due to an excessively large number of samples; therefore, 50 samples can effectively balance implementation costs while ensuring accuracy.

[0035] S303, Identify the target wind turbine and output common background events; Wind turbine affiliation score for all wind turbines Compare, take The fan i corresponding to the maximum value is taken as the target fan: ; Let the maximum belonging score be ,according to Within the specified interval, candidate acoustic emission events are categorized into the following three types: If Then, the candidate acoustic emission event is marked as belonging to the target wind turbine. Valid acoustic emission events; if This segment is marked as low-confidence and does not participate in fault diagnosis. It is only used to update the statistical parameters of the historical normal operating condition reference window in step S202 after removing time windows with uncertain attribution. It is marked as a public background event and does not participate in fault diagnosis. It is used for the adaptive threshold parameter update in step S202 (see S202 for specific fusion method). in, The threshold for wind turbine affiliation. The background noise threshold is given, and it satisfies... The above , The values ​​are all determined by the field calibration data; during calibration, the distribution of wind turbine attribution scores is statistically analyzed based on standard acoustic emission event samples from known sources, and the attribution accuracy, misattribution rate and missed attribution rate on the validation set are jointly determined as optimization objectives; In this embodiment, the wind turbine affiliation threshold The preferred value is 0.65. In practical applications, it can be adjusted within the range of 0.55 to 0.75 according to the crosstalk level of multiple fans on site, in order to balance the accuracy of attribution and robustness; background noise threshold The preferred value is 0.3, but in practical applications it can be adjusted within the range of 0.2 to 0.4 according to the background noise level.

[0036] S304, Output target fan acoustic emission event stream; The valid acoustic emission events belonging to the target wind turbine are output in chronological order as the target wind turbine acoustic emission event stream. This serves as the input for subsequent working condition division and baseline reconstruction.

[0037] S400, Operating Condition Stage Division; S401, Calculate the rate of change of rotational speed and the rate of change of damper opening; By analysis interval Calculate the rate of change of the target wind turbine's rotational speed per second. (Unit: rpm / s), the calculation formula is as follows: In the above formula, Let be the rotational speed at time t, and the speed be rpm; Damper opening change rate Calculation method and speed change rate Consistent (unit: % / s), that is: ;in, Let be the damper opening at time t, in units of %.

[0038] S402, Divide the system into startup phase, climb phase, steady-state phase, and shutdown phase; Based on the rate of change of the target fan speed The start / stop status and changes in damper opening are classified into operating stages according to the following rules: Startup phase: The start / stop status is "start," and Climbing phase: The start / stop status is "started," and... ,and Steady-state phase: ,and And the rate of change of damper opening ;Stop section: The start / stop status is stopped, and ,and In this embodiment, The rated speed of the fan is taken from the rated speed on the fan's nameplate. For example, the induced draft fan is usually 990 rpm or 1480 rpm. It should be noted that the threshold values ​​for the above-mentioned operating condition stages ( , , , , The values ​​are preferred examples in this embodiment, determined based on the wind turbine nameplate parameters and on-site dynamic characteristics. They can be appropriately adjusted within the range of 0.05~0.15, 0.90~0.98, 3~8 rpm / s, 0.5~2 rpm / s, and 1~3% / s according to the specific wind turbine model, peak shaving depth, and operating procedures. This does not constitute a limitation on the scope of protection of this invention.

[0039] S403, Generate operating condition switching flag; set up This indicates a switching flag; when the operating condition of the target wind turbine changes, it will... Set to 1 and maintain it for the next 2-second window. This is used to mark the current operating condition transition interval; when the 2-second window ends or no operating condition phase change occurs... Set to 0; This is the default state and no additional maintenance time is required; the above operating condition switching flag is used to control dynamic baseline updates and pseudo-anomaly detection.

[0040] S404, Output operating condition stage label; Acoustic emission event stream for target wind turbine Each moment in the process is labeled with an additional operating condition stage. .

[0041] S405, Calculate the acoustic emission characteristic values; The specific steps are as follows: C1. Set a fixed time window; This step uses a fixed time window width. Second-to-target wind turbine acoustic emission event stream Segmentation; It should be noted that the fixed time window width of 0.1s is the preferred value for window-level statistical features, and in practical applications, it can be adjusted between 0.05 and 0.2s according to real-time requirements. This fixed time window width is the same as the sliding time window used for candidate event extraction in step S202. The purposes are different: a tiny sliding window is used for high-precision pulse detection, while a fixed time window width is used for window-level statistical feature extraction; The number of sampling points corresponding to this fixed time window is: ; C2. Calculate the window-level acoustic emission eigenvalues ; For waveform samples within the t-th time window The effective value RMS is calculated as the acoustic emission characteristic value. The calculation formula is as follows: ; In the above formula, the acoustic emission characteristic value The dimensions of are consistent with the output voltage of the acoustic emission sensor, typically in mV or V.

[0042] S500, phased dynamic baseline model; S501. Establish a phased baseline model library; An independent baseline model is established for each operating stage of the wind turbine (start-up, ramp-up, steady-state, and shutdown); each baseline model is composed of acoustic emission characteristic values ​​under healthy conditions in that operating stage. mean and standard deviation The two parameters describe the condition, where the subscript 's' indicates the operating stage. Furthermore, the steps for establishing the baseline model described above are as follows: After system installation and commissioning, acoustic emission characteristic values ​​were continuously collected during 72 hours of continuous, fault-free operation of the wind turbine at various operating conditions. Based on the working condition classification results of step S400, each acoustic emission characteristic value is... Mark it to the corresponding operating condition stage s; For each operating condition stage, adopt Principles of elimination and mean Deviation exceeding 3 times the standard deviation To identify obvious outliers (i.e., samples) to avoid occasional disturbances contaminating the baseline; For the health data after removing outliers, calculate all acoustic emission characteristic values ​​within this stage. The arithmetic mean is used as... Calculate its sample standard deviation as Each stage As the initial baseline parameters for this stage, they are stored in the model library; During online operation, the above global statistical parameters No longer updated, only dynamic baseline. The parameters are updated in real time according to the recursive method of subsequent steps S503; the parameters between each working condition stage are independent of each other and do not affect each other.

[0043] S502. Initialize the baseline for the current operating condition stage; When the target wind turbine enters a new operating condition stage s, the historical average parameter of this stage will be used. As the initial value for the dynamic baseline: In the above formula, The mean value of the health data for this working condition stage calculated in step S501; If this operating condition phase occurs for the first time and the historical baseline is insufficient (e.g., no startup data was collected during the training phase), then the initial 10 time windows of this operating condition phase will be used. The median is used as the initial value for the dynamic baseline, i.e.: .

[0044] S503, Perform recursive updates within the same operating condition phase; Within the current operating condition phase s, the acoustic emission characteristic values ​​at each time t. First, determine if all of the following update conditions are met: a. There is no operating condition switching indicator in the current time window, that is... ; b. The current residual satisfies ,in The preset normal residual threshold is defined as follows: ; in, This is the standard deviation of the residual obtained from offline calibration in step S600; this threshold is not used for the fault dynamic threshold in S700, but is only used to determine whether baseline updates are allowed, so as to avoid fault impacts contaminating the baseline. The residual signal is calculated as follows: Based on the acoustic emission characteristic values ​​of the current time window Dynamic baseline with current operating condition stage s Calculate the residual signal The calculation formula is: This residual signal eliminates baseline drift caused by changes in operating conditions (such as speed fluctuations and load variations), ensuring the residual signal remains stable under healthy conditions. The residual signal is close to 0, while the fault impulse manifests as a significant positive or negative pulse, thus highlighting the fault characteristics; the residual signal calculated in this step... This will be used as the direct input for the subsequent step S600; Furthermore, when all the above conditions are met, the dynamic baseline is updated using the following recursive formula: ; In the above formula, Forgetting factors related to different operating conditions are set according to the degree of non-stationarity of each operating condition. Example values ​​are as follows: Startup phase: Climbing section: Steady-state phase: Stop section: ; For the window statistics used for updating, this embodiment preferably takes the current time window and the previous four time windows (a total of 0.5 seconds). The median of: To suppress pulse interference; It should be noted that the start-up and shutdown phases are characterized by drastic fluctuations, so larger values ​​are used to quickly track the baseline; the ramp-up phase is the next best; the steady-state phase is the most stable, so smaller values ​​are used to maintain baseline stability; the above values ​​are preferred examples in this embodiment, and can be adjusted within the range of 0.01 to 0.3 according to the dynamic characteristics of the wind turbine and field tests. Furthermore, when none of the above conditions are met: if If the duration exceeds 60 seconds, regular updates will be frozen and replaced with extremely low weighting. and The median is tracked slowly to prevent long-term baseline deviation.

[0045] S504, Baseline connection during operating condition switching; When a change in operating condition occurs, the dynamic baseline value of the previous operating condition is... It is no longer used, but is only retained in the historical record of this stage (for subsequent analysis or retrospective review), and is not passed to the new stage; the dynamic baseline of the new stage is reinitialized to the historical mean parameter of this operating condition stage. : .

[0046] S600, residual signal intra-stage normalization; This step is divided into two stages: offline calibration and online execution, as detailed below: D1. Offline calibration stage: For each operating condition phase s, temporary residuals are calculated based on health data. Standard deviation The specific steps are as follows: Constructing a temporary reference baseline Preferably, all healthy individuals within the specified operating condition phase s The sliding median, with a window width consistent with the online recursive window (0.5 seconds); the temporary residuals are calculated using the following formula. : ; Statistical temporary residuals Standard deviation : ; In the above formula, Number of healthy samples; This is the temporary residual mean, which is theoretically close to 0; In another preferred embodiment, healthy samples can also be used directly. The sliding standard deviation An approximate estimate is performed as a simplified implementation for the offline calibration stage; It should be noted that, Once the system installation and debugging are complete, it will be fixed and will no longer be updated online. D2. Online operation phase; For residual signals Normalization is performed to obtain the standardized residual signal. The calculation formula is as follows: In the above formula, To prevent extremely small positive numbers from being divided by zero; Due to the use of offline fixed , It has a stable statistical scale within the same operating condition stage, with a mean of 0 and a standard deviation of about 1 when healthy, thus avoiding circular dependencies in online calculations.

[0047] S700, dynamic threshold discrimination and fault early warning; S701, Establish dynamic thresholds for each stage; For standardized residual signals Since the mean is 0 and the standard deviation is 1 under healthy conditions, a fixed constant threshold can be directly used without calculating the mean and standard deviation separately for each operating condition stage. Considering the differences in fault sensitivity of s under different operating conditions, this embodiment sets different threshold coefficients. The dynamic threshold is defined as: In the above formula, The threshold coefficients related to the operating condition stage s are determined by optimizing the ROC curves of historical fault samples to achieve a balance between false alarm rate and false negative rate for different operating condition stages s. This embodiment determines the following example values ​​based on the noise fluctuation characteristics and false alarm / missed alarm balance at different operating stages: Start-up stage: Climbing section: Steady-state phase: Stop section: ; It should be noted that, due to the drastic changes in operating conditions and large background fluctuations during the start-up and shutdown phases, a larger threshold is set to avoid misreporting normal transients as faults; the steady-state phase is relatively stable, so a smaller threshold can be set to improve sensitivity; the climb phase is in between. when At that time, it was considered that the residuals of that time window deviated significantly from the healthy baseline.

[0048] S702, distinguish between start-stop transients, pseudo-abnormalities during operating condition switching and real fault impacts; The following comprehensive criteria are adopted: (1) Pseudo-anomaly during operating condition switching: When the standardized residual signal satisfies However, the duration does not exceed 0.5 seconds, that is, within 5 consecutive time windows, and the anomaly occurs during the operating condition switching indicator. Within the subsequent 2-second window, it was determined to be a pseudo-anomaly during the operating condition switch; (2) Real fault impact: When the standardized residual signal satisfies A true fault impact is defined as an event that occurs when the duration of this state is not less than 0.5 seconds and at least one of the following conditions is met: a) Standardized residual signal The envelope spectrum exhibits periodic peaks related to rotational speed; among them, through the analysis of... The sequence is subjected to Hilbert transform to obtain the envelope signal, and then subjected to Fast Fourier Transform (FFT) to obtain the envelope spectrum. The criteria for judging the periodic peak are: the peak amplitude is greater than 3 times the standard deviation of the mean of the baseline spectral line of the envelope spectrum, and the deviation between the peak frequency and the rotational speed characteristic frequency does not exceed 5%. The rotational speed characteristic frequency is an integer multiple of the frequency obtained by dividing the current rotational speed of the fan by 60. b. The same operating condition recurs in multiple consecutive start-stop cycles, indicating the persistence of the fault rather than an occasional transient event. c. The kurtosis coefficient satisfies: To determine whether the residual signal has a "peak" characteristic (a typical statistical feature of fault impact); in this formula, This represents the number of consecutive time windows within the current operating condition phase. To standardize the residual signal The average value during this operating condition phase; (3) Public background noise: If none of the above conditions are met, and When randomly distributed, it is determined to be common background noise; It should be noted that a kurtosis threshold value of 5 is the preferred value in this embodiment, which was determined through statistical analysis of historical fault samples and can be adjusted within the range of 4 to 6. The aforementioned time threshold, spectrum criterion, and statistical threshold were all obtained through empirical calibration using on-site health data and known fault data. Their value ranges conform to the typical statistical characteristics of acoustic emission signals from wind turbine units, and are used to improve the accuracy and robustness of fault identification.

[0049] S703 outputs fault warning results and links to the DCS system; When a genuine fault impact is determined, the central processing unit outputs a fault warning result containing the following information: target turbine number; current operating stage; fault level, classified as mild, moderate, or severe; and recommended control actions, such as "planned shutdown for inspection" or "emergency load reduction".

[0050] The anomaly level is based on the standardized residual signal. Exceeding the dynamic threshold Divide the multiples into categories: when If the system detects a minor anomaly, it will output an inspection prompt. when When the condition is determined to be a moderate anomaly, the system outputs a load reduction suggestion; when If the system is deemed to have a serious anomaly, it will output a suggested planned shutdown or emergency shutdown. At the same time, the alarm message is sent to the DCS system via OPCUA or Modbus TCP, and the DCS performs alarm prompts, automatic load reduction or shutdown protection according to the preset logic; It should be noted that the above-mentioned boundary coefficients of 1.5 and 2.0 for the abnormality level are preferred example values ​​in this embodiment, which are determined by statistical analysis of historical fault samples and normal samples. They can be adjusted within the range of 1.2~1.8 and 1.8~2.5 according to the specific fan model, on-site noise level and tolerance for false alarms and missed alarms, and do not constitute a limitation on the scope of protection of this invention.

[0051] Example 2 Reference Figure 2 This is the second embodiment of the present invention, which differs from the first embodiment in that: the present invention also provides a system for processing acoustic emission signals from thermal power plant fans to resist interference. This system is used to implement the method for processing acoustic emission signals from thermal power plant fans to resist interference described in Embodiment 1. The system includes: Data acquisition module: used to collect raw acoustic emission signals and operating status parameters of multiple wind turbines; Signal preprocessing module: used to perform time synchronization, alignment, filtering, and envelope extraction on the acquired signals, and to extract candidate acoustic emission events based on adaptive thresholds; Attribution and Operating Condition Analysis Module: Used to determine the wind turbine attribution of candidate acoustic emission events to identify the target wind turbine, and to divide the target wind turbine into operating condition stages, outputting the target wind turbine acoustic emission event stream and common background events; Public background statistics update module: used to update the statistical parameters of the historical normal operating condition reference window based on the public background events, so as to correct the adaptive threshold; Dynamic baseline update module: used to establish a phased dynamic baseline model for different operating conditions and output residual signals based on the target wind turbine acoustic emission event stream; Normalization processing module: used to perform intra-stage normalization processing on the residual signal to obtain a standardized residual signal; Fault discrimination module: used to perform dynamic threshold discrimination and anomaly level determination based on the standardized residual signal; wherein, the dynamic threshold is set with different coefficients according to different working conditions, and the actual fault is judged based on duration, periodicity, repeatability and kurtosis. Linkage and early warning module: Used to send the anomaly judgment results (including target turbine number, current operating condition stage, anomaly level, and recommended control actions) to the DCS system via OPC UA or Modbus TCP protocol, and perform alarm prompts, automatic load reduction or shutdown protection; Storage and Management Module: Used to store historical reference window parameters, dynamic baseline parameters, and calibration parameters (including sound velocity, structural attenuation coefficient, weighting coefficient, etc.) for use by other modules.

[0052] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for anti-interference processing of acoustic emission signals from thermal power plant fans, characterized in that: Includes the following steps: Simultaneously collect raw acoustic emission signals and operating status parameters from multiple wind turbines; The raw acoustic emission signal is time-synchronized, and candidate acoustic emission events are extracted based on an adaptive threshold. Based on the time difference of arrival of multiple channels, correlation characteristics and amplitude attenuation characteristics, the wind turbine attribution of candidate acoustic emission events is determined, and the target wind turbine is determined according to the determination results. Based on the operating status parameters of the target wind turbine, the acoustic emission events corresponding to the target wind turbine are divided into operating condition stages, which include the start-up stage, the ramp-up stage, the steady-state stage, and the shutdown stage. The valid acoustic emission events belonging to the target wind turbine are arranged in chronological order to form the target wind turbine acoustic emission event stream, and the acoustic emission characteristic values ​​are calculated in segments with fixed time windows. A phased dynamic baseline model was established for different operating conditions, and the acoustic emission characteristic values ​​were compensated for to obtain the residual signal. The residual signal is normalized within a stage to obtain a standardized residual signal; The standardized residual signal is judged based on the dynamic threshold, the fault warning result is output, and the linkage control information is sent to the DCS system.

2. The method for anti-interference processing of acoustic emission signals from thermal power plant fans as described in claim 1, characterized in that: The acoustic emission raw signals of each wind turbine are collected in real time by acoustic emission sensors, and the acoustic emission sensor closest to the main bearing housing of the wind turbine is used as the reference channel of the wind turbine; the operating status parameters of each wind turbine are acquired in real time, including at least the speed signal, damper opening degree and start / stop status; after adding a unified timestamp to the acoustic emission raw signals and operating status parameters, a multi-channel unified time series data is formed.

3. The method for anti-interference processing of acoustic emission signals from thermal power plant fans as described in claim 2, characterized in that: Align the signals of each channel according to the unified timestamp, and perform preprocessing and envelope extraction on the aligned signals in sequence; use a sliding time window to perform energy detection on the envelope signal, and when the average energy in the window exceeds the adaptive threshold and multiple consecutive windows meet the condition, mark the corresponding waveform segment as a candidate acoustic emission event.

4. The method for anti-interference processing of acoustic emission signals from thermal power plant fans as described in claim 3, characterized in that: The adaptive threshold is determined based on the envelope statistics within the historical normal operating condition reference window. The reference window is a data segment that has been in a steady state and has not triggered any anomalies within the most recent preset time period, and it is updated by sliding according to a preset step size.

5. The method for anti-interference processing of acoustic emission signals from thermal power plant fans as described in claim 4, characterized in that: The determination of wind turbine ownership includes: Extract the arrival time and event amplitude of candidate acoustic emission events; Based on the peak value of the normalized cross-correlation between the reference channel and other channels, when the maximum normalized cross-correlation coefficient between a channel and the reference channel is lower than the effective channel correlation threshold, the channel is removed from the effective channel set. Based on the set of effective channels, channel correlation index, time difference consistency index and amplitude attenuation matching index are calculated respectively. The channel correlation index is determined based on the average value of the maximum normalized cross-correlation coefficient between the reference channel and the effective channel. The time difference consistency index is calculated based on the equivalent propagation path length and arrival time difference. The amplitude attenuation matching index is calculated based on the actual amplitude attenuation ratio and the theoretical amplitude attenuation ratio. The three indicators are weighted and combined to obtain the wind turbine ownership score. The maximum wind turbine ownership score is then compared with the first threshold and the second threshold, where the first threshold is greater than the second threshold. If the maximum attribution score is greater than or equal to the first threshold, it is classified as a valid acoustic emission event; If the maximum attribution score is greater than or equal to the second threshold and less than the first threshold, it is classified as a low-confidence segment. If the maximum attribution score is less than the second threshold, it is classified as a public background event.

6. The method for anti-interference processing of acoustic emission signals from thermal power plant fans as described in claim 5, characterized in that: The division of operating conditions includes: calculating the rate of change of the target fan speed and the rate of change of the damper opening at fixed time intervals, and dividing the fan operation process into segments based on the start-up and stop states; dividing the operation process into start-up segment, ramp-up segment, steady-state segment or stop segment according to the start-up and stop states, speed range and rate of change of speed; when it is determined that the operating condition stage has changed, generating an operating condition switching flag and maintaining it for a preset time window.

7. The method for anti-interference processing of acoustic emission signals from thermal power plant fans as described in claim 6, characterized in that: The phased dynamic baseline model and residual signal normalization include: establishing independent dynamic baseline models for each different operating stage of the wind turbine, and using the average acoustic emission characteristic value of the wind turbine in its historical healthy state as the initial value of each baseline; in the current operating stage, when the residual threshold constraint is met and there is no operating condition switching flag, an adaptive update strategy is used to update the dynamic baseline; the difference between the current acoustic emission characteristic value and the dynamic baseline is used as the residual signal, and then the residual signal is normalized by combining it with the residual standard deviation obtained from offline calibration, finally obtaining the standardized residual signal.

8. A system for processing acoustic emission signals from thermal power plant fans, employing the method for processing acoustic emission signals from thermal power plant fans as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect the raw acoustic emission signals and operating status parameters of multiple wind turbines; The signal preprocessing module is used to perform time synchronization, alignment, filtering, and envelope extraction on the acquired signals, and to extract candidate acoustic emission events based on adaptive thresholds. The attribution and operating condition analysis module is used to determine the wind turbine attribution of candidate acoustic emission events to identify the target wind turbine, and to divide the target wind turbine into operating condition stages, outputting the target wind turbine acoustic emission event stream and common background events; The public background statistics update module is used to update the statistical parameters of the historical normal operating condition reference window based on the public background events, so as to correct the adaptive threshold. The dynamic baseline update module is used to establish a phased dynamic baseline model for different operating conditions and output residual signals based on the target wind turbine acoustic emission event stream. The normalization processing module is used to perform intra-stage normalization processing on the residual signal to obtain a standardized residual signal; The fault detection module is used to perform dynamic threshold detection and anomaly level determination based on the standardized residual signal; The linkage early warning module is used to send the anomaly judgment result to the DCS system to execute alarm, load reduction or shutdown protection; The storage and management module is used to store historical reference window parameters, dynamic baseline parameters, and calibration parameters, which can then be used by other modules.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the anti-interference processing method for acoustic emission signals of thermal power plant fans as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the anti-interference processing method for acoustic emission signals of thermal power plant fans as described in any one of claims 1 to 7.

Images