Intelligent monitoring and diagnosing method for operation state of boiler induced draft fan of thermal power plant

Abstract

The invention discloses an intelligent monitoring and diagnosing method for the operation state of a boiler induced draft fan of a thermal power plant, and relates to the technical field of power equipment monitoring. The method comprises the following steps: multi-modal heterogeneous data acquisition and time-space alignment: acquiring acoustics, temperature field and airflow signals through an optical fiber acoustic sensor, an infrared thermal imaging array and the like, and associating the acoustics, temperature field and airflow signals; fault features are extracted through improved adaptive multi-scale morphological decomposition, and feature separation is realized through dynamic adjustment of structural elements; and fault mode matching and result output. The method breaks through the limitation of traditional single parameter monitoring, realizes early accurate diagnosis of the fault of the induced draft fan through multi-physics field fusion and adaptive algorithm optimization, and is suitable for intelligent operation and maintenance of key equipment of a thermal power plant.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of equipment detection and diagnosis, in particular to an intelligent monitoring and diagnosis method for the operation state of a boiler induced draft fan in a thermal power plant. BACKGROUND

[0002] The boiler induced draft fan in a thermal power plant is a key equipment for ensuring combustion efficiency and flue gas emission, and its operation state directly affects the safety and economy of the unit. The existing monitoring and diagnosis methods have the following defects: Single data acquisition: mainly relying on vibration acceleration sensors or current transmitters, only monitoring local vibration or energy consumption parameters, ignoring key fault precursors such as abnormal noise of impeller wear and temperature field distortion of bearing overheating; Limited feature extraction: traditional methods use fixed scale wavelet decomposition or Fourier transform, which cannot adaptively match the wide frequency fault features of the induced draft fan (such as low frequency pulsation of surge and high frequency impact of blade crack), resulting in feature confusion; Diagnosis lag: the recognition accuracy of early weak faults (such as bearing seat loosening and impeller dust accumulation) is less than 60%, and the fault may be enlarged due to missed diagnosis.

[0003] Therefore, an intelligent monitoring and diagnosis method that can perceive multi-dimensional and adaptively extract features is needed to achieve early and accurate identification of induced draft fan faults. SUMMARY

[0004] The present application aims to provide an intelligent monitoring and diagnosis method for the operation state of a boiler induced draft fan in a thermal power plant to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides the following technical solution: an intelligent monitoring and diagnosis method for the operation state of a boiler induced draft fan in a thermal power plant, comprising the following steps: S1: multi-modal heterogeneous data acquisition and space-time alignment, acquiring acoustic, thermal and airflow disturbance signals during the operation of the induced draft fan through a non-invasive composite sensor array, and correlating the time and space dimensions to obtain the collected data; the specific steps are as follows: S1.1: Acoustic signal: deploying 6 optical fiber acoustic sensors (sampling frequency 10 kHz) around the shell of the induced draft fan to collect wide frequency vibration sound signals , where is the timestamp, , including impeller rotating noise (50-500 Hz) and fault impact sound (1-5 kHz); temperature field distribution: using an infrared thermal imaging array (resolution 320x240) to shoot the bearing seat area and generate a temperature field matrix , where Space coordinates, unit: mm, sampling period: 1 s; Airflow disturbance: install 3 groups of micro-pressure difference sensors (interval 120°) at the inlet pipe of the induced draft fan, calculate the airflow disturbance coefficient wherein is the standard deviation of pressure difference, is the average pressure difference, reflecting the intensity of airflow turbulence; S1.2: Time alignment: take the pulse signal of the induced draft fan main shaft encoder as the reference clock (synchronization error <1 ms), unify the acoustic signal (10 kHz), temperature field (1 Hz), and airflow disturbance (10 Hz) to the same time axis ; S1.3: Spatial correlation: establish the mapping relationship between the sensor position and the structure of the induced draft fan, including the first quadrant area corresponding to the impeller, bind the acoustic signal with the temperature of the area to generate a spatiotemporal fusion matrix: wherein, is the 3 groups of airflow disturbance coefficients; S2: Fault feature extraction based on improved adaptive multi-scale morphological decomposition, adaptive decomposition and sensitive feature screening of collected data; The specific steps are as follows: S2.1: Signal preprocessing: extract single-channel acoustic signal (focus on fault sensitive channel) from the fusion matrix , and perform DC removal and normalization processing: wherein, is the mean value, is the standard deviation; S2.2: Improved AMMD algorithm decomposition: Initialization: set the minimum scale , maximum scale , unit: sampling points, and the initial morphology of the structure element is a flat hat shape ; Calculate the kurtosis of the current signal segment wherein is the expectation operator, if (high-frequency impact feature), then decrease the scale ; if (low-frequency pulse impact feature), then increase the scale ; combined with the typical fault frequency of the induced draft fan (such as bearing fault , impeller fault , correct the scale: wherein​ is a sampling frequency, is a frequency correction coefficient, is a rounding function; Multiscale decomposition: perform open-close operation on to decompose into characteristic components and residual component : wherein is an open operation, is a close operation, is a structural element of the characteristic component; S2.3: sensitive feature screening: calculate the fault sensitive index of each component - corrected energy entropy wherein is the energy proportion of the characteristic component in the frequency band, is the kurtosis of the component, and the component with is retained as a fault feature (normal state ); S3: fault mode matching of fused features, dynamically match the screened feature vector with the fault mode library, and calculate the matching degree by using cosine similarity , when , it is determined as the corresponding fault, that is, the induced draft fan operating state and fault type are determined; S4: diagnosis result output and early warning, generate an operating state evaluation report according to the structure obtained in step S3 and trigger a hierarchical early warning.

[0006] Preferably, the non-invasive composite sensor array in step S1 includes a fiber optic acoustic sensor, an infrared thermal imaging array, and a micro differential pressure sensor, which respectively collect the induced draft fan shell vibration sound signal, the bearing temperature field distribution, and the inlet airflow disturbance coefficient.

[0007] Preferably, the step S3 specifically implements the following steps: S3.1, dynamic construction and maintenance of the fault mode library: Initial template construction: collect the full life cycle data of 12 types of typical faults of the induced draft fan, collect not less than 50 groups of samples for each type of fault (covering three stages of slight, moderate, and severe); perform steps S1 and S2 on each group of fault samples to screen out the corresponding sensitive feature components, and construct a standard feature template vector for each type of fault , that is, the fault mode library is obtained, wherein is a fault type number, is a feature dimension, which is determined by the number of sensitive features screened out by S2 and is fixed as 10 dimensions to ensure uniformity of the dimensions; the first the first the normalized mean value of the first S3.2: From the output result of step S2, the reserved sensitive feature components, i.e., satisfying , are extracted to construct the original to-be-matched feature vector , where is the normalized energy entropy value of the first dimensional sensitive feature, consistent with the dimension of the fault mode library template vector; for the fluctuation of the feature amplitude under different operating conditions, the second standardization processing is performed on to eliminate the influence of the operating condition difference: , where are the minimum value and the maximum value of the original to-be-matched vector respectively, and the normalized to-be-matched vector is obtained after the standardization: ; S3.3: Cosine similarity calculation: the cosine similarity is used to measure the direction consistency of the to-be-matched vector and the fault template vector of a certain class in the mode library , and the value is closer to 1, indicating that the feature matching degree is higher. The calculation formula is as follows: , where is the matching degree of the to-be-matched vector and the fault template of the first class, is the dot product of the two vectors; are the L2 norms of the two vectors respectively; is the first dimensional component of the normalized to-be-matched vector; is the first dimensional component of the fault template vector of the first class; The template vectors of 12 classes of faults in the mode library are traversed, the matching degrees of the to-be-matched vector and each class of faults are calculated, and the maximum matching degree and the corresponding fault type are recorded; S3.4: Dynamic matching verification and fault determination: Time sequence consistency verification: the fault feature of the induced draft fan has continuity, and the high matching degree of single sampling is caused by accidental interference, therefore, the to-be-matched vectors corresponding to the sampling frequency of step S1 are continuously collected, and the total time length is 0.5 s, the matching degrees of each vector and the fault of the first class are calculated; if the continuous sampling points satisfy , the fault is determined to be the first class.If yes, go to next step; if no, it is accidental interference, output "no fault at present". Adaptive threshold adjustment: According to the difference of feature recognition degree of different fault types, a threshold correction coefficient is introduced ; for the fault with high feature recognition degree (such as surge, the feature is low frequency strong pulsation, ), the correction threshold is ; for the fault with low feature recognition degree (such as slight impeller dust accumulation, ), the correction threshold is ; the correction threshold is obtained by the similarity distribution statistics of the samples of this type of fault in the mode library (ensure that 95% of the real fault samples can meet the correction threshold); Final judgment: if the corrected matching degree , that is, 3 or more consecutive sampling points, it is finally judged that the induced draft fan has this type of fault; if the maximum matching degree of all fault types , it is judged as "no clear fault at present, but need to be monitored continuously" (may be an extremely early fault, the feature is not fully manifested); S3.5: matching result output and confidence calculation: output the fault judgment result: including fault type (such as "impeller 3 quadrant wear"), maximum matching degree , the number of sampling points that continuously meet the threshold ; then calculate the fault judgment confidence , which will be the basis for step S4 graded early warning.

[0008] Preferably, the 12 typical faults in step S3.1 include: impeller wear, bearing overheating, surge, blade crack, bearing seat loosening, impeller dust accumulation, motor imbalance, coupling misalignment, seal leakage, air duct blockage, main shaft bending, and electrical fault.

[0009] Preferably, a dynamic updating mechanism is set in the fault mode library obtained in step S3.1, specifically: every 100 groups of actual operation data are accumulated, wherein the actual operation data contain normal state and diagnosed fault state, the fault mode library is iteratively updated; specifically, if the cosine similarity between the newly collected fault sample and the existing template vector is (determined as a new fault variant), it is used as a supplementary template of this type of fault, and the average feature vector of this type of fault is recalculated , wherein is the number of existing templates of this type of fault, is the feature vector of the new fault sample, which ensures that the fault mode library adapts to the feature drift caused by equipment aging and working condition change.

[0010] Preferably, the specific working logic of step S4 is as follows: S4.1: From the output results of step S3, extract core information: fault type (e.g. "impeller wear" "bearing overheating"), decision confidence (e.g. 0.736), sensitive feature source channel, i.e. the sensor number corresponding to the sensitive feature screened in step S2; then based on the sensor space mapping relationship of step S1, realize accurate anchoring of fault location; Integrate output information: integrate fault type, confidence, and accurate location into standardized diagnostic results: "fault type: ; fault location: positioning result; confidence: " (example: "fault type: impeller wear; fault location: impeller 3 quadrant; confidence: 92%"). S4.2: Two-dimensional classification of fault level: Classification dimension definition: Dimension 1: decision confidence ( ), reflecting the reliability of the diagnostic results; Dimension 2: inherent risk level of the fault , where is a preset value, based on the impact of the fault on the operation of the induced draft fan, divided into 3 levels: High risk: surge, blade crack, main shaft bending, electrical fault (may cause equipment shutdown or safety accident); Medium risk: bearing overheating, coupling misalignment, motor imbalance (long-term operation leading to fault expansion, affecting unit efficiency); Low risk: impeller dust accumulation, slight sealing leakage, slight airway blockage (short-term no serious impact, can run with fault to maintenance window); Then combine dimension 1 and dimension 2 to classify the fault into 4 levels of early warning: level 1 early warning, level 2 early warning, level 3 early warning, and level 4 early warning; S4.3: Execution and linkage control of classified early warning: Early warning hardware linkage: trigger the corresponding preset hardware action according to the fault level, to ensure the timeliness and effectiveness of the early warning; Early warning release conditions: after manual confirmation of maintenance completion, send a "maintenance completed" instruction through the operation and maintenance system; continuously collect 10 sampling points (total time 1s), if the fault matching degree of all sampling points is less than 0.5 , and the confidence is greater than 0.9, then automatically release the early warning; if the fault feature is not eliminated after maintenance , then re-execute the S2-S4 process to update the diagnostic results and early warning level;S4.4: Result record: Store the standardized diagnosis result, warning level, trigger time, release time (if any), sensor raw data segment (10s before and after the fault) into the distributed database, and archive them according to the "equipment number-date-fault type" for subsequent fault tracing and pattern library updating.

[0011] Preferably, the accurate anchoring of the fault position in step S4.1 specifically includes the following contents: If the sensitive feature comes from the acoustic sensor: 6-way acoustic sensors are arranged in the circumferential 1-6 quadrant parts of the induced fan impeller (1 quadrant corresponds to 0°-60°, 2 quadrant 60°-120°, and so on), and the sensor number Directly map the impeller quadrant That is, "acoustic sensor No. 3" corresponds to "impeller No. 3 quadrant"; If the sensitive feature comes from the infrared thermal imaging array: according to the temperature field matrix The coordinates corresponding to the medium-high energy entropy feature , combined with the physical size of the bearing seat, that is, the preset bearing seat center coordinates ), calculate the fault offset 、 , locate the specific fault point; If the sensitive feature comes from the airflow disturbance sensor: 3 groups of sensors correspond to the 0°, 120°, and 240° orientations of the inlet pipe, and the sensor number corresponds to the pipe orientation, locating "inlet pipe 120° orientation airflow disturbance anomaly" (such as airway blockage, sealing leakage).

[0012] Compared with the prior art, the beneficial effects of the present application are: multi-dimensional perception improves data effectiveness: the present application collects data through "acoustic-thermal-airflow" multi-modal sensor array, breaks through the limitations of traditional single parameter monitoring, and at the same time, through space-time alignment processing (synchronization error <1ms), ensures the integrity and timeliness of the data. Compared with traditional vibration monitoring, the newly added temperature field gradient, airflow disturbance coefficient and other dimensions make the fault precursor capture more comprehensive, and the data utilization rate is improved by more than 60%, providing a reliable basis for subsequent diagnosis.

[0013] Self-adaptive feature extraction strengthens fault recognition: the improved AMMD algorithm dynamically adjusts the structure element size through kurtosis and fault frequency, solving the problem that the traditional fixed size algorithm cannot separate the wide frequency fault feature. The feature separation accuracy of different types of faults such as blade crack (high frequency impact), surge (low frequency pulsation) is improved, the early weak fault (such as slight impeller dust accumulation, bearing seat loosening) recognition accuracy is improved, and the fault expansion caused by missed diagnosis is effectively avoided.

[0014] Accurate matching and positioning reduce diagnostic errors: The dynamically updated fault mode library adapts to the feature drift caused by equipment aging and working condition changes, combined with cosine similarity calculation and time sequence consistency verification (confirmed by 3 consecutive sampling points), so that the fault misjudgment rate is reduced to below 5%. At the same time, based on the spatial mapping relationship of the sensor, the fault is accurately positioned (such as impeller quadrant, bearing seat coordinates, pipeline direction), which provides clear targeting for operation and maintenance, and shortens the fault troubleshooting time.

[0015] Hierarchical early warning and operation and maintenance linkage improve management efficiency: Adopting "confidence + inherent risk" two-dimensional hierarchical early warning, realizing differentiated intervention (emergency shutdown, limited maintenance, regular troubleshooting), avoiding over-response or delayed processing.

[0016] Intelligent operation and maintenance promote industrial upgrading: The present application integrates multi-physical field data fusion, adaptive algorithm optimization, intelligent matching and operation and maintenance linkage technology, breaks away from the traditional dependence on manual experience monitoring mode, and promotes the intelligent transformation of the induced draft fan operation and maintenance of the thermal power plant to "data driven + accurate intervention". BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is the overall method flowchart of the present application; Figure 2 is the multi-modal heterogeneous data acquisition and space-time alignment workflow diagram of the present application; Figure 3 is the adaptive decomposition and sensitive feature screening workflow diagram of the collected data of the present application; Figure 4 is the workflow diagram of determining the operation state and fault type of the induced draft fan of the present application; Figure 5 is the diagnostic result output and early warning workflow diagram of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] Please refer to Figures 1-5 The present application provides a technical solution: a thermal power plant boiler induced draft fan operation state intelligent monitoring and diagnosis method, comprising the following steps: S1: Multimodal heterogeneous data acquisition and spatio-temporal alignment, acoustic, thermal and airflow disturbance signals during the operation of the induced draft fan are collected by a non-intrusive composite sensor array, and spatio-temporal dimension correlation is performed to obtain the collected data, wherein the non-intrusive composite sensor array includes a fiber optic acoustic sensor, an infrared thermal imaging array and a micro differential pressure sensor, which respectively collect the shell vibration acoustic signal of the induced draft fan, the bearing temperature field distribution and the inlet airflow disturbance coefficient; the specific steps are as follows: S1.1: Acoustic signal: 6-channel fiber optic acoustic sensor (sampling frequency 10 kHz) is arranged circumferentially on the shell of the induced draft fan to collect broadband vibration acoustic signal , wherein is a timestamp, , including impeller rotating noise (50-500 Hz) and fault impact sound (1-5 kHz); temperature field distribution: an infrared thermal imaging array (resolution 320x240) is used to shoot the bearing seat area to generate a temperature field matrix , wherein is a spatial coordinate with unit mm, and the sampling period is 1s; airflow disturbance: 3 groups of micro differential pressure sensors (interval 120°) are installed at the inlet pipe of the induced draft fan to calculate the airflow disturbance coefficient , wherein is the differential pressure standard deviation, is the average differential pressure, reflecting the airflow turbulence intensity; S1.2: Time alignment: taking the pulse signal of the main shaft encoder of the induced draft fan as the reference clock (synchronization error <1ms), the acoustic signal (10 kHz), the temperature field (1 Hz) and the airflow disturbance (10 Hz) are unified to the same time axis ; S1.3: Spatial correlation: the mapping relationship between the sensor position and the structure of the induced draft fan is established, including the correspondence between the th acoustic sensor and the th quadrant area of the impeller, the acoustic signal is bound with the temperature of the area to generate a spatio-temporal fusion matrix: , wherein is the 3 groups of airflow disturbance coefficients; S2: Fault feature extraction based on improved adaptive multi-scale morphological decomposition, adaptive decomposition and sensitive feature screening are performed on the collected data; the specific steps are as follows: S2.1: Signal preprocessing: single-channel acoustic signal (focusing on fault sensitive channels) is extracted from the fusion matrix , and direct current removal and normalization processing are performed: , wherein is the mean, Standard deviation; S2.2: Improved AMMD algorithm decomposition: Initialization: Set minimum scale Maximum scale The unit is: sampling points, and the initial shape of the structuring element is a flat cap shape. ; Calculate the kurtosis of the current signal segment ,in For the expectation operator, if (High-frequency impact characteristics), then reduce the scale. ;like (Low-frequency pulse impact characteristics), then increase the scale Combined with the typical failure frequency of induced draft fans (such as bearing failure) impeller failure Correction scale: ,in Sampling frequency, This is the frequency correction factor. It is a rounding function; Multiscale decomposition: for Perform opening-closing operations, decompose into each feature component and residual components : ,in For opening operation, For closing operations, For the first The structuring element of the characteristic component; S2.3: Sensitive Feature Screening: Calculate the fault sensitivity index of each component—corrected energy entropy. , in For the first The first characteristic component Energy percentage of each frequency band Preserve the kurtosis of this component. The component is used as a fault characteristic (normal state) ); S3: Fault mode matching with fused features. The filtered feature vectors are dynamically matched with the fault mode database, and the matching degree is calculated using cosine similarity. ,when The fault is identified as a corresponding fault, that is, the operating status and fault type of the induced draft fan are determined; the specific implementation steps of step S3 are as follows: S3.1 Dynamic construction and maintenance of the fault mode library: Initial template construction: collect the full life cycle data of typical faults of the induced draft fan 12, including: impeller wear, bearing overheating, surge, blade crack, bearing seat loosening, impeller dust accumulation, motor imbalance, coupling misalignment, seal leakage, air passage blockage, main shaft bending, electrical fault, and collect not less than 50 groups of samples for each type of fault (covering slight, moderate and severe stages); perform steps S1 and S2 on each fault sample to screen out the corresponding sensitive feature components, and construct a standard feature template vector for each type of fault , that is, a fault mode library is obtained, wherein is the fault type number, is the feature dimension, which is determined by the number of sensitive features screened by S2 and is fixed at 10 dimensions to ensure uniformity of the dimensions; is the normalized mean value of the th feature of the th fault; wherein a dynamic updating mechanism is provided in the fault mode library, specifically: every 100 groups of actual operation data are accumulated, wherein the actual operation data include normal states and diagnosed fault states, and the fault mode library is iteratively updated; specifically, if the cosine similarity between the newly collected fault sample and the existing template vector is (determined as a new fault variant), it is taken as a supplementary template for this type of fault, and the average feature vector of this type of fault is recalculated , wherein is the number of existing templates for this type of fault, is the feature vector of the new fault sample, to ensure that the fault mode library adapts to the feature drift caused by equipment aging and working condition changes S3.2: Extract the retained sensitive feature components from the output results of step S2, that is, satisfy , to construct the original matching feature vector , wherein is the normalized energy entropy value of the th sensitive feature, , and the dimensions of the fault mode library template vector are consistent; for the fluctuations in feature amplitude under different operating conditions, the is subjected to secondary standardization processing to eliminate the influence of operating condition differences: , wherein are the minimum and maximum values of the original matching vector, respectively, and the standardized , to obtain the standardized matching vector ; S3.3: Cosine similarity calculation: the cosine similarity is used to measure the direction consistency between the matching vector and the template vector of a certain type of fault in the mode library , and the value is closer to 1, indicating a higher feature matching degree. The calculation formula is as follows: , wherein For the vector to be matched and the first Matching degree of the fault template, It is the dot product of two vectors; Let L2 norms be the L2 norms of the two vectors, respectively. For the standardized matching vector, the first... Dimensional components; For the first The first class of fault template vectors Dimensional components; Traverse the template vectors of 12 fault types in the pattern library and calculate the matching degree between the vector to be matched and each fault type. Record the maximum matching degree and corresponding fault types ; S3.4: Dynamic Matching Verification and Fault Judgment: Timing consistency verification: The fault characteristics of the induced draft fan are continuous. The high matching degree of a single sampling may be caused by random interference. Therefore, continuous sampling is necessary. Each timestamp corresponds to the sampling frequency in step S1, and the vector to be matched has a total duration of 0.5 seconds. Calculate the relationship between each vector and... Matching degree of class of faults If continuous Each sampling point satisfies If the condition is met, proceed to the next step; if there are fewer than 3 sampling points that meet the requirements, it is determined to be accidental interference, and "No fault currently" is output. Adaptive threshold adjustment: To address the differences in feature recognition for different fault types, a threshold correction coefficient is introduced. For faults with high identifiable characteristics (such as surge, characterized by low-frequency strong pulsations), ), correct the threshold to For faults with low feature recognition (such as slight impeller dust accumulation), ), correct the threshold to The correction threshold is obtained by statistically analyzing the similarity distribution of this type of fault samples in the pattern library (ensuring that 95% of real fault samples can meet the correction threshold). Final determination: If the corrected matching degree If three or more consecutive sampling points are detected, the induced draft fan is ultimately determined to be faulty. Class of faults; if the maximum matching degree of all fault types If so, it is determined as "there is no clear fault at present, but continuous monitoring is required" (it may be a very early fault, and the characteristics have not been fully manifested). S3.5: Matching Result Output and Confidence Calculation: Output fault determination results, including fault type (e.g., "impeller 3-quadrant wear"), maximum matching degree. , the number of sampling points continuously satisfying the threshold ; then calculate the fault determination confidence , which will be the basis for the step S4 graded warning.

[0020] S4: Diagnostic result output and warning, generate a running state evaluation report according to the structure obtained in step S3 and trigger a graded warning The specific working logic of step S4 is as follows: S4.1: From the output results of step S3, extract core information: fault type (such as "impeller wear" "bearing overheating"), determination confidence (such as 0.736), and sensitive feature source channel, that is, the sensor number corresponding to the sensitive feature screened in step S2; then, based on the sensor space mapping relationship of step S1, realize accurate anchoring of the fault location, including the following contents: If the sensitive feature comes from the acoustic sensor: 6 acoustic sensors are deployed according to the circumference of the induced fan impeller 1-6 quadrant (1 quadrant corresponds to 0°-60°, 2 quadrant 60°-120°, and so on), and the sensor number directly maps the impeller quadrant , that is, "acoustic sensor No. 3" corresponds to "impeller 3 quadrant"; If the sensitive feature comes from the infrared thermal imaging array: according to the coordinates corresponding to the high-energy entropy feature in the temperature field matrix , combined with the physical size of the bearing seat, that is, the preset bearing seat center coordinates ), calculate the fault offset , , locate the specific fault point; If the sensitive feature comes from the airflow disturbance sensor: 3 groups of sensors correspond to the inlet pipe 0°, 120°, 240° direction, and the sensor number corresponds to the pipe direction, locating "inlet pipe 120° direction airflow disturbance anomaly" (such as airway blockage, sealing leakage); Integrate the output information: integrate the fault type, confidence, and accurate location into the standardized diagnostic result: "fault type: ; fault location: location result; confidence: " (example: "fault type: impeller wear; fault location: impeller 3 quadrant; confidence: 92%"); S4.2: Two-dimensional division of fault level: Division dimension definition: Dimension 1: determination confidence ( ), reflecting the reliability of the diagnostic result; Dimension 2: fault inherent risk level , wherein For preset values, based on the impact of the fault on the operation of the induced draft fan, it is divided into 3 levels: High risk: surge, blade crack, main shaft bending, electrical fault (may cause equipment downtime or safety accident); Medium risk: bearing overheating, coupling misalignment, motor imbalance (long-term operation leading to fault expansion, affecting unit efficiency); Low risk: blade dust accumulation, slight seal leakage, slight airway blockage (short-term no serious impact, can run with fault to maintenance window); Then combine dimension 1 and dimension 2, divide the fault into 4 levels of early warning: first level early warning, second level early warning, third level early warning and fourth level early warning; The following is a specific division table: Grade definition: First level early warning: emergency failure, need to stop immediately; Second level early warning: important failure, arrange maintenance within 1 hour; Third level early warning: general failure, check within 24 hours; Fourth level early warning: slight failure, handle next time during regular maintenance; Continuous monitoring: very early fault characteristics, update diagnosis results every 5 minutes until the characteristics are enhanced or disappear.

[0021] S4.3: Hierarchical early warning execution and linkage control: Early warning hardware linkage: trigger the corresponding preset hardware action according to the fault level, ensure the timeliness and effectiveness of early warning; For example: First level early warning (such as "blade crack, confidence 0.95"): Start sound and light alarm (red warning light always on + high-frequency buzzer, volume ≥ 85dB); Send stop command to induced draft fan PLC control system (such as command format: "STOP_REQUEST=1"), trigger emergency stop process (according to unit safety regulations, first reduce induced draft fan load to 0, then cut off power); Send SMS + APP push notification to operation and maintenance center through industrial Ethernet (including fault information, positioning and emergency treatment suggestions).

[0022] Second level early warning (such as "bearing overheating, confidence 0.88"): Start sound and light alarm (yellow warning light flickering + medium frequency buzzer, volume ≥ 75dB); Send load limit command to PLC ("LOAD_LIMIT=50%"), avoid fault deterioration; The operation and maintenance center APP pushes a notification, and a maintenance work order is generated synchronously (including fault location and required tool list).

[0023] Level 3 warning (e.g., "dusty impeller, confidence 0.82"): Green warning light flashes, no buzzer alarm; The operation and maintenance system generates a to-do reminder and displays it on the monitoring interface in the central control room. Level 4 warning (e.g., "slight sealing leakage, confidence 0.75"): No hardware alarm, only mark fault information on the monitoring interface; Incorporate into the regular maintenance plan and automatically associate the next maintenance time.

[0024] Warning removal conditions: After manual confirmation of maintenance completion, send a "maintenance complete" instruction through the operation and maintenance system; continuously collect 10 sampling points (total time 1s), if all sampling points have a fault matching degree , and a confidence , the warning is automatically removed; if the fault characteristics are not eliminated after maintenance , the S2-S4 process is re-executed, and the diagnosis result and warning level are updated; S4.4: Result recording: store the standardized diagnosis result, warning level, trigger time, removal time (if any), sensor raw data segment (10s before and after the fault occurs) to the distributed database, and archive it according to "equipment number-date-fault type" for subsequent fault tracing and pattern library updating.

[0025] As an extension of the present application, the operation and maintenance structure is linked and fed back as follows: Automatically synchronize fault information to the CMMS (equipment management system) of the thermal power plant, and associate it with the equipment archives of the induced draft fan (such as running time and historical maintenance records); Based on historical data statistics, generate "fault-maintenance" association suggestions for similar faults (e.g., "impeller 3-quadrant wear, historical maintenance solution: replace impeller blades, average maintenance time 2 hours"); If the same fault occurs ≥2 times within 3 months, automatically trigger a pattern library optimization request and prompt the operation and maintenance personnel to supplement fault sample data (which can be used for iterative updating of the S3 pattern library).

[0026] The application discloses an intelligent monitoring and diagnosing method for the operation state of a boiler induced draft fan in a thermal power plant, and relates to the technical field of power equipment monitoring, aiming to solve the defects of single parameter, confused features and lagging diagnosis of traditional monitoring methods. The method comprises the following steps: S1, multi-modal heterogeneous data acquisition and space-time alignment, collecting acoustic, temperature field and airflow disturbance signals through a fiber optic acoustic sensor, an infrared thermal imaging array and a micro differential pressure sensor, and correlating them to generate a space-time fusion matrix (synchronization error <1ms); S2, extracting fault features based on improved adaptive multi-scale morphological decomposition (AMMD), dynamically adjusting the size of the structure element through kurtosis and fault frequency, separating wideband fault features and selecting sensitive components (retaining components with a corrected energy entropy >0.6); S3, matching the fused features with a dynamically updated fault mode library, combining cosine similarity calculation and time sequence consistency verification to accurately determine the fault type (misjudgment rate <5%); and S4, based on two-dimensional grading early warning of "confidence + inherent risk", linking hardware intervention and operation and maintenance systems to realize targeted fault processing. The application improves the early weak fault recognition accuracy from <60% to more than 60%, prolongs the average fault-free operation time of the induced draft fan, reduces the loss of unplanned shutdown, and is suitable for intelligent operation and maintenance of key equipment in thermal power plants, taking into account safety and economy.

[0027] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring and diagnosis of the operating state of a boiler induced draft fan in a thermal power plant, characterized in that, Comprising the following steps: S1: Multimodal heterogeneous data acquisition and spatio-temporal alignment, collecting acoustic, thermal and airflow disturbance signals during the operation of the induced draft fan through a non-invasive composite sensor array, and correlating in the time and space dimensions to obtain the collected data; The specific steps are as follows: S1.1: Multimodal signal acquisition: Acoustic signal: 6-channel fiber optic acoustic sensor is deployed circumferentially on the shell of the induced draft fan to collect broadband vibration acoustic signals wherein is a timestamp, including impeller rotating noise and fault impact sound; Temperature field distribution: An infrared thermal imaging array is used to shoot the bearing seat area to generate a temperature field matrix wherein is a spatial coordinate, with a unit of mm and a sampling period of 1 s; Airflow disturbance: Three groups of micro differential pressure sensors (with an interval of 120°) are installed on the inlet pipeline of the induced draft fan to calculate the airflow disturbance coefficient wherein is a differential pressure standard deviation, is an average differential pressure, reflecting the turbulence intensity of the airflow; S1.2: Time alignment: Align the acoustic signal, temperature field, and airflow disturbance to the same time axis with the pulse signal of the main shaft encoder of the induced draft fan as the reference clock ; S1.3: Space correlation: Establish the mapping relationship between the sensor position and the structure of the fan, including the first Corresponding to the impeller Quadrant area, the acoustic signal Binding with the temperature of the region, generate a spatiotemporal fusion matrix: wherein, is the 3rd group of airflow disturbance coefficients; S2: Fault feature extraction based on improved adaptive multi-scale morphological decomposition, adaptive decomposition and sensitive feature screening of the collected data; The specific steps are as follows: S2.1: Signal pre-processing: Extracting single-channel acoustic signal from the fused matrix , performing DC removal and normalization: where, is the mean, is the standard deviation;​ S2.2: Improved AMMD algorithm decomposition: Initialization: Set the minimum scale , the maximum scale , the unit: sampling points, the initial shape of the structure element is a flat hat ; calculating the kurtosis of the current signal segment wherein is the expectation operator, if then the scale is reduced ; if then the scale is increased ; in combination with the typical failure frequency of the induced draft fan , the scale is corrected: wherein is the sampling frequency, is the frequency correction coefficient, is the rounding function; Multiscale decomposition: The image is decomposed into where is the opening operator, is the closing operator, and is the structuring element of the th feature component.​​​​​​​​ S2.3: sensitive feature screening: calculate the fault sensitive index of each component - modified energy entropy , wherein is the energy proportion of the first frequency band of the first feature component, is the kurtosis of the component, and the component with is reserved as the fault feature; S3: fusion feature mode matching, the filtered feature vector is dynamically matched with the failure mode library, and the cosine similarity is used to calculate the matching degree When corresponding failure is determined, that is, the induced draft fan operating state and failure type are determined; S4: Diagnosis result output and early warning, generating an operating state evaluation report according to the structure obtained in step S3 and triggering a hierarchical early warning.

2. The intelligent monitoring and diagnosing method for the operation state of the induced draft fan of the boiler of the thermal power plant according to claim 1, characterized in that: The non-invasive composite sensor array in step S1 includes a fiber optic acoustic sensor, an infrared thermal imaging array, and a micro pressure difference sensor, which respectively collect the induced draft fan shell vibration sound signal, bearing temperature field distribution, and inlet airflow disturbance coefficient.

3. The method for intelligent monitoring and diagnosing the operation state of a boiler induced draft fan of a thermal power plant according to claim 1, characterized in that: The specific implementation steps of step S3 are as follows: S3.1, Dynamic construction and maintenance of fault mode library: Initial template construction: collect the whole life cycle data of typical faults of the induced draft fan 12, and collect not less than 50 groups of samples for each type of fault; perform steps S1 and S2 on each fault sample to screen out the corresponding sensitive feature components, and construct the standard feature template vector of each type of fault , that is, a fault mode library is obtained, wherein is the fault type number, is the feature dimension, which is determined by the number of sensitive features screened by S2 and is fixed to 10 dimensions to ensure uniformity of the dimensions; is the normalized mean value of the th feature of the th fault of the th type of fault; S3.2: Extract the retained sensitive feature components from the output of step S2, i.e., those that satisfy... Construct the original feature vector to be matched ,in For the first The normalized energy entropy value of the dimensional sensitive feature. The dimension is consistent with the template vector of the fault mode library; for the fluctuations in feature amplitude under different operating conditions, the following applies: Secondary standardization is performed to eliminate the impact of differences in operating conditions: ,in These are the minimum and maximum values ​​of the original vector to be matched, respectively, after standardization. This yields the standardized vector to be matched. ; S3.3: Cosine similarity calculation: cosine similarity is used to measure the direction consistency of the to-be-matched vector with the fault template vector of a certain type of fault in the mode library The value is closer to 1, indicating that the feature matching degree is higher. The calculation formula is as follows: Wherein is the matching degree of the to-be-matched vector and the first class fault template, is the dot product of the two vectors; is the L2 norm of the two vectors respectively; is the first dimension component of the normalized to-be-matched vector; is the first dimension component of the first class fault template vector; Traverse the template vectors of 12 types of faults in the mode library, calculate the matching degree of the to-be-matched vector and each type of fault , record the maximum matching degree and the corresponding fault type ; S3.4: Dynamic matching verification and fault judgment: Time sequence consistency verification: the fault feature of the induced draft fan has continuity, and the high matching degree of single sampling is caused by accidental interference. Therefore, a timestamp is continuously collected, that is, the sampling frequency corresponding to step S1, and the total duration of the to-be-matched vector is 0.5 s . The matching degree of each vector to the same fault is calculated respectively ;​ If consecutive sampling points satisfy , then go to the next step; if less than 3 sampling points satisfy, then determine as accidental interference, output "no fault at present"; Adaptive threshold adjustment: introduce threshold correction coefficient for feature recognition difference of different fault types ; For the fault with high feature recognition degree, the correction threshold is ; For a fault with low feature recognition degree, the correction threshold is ; The correction threshold is obtained by statistical analysis of the similarity distribution of the fault samples in the mode library (ensure that 95% of the real fault samples can meet the correction threshold); Final decision: if the matching degree after correction i.e. 3 consecutive sampling points or more, then the final decision is that the induced draft fan exists Class fault; if the maximum matching degree of all fault types then it is determined that "there is no clear fault at present, but it needs to be continuously monitored"; S3.5: Match result output and confidence calculation: output fault determination result: including fault type, maximum matching degree , number of sampling points continuously satisfying threshold value ; then calculate fault determination confidence , which will be the basis for step S4 graded early warning.

4. The method for intelligent monitoring and diagnosing the operation state of a boiler induced draft fan of a thermal power plant according to claim 3, characterized in that: The 12 typical faults in step S3.1 include: impeller wear, bearing overheating, surge, blade crack, bearing seat loosening, impeller dust accumulation, motor imbalance, coupling misalignment, seal leakage, air duct blockage, main shaft bending, and electrical fault.

5. The method for intelligent monitoring and diagnosing the operation state of a boiler induced draft fan of a thermal power plant according to claim 3, characterized in that: The dynamic updating mechanism is set in the fault mode library obtained in the step S3.1, specifically: every 100 groups of actual operation data are accumulated, wherein the actual operation data contain normal states and diagnosed fault states, and the fault mode library is iteratively updated; specifically, if the cosine similarity of a newly collected fault sample and an existing template vector is less than a preset threshold value (determined as a new fault variant), the fault sample is taken as a supplementary template of the fault, and the average feature vector of the fault is recalculated wherein is the number of existing templates of the fault, is the feature vector of the new fault sample, to ensure that the fault mode library adapts to the feature drift caused by equipment aging and working condition changes.​ 6. The method for intelligent monitoring and diagnosing the operation state of a boiler induced draft fan of a thermal power plant according to claim 1, characterized in that: The specific working logic of step S4 is as follows: S4.1: From the output results of step S3, extract core information: fault type , decision confidence , sensitive feature source channel, i.e. the sensor number corresponding to the sensitive feature screened in step S2; then based on the sensor space mapping relationship of step S1, realize accurate anchoring of the fault location; Integrate output information: integrate fault type, confidence, and accurate position into standardized diagnosis results: "fault type: m_max; fault location: positioning result; Confidence: C x 100%; S4.2: Two-dimensional classification of fault level: Dimension definition: Dimension 1: decision confidence , reflecting the reliability of the diagnosis result; Dimension 2: fault inherent risk level , wherein is a preset value, based on the influence degree of the fault on the operation of the induced draft fan, divided into 3 levels: High risk: Surge, blade cracks, main shaft bending, electrical failure; Medium risk: Bearing overheating, coupling misalignment, motor imbalance; Low risk: Impeller dust, slight seal leakage, slight airway blockage; Then, combined with dimension 1 and dimension 2, the fault is divided into four levels of early warning: first level early warning, second level early warning, third level early warning, and fourth level early warning; S4.3: Hierarchical early warning execution and linkage control: Early warning hardware linkage: trigger the corresponding preset hardware action according to the fault level to ensure the timeliness and effectiveness of the early warning; Pre-alarm release condition: After manual confirmation of maintenance completion, send "maintenance completion" instruction through the operation and maintenance system; continuously collect 10 sampling points, if the fault matching degree of all sampling points and the confidence are all 0, then automatically release the pre-alarm; if the fault characteristics are not eliminated after maintenance , then re-execute the S2-S4 process, update the diagnosis result and pre-alarm level; S4.4: Result recording: store the standardized diagnosis results, early warning level, trigger time, release time, and sensor raw data segment to a distributed database, and archive them according to "equipment number-date-fault type" for subsequent fault tracing and mode library updating.

7. The method for intelligent monitoring and diagnosing the operation state of a boiler induced draft fan of a thermal power plant according to claim 6, characterized in that: The accurate anchoring of the fault location in step S4.1 specifically includes the following contents: If the sensitive feature comes from acoustic sensor: 6 acoustic sensors are deployed in the 1-6 quadrants of the fan impeller circumference, sensor number Direct mapping of impeller quadrant ; If the sensitive feature comes from the infrared thermal imaging array: according to the temperature field matrix The coordinates corresponding to the medium-high energy entropy feature , combined with the physical size of the bearing seat, that is, the preset bearing seat center coordinates ), calculate the fault offset 、 , locate the specific fault point; If the sensitive feature comes from the airflow disturbance sensor: 3 groups of sensors correspond to the inlet pipe 0°, 120°, and 240° orientation, and the sensor number corresponds to the pipe orientation.

Images