Sound-based methods and systems for monitoring pipe blockage and damage.

Abstract

This invention discloses a sound-based method and system for monitoring pipe blockage and damage. It relates to the field of intelligent acoustic signature monitoring technology for industrial pipelines, integrating acoustic signature recognition and artificial intelligence technologies to achieve real-time perception of pipeline health status. The system includes a multi-source data real-time acquisition module, a data processing module, a calculation and management module, and a handling and analysis module. The invention sets up three units: acoustic signature characteristics, fault type, and comprehensive risk. The acoustic signature characteristics unit outputs a stable anomaly index through multiple correction parameters and global balanced weights, reducing false alarm probability and maintenance costs from the source. The fault type unit directly outputs a clear fault type based on historical fault characteristics, physical signal features, and multi-source data, eliminating the need for manual judgment. The comprehensive risk unit integrates parameters such as media hazards and environmental attributes to synchronously output the risk level and investigation scope, allowing maintenance personnel to directly carry out handling work.

Description

Technical Field

[0001] This invention relates to the field of intelligent acoustic monitoring technology for industrial pipelines, specifically to a method and system for monitoring pipeline blockage and damage based on sound. Background Technology

[0002] In industrial sectors such as mining, long-distance transportation of slurry generally relies on pipeline transportation. This method has become the mainstream choice due to its high transmission efficiency and small footprint. However, during long-term operation, pipelines are easily affected by multiple factors such as the characteristics of the transported medium, pressure fluctuations, environmental erosion, and construction quality, making them prone to two core safety hazards: blockage and leakage. When a pipeline breaks and leaks, it not only wastes a large amount of slurry materials and increases production and operating costs, but the leaked slurry also causes serious pollution to the surrounding soil, water and other ecological environments, posing environmental risks. If signs of blockage appear in the pipeline and are not detected and dealt with in time, the blockage will expand rapidly, eventually leading to a complete pipeline blockage accident, paralyzing the conveying system and seriously affecting the continuity of production. More importantly, if pipeline leaks and blockages cannot be identified in their early stages, the accident situation will continue to escalate, potentially triggering a chain reaction such as pipeline rupture and equipment damage, and even threatening the safety of on-site personnel, causing incalculable economic losses and safety consequences.

[0003] Currently, pipeline blockage and leakage monitoring may rely heavily on manual inspections along the pipeline or single-point detection by pressure and flow sensors. The former has the disadvantages of poor monitoring timeliness, difficulty in detecting faults in hidden pipe sections, and high labor costs, while the latter relies on the sudden change in fluid parameters caused by the development of the fault to a certain extent, which has obvious response lag and makes it difficult to identify early minor faults. Existing acoustic feature-based monitoring methods may rely on direct comparison between the original acoustic signature and a fixed reference acoustic signature for anomaly detection. They do not introduce correction mechanisms adapted to the pipeline's service status and real-time operating conditions, nor do they perform targeted signal enhancement processing on the characteristic frequency bands corresponding to pipe blockage and leakage. Therefore, they may be misjudged as faults. Existing acoustic feature-based monitoring methods, after identifying abnormal signals, mostly distinguish fault types based solely on the spectral characteristics of the acoustic signal itself, without integrating the correlation characteristics of heterogeneous data such as pressure changes and pump operation status. They lack multi-modal feature matching and verification mechanisms, making it impossible to reliably and accurately distinguish between pipe blockage and leakage. They also lack the multi-dimensional feature fusion capabilities of artificial intelligence technology, making it difficult to meet the needs of acoustic signature recognition monitoring and pipeline health status prediction for long-distance industrial pipelines, thus resulting in poor practicality. Summary of the Invention

[0004] The purpose of this invention is to provide a sound-based method and system for monitoring pipe blockage and pipe damage, which solves the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a sound-based method for monitoring pipe blockage and pipe damage, comprising the following steps: Step S1: Obtain acoustic signature data, fault signature data, and risk correlation data of the pipeline transporting medium through a multi-source real-time data acquisition module; Step S2: Input the voiceprint feature data, fault feature data and risk association data into the data processing module. The data processing module cleans the input data and then inputs the cleaned data into the calculation management module. Step S3: The calculation and management module receives the data output in step S2 and uses the voiceprint feature unit, fault type unit, and comprehensive risk unit to realize the analysis and location of anomaly identification, fault classification, and risk rating. The processing flow of the computing management module is as follows: Step A: Based on the pipe wall wear compensation coefficient, support resonance compensation coefficient, slurry concentration correction coefficient, valve opening linkage correction coefficient, dynamic benchmark adaptive update coefficient, total number of frequency segments of the acoustic signal, and the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, output the acoustic anomaly degree. Step B: Based on the fault history weight coefficient, pump frequency coupling matching degree, signal ratio weight, sound pressure cross-modal matching weight, pressure correlation coefficient, difference between current pressure and reference pressure, and matching degree between current abnormal soundprint and historical fault soundprint in the fault feature data, and combined with soundprint anomaly degree, output the fault type attribution degree. Step C: Based on the risk coefficient of the transport medium, the regional sensitivity coefficient, the duration of the abnormal signal, the matching coefficient of the special pipe section structure, the location confidence correction coefficient, and the time difference of the abnormal acoustic signal arriving at two adjacent sensors in the risk association data, and combined with the acoustic anomaly degree and the fault type attribution degree, output the comprehensive risk value and the location confidence. Step S4: Input the voiceprint anomaly degree, fault type attribution degree, comprehensive risk value and location reliability into the handling analysis module. The handling analysis module classifies and executes risk levels based on the input data.

[0006] Optionally, the processing procedure for the voiceprint feature unit is as follows: Step A1: By combining the measured pipe wall thickness with the initial pipe wall thickness, the wear rate is obtained, and the attenuation deviation of the acoustic signal propagation caused by pipe wall wear is analyzed, thereby obtaining the pipe wall wear compensation coefficient. Step A2: By combining the resonant frequency of the pipeline with the reference resonant frequency, the frequency offset rate is obtained to analyze the inherent resonant offset of the pipeline caused by the loosening of the support and hanger, and then the resonant compensation coefficient of the support and hanger is obtained. Step A3: By combining the slurry concentration and the reference delivery concentration, the concentration deviation rate is obtained to analyze the flow acoustic signal intensity deviation caused by the change in slurry concentration, and then the slurry concentration correction coefficient is obtained. Step A4: By combining the opening degree of the pipeline valve with the reference opening degree, the maximum opening degree deviation rate is obtained to analyze the flow field noise deviation caused by valve opening adjustment, and then the valve opening degree linkage correction coefficient is obtained. Step A5: By inputting the fault-free normal sound signals from the past 7 days into the lightweight adaptive learning model, the dynamic baseline adaptive update coefficients are obtained, and then the matching degree between the current normal voiceprint and the original baseline voiceprint is calculated. Step A6: Combine the total number of frequency segments of the acoustic signal, the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, and the reference power spectral density of the i-th frequency segment when the pipeline is running normally under the same working conditions, and combine the global balance weight, hardware attribute parameter weight, working condition fluctuation parameter weight, AI adaptive parameter weight, and feature weight of the i-th frequency segment to finally output the acoustic anomaly degree. Step A7: If the voiceprint anomaly score is <30, it is determined to be a normal working condition. All collected data will only be cached for a short period of time and will not trigger subsequent processes. The process will directly enter the next collection cycle. If the voiceprint anomaly degree is ≥30, it is determined that there is an abnormal operating condition, triggering the subsequent fault analysis process. The voiceprint anomaly degree value, voiceprint feature data, fault feature data, and risk association data are simultaneously input into the fault type unit and the comprehensive risk unit.

[0007] Optionally, the processing procedure of the fault type unit is as follows: Step B1: By introducing voiceprint anomaly degree as the basic input for fault type determination, the higher the anomaly degree, the higher the probability of fault occurrence. Step B2: By statistically analyzing the number of pipe blockages and leaks in the past three months, the proportion of pipe blockage and leak failures is analyzed to obtain the historical weighting coefficient of the failures. Step B3: Calculate the pump frequency coupling matching degree by analyzing the degree of matching between the abnormal sound signal fluctuation frequency and the operating frequency of the delivery pump; Step B4: Obtain the signal ratio weight by analyzing the intensity ratio of the air-conducted sound signal to the solid-conducted sound signal in the pipe wall; Step B5: By inputting the current acoustic signature spectrum features and the pressure time series features of the past 5 minutes into the Transformer model, the model outputs the matching degree with the standard pipe blockage and leakage failure modes to obtain the sound pressure cross-modal matching weights. Step B6: The pressure correlation coefficient, the difference between the current pressure and the reference pressure, and the matching degree between the current abnormal soundprint and the historical fault soundprint are combined and calculated. The global balance weight, the working condition correlation parameter weight, the signal feature parameter weight and the AI ​​multimodal parameter weight are combined and normalized to finally output the fault type attribution degree. Step B7, based on the fault type attribution degree, specifically obtains the probability of pipe blockage attribution and the probability of leakage attribution; Step B8: If the probability of pipe blockage is greater than the probability of leakage, then the current fault type is determined to be pipe blockage; if the probability of leakage is greater than the probability of pipe blockage, then the current fault type is determined to be leakage.

[0008] Optionally, the processing procedure for the integrated risk unit is as follows: Step C1: Input the maximum value of the two fault types, pipe blockage and leakage, obtained from the fault type unit into the comprehensive risk unit to obtain the maximum fault assignment probability, thereby quantifying the credibility basis of the risk assessment. Step C2: By combining the pH value and heavy metal content of the pipeline transport medium, we can obtain the risk amplification factors caused by the pollution and corrosiveness of the transported slurry, and thus calculate the risk coefficient of the transport medium. Step C3: By connecting to the mine's GIS system, the environmental attributes around the fault point are identified based on the spatial location of the pipeline, and then values ​​are assigned based on the sensitivity of the surrounding environment to obtain the regional sensitivity coefficient. Step C4: By analyzing the duration of the current abnormal sound signal from the first trigger threshold to the current moment, the duration of the abnormal signal is obtained, and combined with the static attribute parameter weights, AI inference parameter weights, inference risk correction coefficients and fault duration coefficients, a comprehensive risk value is output. Step C5: Analyze the mileage coordinates of all special structures in the pipeline to calculate the distance between the fault location and the nearest special structure point, so as to obtain the matching coefficient of the special pipe section structure. Step C6: By inputting the time difference of the abnormal acoustic signal arriving at two adjacent sensors, the number of historical faults in the pipe section, and the signal amplitude of the two adjacent sensors into the Bayesian network model, the location confidence correction coefficient is obtained. Step C7: Combine the time difference of the abnormal sound wave signal reaching the two adjacent sensors, the theoretical time difference of normal sound wave propagation at the same distance, and the maximum theoretical time difference of sound wave propagation between the two sensors, and combine the structural attribute class parameter weights and AI inference class weight coefficients to output the location confidence.

[0009] Optionally, the processing analysis module includes a division unit and an action execution unit.

[0010] Optionally, the partitioning unit specifically comprises: When the overall risk value is less than 0.3, it is considered a low-risk level. When 0.3 ≤ comprehensive risk value < 0.7, it is classified as medium risk. When the comprehensive risk value is ≥0.7, it is classified as a high-risk level; When the location reliability is ≥0.8, the investigation range is 3 meters before and after the fault location point; When 0.5 ≤ location confidence level < 0.8, the investigation range is 10 meters before and after the fault location point; When the location confidence level is less than 0.5, the investigation scope is the entire 200-meter section of pipeline covered by the two adjacent sensors upstream and downstream of the fault location point.

[0011] Optionally, the action execution unit specifically comprises: When the risk level is low, abnormal events are recorded in the system background without pushing alarm information. Maintenance personnel can simply perform routine inspections on the corresponding pipe sections during regular inspection periods. When the risk level is medium, push fault warning messages to the mobile terminal of the operation and maintenance team, simultaneously display the fault type and the scope of investigation, and require the handling to be completed within 24 hours; When the risk level is high, an audible and visual alarm is sent to the dispatch and monitoring room, and a text message is sent to the operation and maintenance manager, requiring them to go to the site immediately for handling.

[0012] Secondly, the present invention also provides a sound-based pipe blockage and pipe damage monitoring system, including a multi-source data real-time acquisition module, a data processing module, a calculation and management module, and a disposal and analysis module; The multi-source data real-time acquisition module: Used to acquire acoustic signature data, fault signature data, and risk correlation data when transporting media through pipelines; The data processing module: It is used to receive voiceprint feature data, fault feature data and risk association data, and clean the received data so that the cleaned data can be input into the calculation management module. The computing management module: It is used to receive voiceprint feature data, fault feature data and risk correlation data output by the data processing module, so as to perform analysis and location of anomaly identification, fault classification and risk rating; The computing management module includes: Voiceprint feature unit, fault type unit, and comprehensive risk unit; The voiceprint feature unit: Based on the pipe wall wear compensation coefficient, support and hanger resonance compensation coefficient, slurry concentration correction coefficient, valve opening linkage correction coefficient, dynamic reference adaptive update coefficient, the total number of frequency segments of the acoustic signal and the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, the acoustic texture anomaly is output. The fault type unit: Based on the fault history weight coefficient, pump frequency coupling matching degree, signal ratio weight, sound pressure cross-modal matching weight, pressure correlation coefficient, the difference between current pressure and reference pressure, and the matching degree between current abnormal soundprint and historical fault soundprint, combined with soundprint anomaly degree, the fault type attribution degree is output. The comprehensive risk unit: Based on the risk coefficient of the transport medium, the regional sensitivity coefficient, the duration of the abnormal signal, the matching coefficient of the special pipe section structure, the location reliability correction coefficient, and the time difference of the abnormal acoustic signal arriving at two adjacent sensors, combined with the acoustic anomaly degree and the fault type attribution degree, a comprehensive risk value and location reliability are output. The disposal analysis module: The system receives the voiceprint anomaly level, fault type attribution, comprehensive risk value, and location confidence level output by the calculation and management module. Based on this, the handling and analysis module classifies the risk level and performs the appropriate actions. Compared with the prior art, the beneficial effects of the present invention are as follows: I. This invention outputs voiceprint anomaly levels through a voiceprint feature unit, and amplifies the contribution of characteristic signals corresponding to pipe blockage and leakage through characteristic frequency band weighting parameters. It also uses hardware attribute correction parameters to offset signal transmission deviations caused by long-term pipe wear and support resonance, and operating condition fluctuation correction parameters to filter out normal flow noise interference caused by changes in slurry concentration and valve opening adjustments. AI adaptive correction parameters enable dynamic iteration of the baseline voiceprint without frequent manual calibration. The global balance weight constrains the output range to remain stable within a preset range. The final output can be directly used to trigger anomaly indicators for subsequent processes, reducing the probability of false alarms from the source while reducing system operation and maintenance costs. This provides a low-cost health monitoring solution for industrial pipelines based on voiceprint recognition and artificial intelligence technology.

[0013] Second, this invention outputs the fault type attribution degree through the fault type unit, and uses the fault history weight parameter to match the historical fault occurrence characteristics of the pipeline. The pump frequency coupling matching degree parameter uses the physical characteristics of strong coupling between pipe blockage and pump frequency and no correlation between leakage and pump frequency to distinguish the signal characteristics of the two types of faults. The air conduction solid conduction signal ratio weight parameter uses the difference in sound propagation path between the two types of faults to enhance the feature discrimination. The sound pressure cross-modal matching weight parameter integrates the correlation characteristics of acoustic and pressure heterogeneous data to improve the classification credibility. The global balance weight, combined with the normalization function, outputs a standard probability value. Finally, it directly outputs a clear fault type determination result, without requiring maintenance personnel to carry out additional manual fault identification work.

[0014] Third, this invention outputs a comprehensive risk value and location reliability through a comprehensive risk unit. It quantifies the consequences of a failure by combining the hazard level of the transported medium and the surrounding environmental attributes with parameters such as the medium risk coefficient and the regional sensitivity coefficient. The Bayesian chain inference correction coefficient assesses the potential evolution risk of the failure and outputs a comprehensive risk value to classify the priority of handling. The special pipe section matching coefficient, combined with the prior knowledge of the high-incidence location of pipeline failure, improves the rationality of the location. The location reliability combines the historical failure probability and signal time sequence characteristics to verify the credibility of the location result. Corresponding weight parameters balance the contribution ratio of different dimension parameters. Finally, the risk level and investigation scope are output synchronously. Operation and maintenance personnel can directly carry out handling work based on the output results without additional manual analysis. Attached Figure Description

[0015] Figure 1 This is a system flowchart of the computing management module and the processing and analysis module of the present invention; Figure 2 This is a system module block diagram of the present invention; Figure 3 This is a flowchart of the steps of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figures 1 to 3 This embodiment provides a sound-based method for monitoring pipe blockage and pipe damage, including the following steps: Step S1: Obtain acoustic signature data, fault signature data, and risk correlation data of the pipeline transporting medium through a multi-source real-time data acquisition module; Step S2: Input the voiceprint feature data, fault feature data and risk association data into the data processing module. The data processing module cleans the input data and then inputs the cleaned data into the calculation management module. Step S3: The calculation management module receives the data output in step S2 to perform analysis and location of anomaly identification, fault classification, and risk rating. Furthermore, the computational management module includes a voiceprint feature unit, a fault type unit, and a comprehensive risk unit; Specifically, the processing flow of the computing management module is as follows: Voiceprint feature unit: Based on the pipe wall wear compensation coefficient, support resonance compensation coefficient, slurry concentration correction coefficient, valve opening linkage correction coefficient, dynamic reference adaptive update coefficient, total number of frequency segments of the acoustic signal, and the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, the acoustic anomaly degree is output. Fault type unit: Based on the fault history weight coefficient, pump frequency coupling matching degree, signal ratio weight, sound pressure cross-modal matching weight, pressure correlation coefficient, difference between current pressure and reference pressure, and matching degree between current abnormal soundprint and historical fault soundprint in the fault feature data, and combined with soundprint anomaly degree, the fault type attribution degree is output. Comprehensive Risk Unit: Based on the risk coefficient of the transport medium, the regional sensitivity coefficient, the duration of abnormal signals, the matching coefficient of special pipe section structure, the location reliability correction coefficient, and the time difference of abnormal acoustic signals arriving at two adjacent sensors in the risk association data, and combined with the acoustic anomaly degree and fault type attribution degree, a comprehensive risk value and location reliability are output. Step S4: Input the voiceprint anomaly degree, fault type attribution degree, comprehensive risk value and location reliability into the handling analysis module. The handling analysis module classifies and executes risk levels based on the input data. Furthermore, the handling analysis module includes a division unit and an action execution unit; Specifically, the units are divided as follows: When the overall risk value is less than 0.3, it is considered a low-risk level. When 0.3 ≤ comprehensive risk value < 0.7, it is classified as medium risk. When the comprehensive risk value is ≥0.7, it is classified as a high-risk level; When the location reliability is ≥0.8, the investigation range is 3 meters before and after the fault location point; When 0.5 ≤ location confidence level < 0.8, the investigation range is 10 meters before and after the fault location point; When the location confidence level is less than 0.5, the investigation scope is the entire 200-meter section of pipeline covered by the two adjacent sensors upstream and downstream of the fault location point. It should be noted that the specific fault location is as follows: The output of the combined risk value RG and location confidence CC are calculated simultaneously. The calculation process completely reuses the existing collected parameters. The specific calculation logic is as follows: First: Locate the sensor corresponding to the positioning; When the acoustic anomaly TA≥30 triggers the abnormal process, the system automatically identifies the two adjacent distributed fiber acoustic sensors (DAS) that first detect the abnormal signal. The one that receives the abnormal signal first is the upstream sensor, and the one that receives it later is the downstream sensor. The pipe section covered by this group of sensors is the range where the fault may exist. Secondly: Recall pre-stored and real-time parameters; Installation spacing between adjacent sensors: This is a fixed value pre-stored during the system deployment phase, corresponding to the actual length of the pipe between two sensors, in meters; Sound wave propagation calibration speed: This is a fixed value calibrated on-site during the system initialization phase. It is calculated by tapping the pipe at the midpoint between two sensors and measuring the time difference between the arrival of the signal at the two sensors. It characterizes the propagation speed of sound waves in the current combination of the pipe and the conveyed slurry, and the unit is meters per second. Abnormal signal arrival time difference (DDT): This is a parameter obtained from real-time acquisition and preprocessing. It is the time it takes for the characteristic peak of the abnormal signal to reach the downstream sensor minus the time it takes to reach the upstream sensor, and the unit is seconds. Then: calculate the coordinates of the fault location point; The mileage coordinates of the fault point are calculated using the time difference location formula: the pre-stored mileage coordinates of the upstream sensor are X0, and the mileage coordinates of the fault point are XX = X0 + (installation spacing between adjacent sensor pairs - acoustic wave propagation calibration speed × DDT) ÷ 2. And: mapping the location point to its actual physical location; The calculated mileage coordinates are automatically matched with the pipeline GIS layer pre-stored in the system, and the corresponding physical location is directly marked on the operation and maintenance map. This location is the benchmark point for the subsequent investigation scope delineation. The calculation result of this location point is directly related to the location reliability CC: the location reliability CC is used to evaluate the reliability of this location point. Therefore, based on the value of the location reliability CC, the corresponding range is expanded outward from this location point as a benchmark, which fully matches the previous investigation range delineation rules. Specifically, the action execution unit is: When the risk level is low, abnormal events are recorded in the system background without pushing alarm information. Maintenance personnel can simply perform routine inspections on the corresponding pipe sections during regular inspection periods. When the risk level is medium, push fault warning messages to the mobile terminal of the operation and maintenance team, simultaneously display the fault type and the scope of investigation, and require the handling to be completed within 24 hours; When the risk level is high, an audible and visual alarm is sent to the dispatch and monitoring room, and a text message is sent to the operation and maintenance manager, requiring them to go to the site immediately for handling. After the maintenance personnel have completed the handling, they will feed back the actual fault type, location, and severity to the system. The system will automatically store the sample in the fault sample database, update the historical fault statistics, and simultaneously fine-tune the conditional probability table of the Bayesian network model to complete the operation loop. Based on the above, the system's voiceprint feature unit, fault type unit, and comprehensive risk unit form a progressive three-level judgment logic, constituting a complete operational closed loop from signal recognition to type judgment and then to decision output. All three are indispensable. The layered and progressive structure avoids the system performing ineffective calculations on all data, greatly reducing the system's operational load. Simultaneously, the calculation results of each layer serve as the basic weight for the next layer's calculation, achieving layer-by-layer fusion of multi-dimensional information and ensuring the rationality and credibility of the final output. The combined system requires no manual intervention to analyze the raw data, directly outputting complete decision information including fault type, risk level, and investigation scope. Maintenance personnel can directly carry out handling work based on the output results without additional analysis and judgment. The entire process uses sound signals as the core data source, perfectly aligning with the system's design positioning. This avoids increased system complexity due to reliance on too many other types of sensors and ensures that even if other auxiliary sensors fail, the system can still perform basic health status monitoring and anomaly alarms for industrial pipelines based on voiceprint recognition and artificial intelligence technology.

[0018] Please refer to Figure 1 , Figure 2 and Figure 3 The processing flow of the voiceprint feature unit is as follows: Step A1: By combining the measured pipe wall thickness with the initial pipe wall thickness, the wear rate is obtained, and the attenuation deviation of the acoustic signal propagation caused by pipe wall wear is analyzed, thereby obtaining the pipe wall wear compensation coefficient. Step A2: By combining the resonant frequency of the pipeline with the reference resonant frequency, the frequency offset rate is obtained to analyze the inherent resonant offset of the pipeline caused by the loosening of the support and hanger, and then the resonant compensation coefficient of the support and hanger is obtained. Step A3: By combining the slurry concentration and the reference delivery concentration, the concentration deviation rate is obtained to analyze the flow acoustic signal intensity deviation caused by the change in slurry concentration, and then the slurry concentration correction coefficient is obtained. Step A4: By combining the opening degree of the pipeline valve with the reference opening degree, the maximum opening degree deviation rate is obtained to analyze the flow field noise deviation caused by valve opening adjustment, and then the valve opening degree linkage correction coefficient is obtained. Step A5: By inputting the fault-free normal sound signals from the past 7 days into the lightweight adaptive learning model, the dynamic baseline adaptive update coefficients are obtained, and then the matching degree between the current normal voiceprint and the original baseline voiceprint is calculated. Step A6: Combine the total number of frequency segments of the acoustic signal, the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, and the reference power spectral density of the i-th frequency segment when the pipeline is running normally under the same working conditions, and combine the global balance weight, hardware attribute parameter weight, working condition fluctuation parameter weight, AI adaptive parameter weight, and feature weight of the i-th frequency segment to finally output the acoustic anomaly degree. Step A7: If the voiceprint anomaly (preset trigger threshold is 30) < 30, it is determined to be a normal working condition. All collected data is only cached for a short period of time and does not trigger subsequent processes. It directly enters the next collection cycle. If the voiceprint anomaly is ≥30, it is determined that there is an abnormal working condition, triggering the subsequent fault analysis process. The voiceprint anomaly value, voiceprint feature data, fault feature data, and risk association data are synchronously input into the fault type unit and the comprehensive risk unit. The calculation formula for voiceprint feature units is as follows: ; in: TA stands for Voiceprint Anomaly, which represents the degree of difference between the currently collected voiceprint and the baseline voiceprint under normal operating conditions. It serves as the first-level judgment threshold of the entire monitoring system. Only when the anomaly exceeds the set threshold will the subsequent fault identification process be triggered, thereby reducing the amount of unnecessary calculations. TAA refers to the pipe wall wear compensation coefficient, which is a correction coefficient to compensate for the attenuation deviation of acoustic signal propagation caused by pipe wall wear. The value ranges from 0.8 to 1.2. The wall thickness data of each section of the pipe can be collected quarterly by ultrasonic wall thickness testing equipment and compared with the initial wall thickness of the pipe. The compensation coefficient is calculated based on the wear rate. The higher the wear rate, the larger the coefficient value. The introduction of this parameter offsets the attenuation deviation of acoustic signal caused by long-term service wear of the pipe and avoids misjudgment of anomalies caused by signal attenuation. Specifically, the method of obtaining the data involves using an ultrasonic wall thickness gauge every quarter to measure the wall thickness at three fixed testing points on each section of the pipeline. The average value is taken as the measured wall thickness h, with the initial wall thickness of the pipeline being h0. The wear rate rw is first calculated as (h0-h)÷h0 ×100%, and then mapped according to the rules. When rw < 5%, then TAA is 1; When 5%≤rw<10%, then TAA=1+0.01×(rw-5) When rw≥10%, then TAA=1.05+0.01×(rw-10), and the maximum shall not exceed 1.2; TAB refers to the pipe support resonance compensation coefficient, which is the correction coefficient for the inherent resonance offset of the pipeline caused by the loosening of the pipe support. The value ranges from 0.7 to 1.3. The inherent resonance frequency of the pipeline can be detected monthly and compared with the initial reference resonance frequency of the pipeline. The compensation coefficient is calculated based on the frequency offset. The larger the offset, the lower the value of the coefficient. The introduction of this parameter is used to cancel the pipeline resonance noise caused by the loosening of the pipe support and avoid misjudging the resonance caused by normal flow as a fault signal. Specifically, the acquisition method involves deploying wireless piezoelectric vibration sensors (built-in battery powered, LoRa / NB-IoT transmission, 5kHz sampling rate, IP68 protection rating) on ​​key supports (elbows, reducers, and cross-section supports). These sensors automatically upload the resonant frequency f data hourly. The initial reference resonant frequency of the pipeline is f0. The frequency offset rate rf is first calculated as |f - f0| ÷ f0 × 100%, and then mapped according to rules. When rf < 3%, TAB is set to 1; When 3%≤rf<10%, then TAB=1-0.04×(rf-3) When rf ≥ 10%, then TAB = 0.7; Long-distance pipeline supports and hangers may loosen due to long-term impact from slurry and ground subsidence. After loosening, the natural resonant frequency of the pipeline will shift. Normally flowing slurry will trigger abnormal resonance in the pipeline, which can be easily misjudged as a fault signal. Current technology hardly pays attention to the impact of the support and hanger status on the acoustic signal. TAC refers to the slurry concentration correction coefficient, which is a correction coefficient to compensate for the deviation in the intensity of the flow acoustic signal caused by changes in slurry concentration. The value ranges from 0.9 to 1.1. The concentration value of the conveyed slurry can be collected in real time by an ultrasonic slurry concentration meter installed at the pipeline inlet, and compared with the concentration under the reference condition. The correction coefficient is calculated based on the concentration deviation. The higher the concentration, the lower the coefficient value. The introduction of this parameter is used to offset the normal flow noise changes caused by slurry concentration fluctuations and avoid false alarms under high concentration conditions. Specifically, an ultrasonic slurry concentration meter is installed at the pipeline inlet, 5 meters from the first bend, to collect the slurry concentration *c* in real time. The design reference delivery concentration is *c0*. First, the concentration deviation rate *rc* = (*c* - *c0*) ÷ *c0* × 100% is calculated, and then mapped according to the rules. When rc > 0, it indicates an increase in concentration; that is, for every 2% increase, TAC decreases by 0.01, with a minimum of 0.9. When rc < 0, it indicates a decrease in concentration; that is, for every 2% decrease, TAC is increased by 0.01, with a maximum of 1.1. When |rc|≤2%, then TAC is 1; TAD refers to the valve opening linkage correction coefficient, which is the correction coefficient to compensate for the flow field noise deviation caused by valve opening adjustment. The value range is 0.6-1.2. The opening data of all regulating valves and branch valves can be collected in real time by the pipeline automatic control system and compared with the valve opening under the reference operating condition. The correction coefficient is calculated based on the opening deviation. The larger the opening deviation, the lower the coefficient value. The introduction of this parameter is used to offset the flow field sudden noise caused by valve opening adjustment and avoid misjudging the noise of normal operating condition adjustment as a fault signal. Specifically, the acquisition method involves reading the opening degree data of all pipeline valves in real time from the on-site PLC automatic control system. i The reference opening degree of each valve is k0. i (i is the valve number), first calculate the maximum opening deviation rate of all valves rk = max(|k i -k0 i |÷k0 i (×100%), then map according to the rules: When rk < 5%, TAD is 1; When 5%≤rk<20%, ​​then TAD=1-0.027×(rk-5; When rk≥20%, then TAD is 0.6; Long-distance slurry pipelines are usually equipped with multiple regulating valves and branch valves. When the valve opening is adjusted, the flow field inside the pipeline will change abruptly, generating flow noise that is highly similar to the blockage / leakage signal. This is a normal operating condition fluctuation. Existing acoustic monitoring systems almost never connect to the valve operation data of the automatic control system, and are very likely to misjudge this normal fluctuation as a fault. This parameter realizes cross-domain data linkage between the monitoring system and the automatic control system. TAE stands for Dynamic Reference Adaptive Update Coefficient, which is the reference voiceprint adaptive correction coefficient output by the AI ​​model. The value ranges from 0.8 to 1.2. Based on a lightweight adaptive learning model, it can take the fault-free normal sound signal of the last 7 days as input, calculate the matching degree between the current normal voiceprint and the original reference voiceprint, and automatically output the corresponding correction coefficient. The introduction of this parameter is used to adaptively adapt to normal working conditions that change slowly, such as seasonal temperature changes and mineral type switching, without the need for manual recalibration of the reference voiceprint periodically. Specifically, the acquisition method is based on a lightweight moving average adaptive model. The input is the average power spectral density (Savg,i) of each frequency band during all alarm-free periods over the past 7 days. The average similarity between the current benchmark and the 7-day average is then calculated. Then map according to the rules: If sim≥95%, then TAE=1; If 85%≤sim<95%, then TAE=1-0.013×(95-sim) If sim < 80%, then TAE = 0.8; Because the acoustic signature of long-distance pipelines will slowly shift with seasonal temperature changes, mineral type switching, and natural pipeline aging, traditional solutions require manual recalibration of the reference acoustic signature periodically, which is prone to false alarms due to outdated references. This parameter is automatically learned by an artificial intelligence model from the acoustic signature samples of the past 7 days of fault-free operating conditions, and dynamically calculates the adaptation range of the reference acoustic signature. It does not require manual intervention and is a typical application of acoustic signature recognition technology to adapt to changes in operating conditions, which can effectively improve the stability of long-term health monitoring of industrial pipelines. TAF refers to the total number of frequency segments of the acoustic signal, which is a fixed parameter preset by the system. In this embodiment, it is preset to 20 segments, which is used to divide the continuous acoustic signal spectrum into discrete segments, simplifying the computational complexity of feature comparison. TAG i This refers to the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment. The axial vibration acoustic signal of the pipeline can be acquired by distributed fiber optic acoustic sensors deployed at equal intervals along the outer wall of the pipeline. The acquired time-domain acoustic signal is subjected to a fast Fourier transform, and the power spectral density of each segment is statistically calculated according to the preset frequency segmentation rules. The introduction of this parameter is used to quantify the energy distribution of the acoustic signal in different frequency segments under the current pipeline operating state, and it is the core data source for acoustic feature comparison. Specifically, the acquisition method involves laying a sensing optical fiber along the outer wall of the pipe, running in the same direction as the pipe, and connecting it to a distributed optical fiber acoustic sensor (DAS) host. A spatial sampling resolution of 200 meters is set, with a fixed sampling rate of 20kHz. A 2048-point Fast Fourier Transform is performed on the acquired time-domain acoustic signal every second to obtain the 0-10kHz spectrum. This spectrum is divided into 20 frequency bands, each segmented in 500Hz increments. The integral result of the power spectral density within each frequency band is the power spectral density (TAG) of the real-time acquired acoustic signal in the i-th frequency segment. i ; TAH i It refers to the reference power spectral density of the i-th frequency segment when the pipeline is operating normally under the same working conditions. During the pipeline's fault-free operation phase, normal sound signals can be continuously collected for 7 days, and the average value of the power spectral density of each frequency segment can be used as the reference value. The introduction of this parameter serves as a reference for comparing soundprint anomalies and is used to distinguish between normal working condition sound signals and fault sound signals. Specifically, during a 7-day period of continuous, trouble-free pipeline operation, the power spectral density of each frequency band was calculated every hour, and the average of all calculation results was taken as the initial TAH. i It will be updated every 7 days thereafter, and the update rule is to take the average power spectral density of each frequency band during all alarm-free periods in the last 7 days. It should be noted that, after obtaining the power spectral density (TAG) of the real-time acquired acoustic signal in the i-th frequency segment... i The reference power spectral density TAH of the i-th frequency segment when the pipeline is operating normally under the same conditions. i When doing this, the two parameters need to be restricted before being input into the voiceprint feature unit; Both of these parameters should be divided by max(TAGA, TAHA) to impose limits. TAHA is the maximum real-time power spectrum value for all frequency bands, and TAGA is the maximum reference power spectrum value for all frequency bands. The difference is constrained to the interval [-1, 1], and the value range is stabilized to 0-1 after squaring, thus improving the calculation stability. T1 refers to the global balance weight one, which is the global output balance coefficient of the voiceprint feature unit. The initial preset value is 70. After the system is installed, the setting is adjusted according to the actual working conditions during the on-site calibration stage. After calibration, it is a fixed value and can be manually fine-tuned later. The introduction of this parameter balances the output range after multi-parameter weighting. T2 refers to the weight of hardware attribute parameters, which is the weighting coefficient of the pipeline hardware attribute correction parameters (pipeline wall wear compensation coefficient TAA and support resonance compensation coefficient TAB). In this embodiment, it is preset to 0.4. It can be a fixed value preset based on the actual measurement data of a large number of mining pipeline monitoring projects. It can be finely adjusted according to the on-site working conditions. The introduction of this parameter is used to balance the contribution ratio of hardware attribute correction parameters to the anomaly calculation. T3 refers to the weight of the operating condition fluctuation parameters, that is, the weighting coefficient of the operating condition fluctuation correction parameters (slurry concentration correction coefficient TAC and valve opening linkage correction coefficient TAD). In this embodiment, it is preset to 0.35. It can be a fixed value preset based on the actual measurement data of a large number of mine pipeline monitoring projects. It can be finely adjusted according to the on-site operating conditions. The introduction of this parameter balances the contribution ratio of the operating condition fluctuation correction parameters to the anomaly calculation. T4 refers to the AI ​​adaptive class parameter weight, that is, the weighting coefficient of the AI ​​adaptive class correction parameter (dynamic benchmark adaptive update coefficient TAE). In this embodiment, it is preset to 0.25. It can be a fixed value preset based on the actual measurement data of a large number of mine pipeline monitoring projects. It can be finely adjusted according to the on-site working conditions. The introduction of this parameter is used to balance the contribution ratio of the AI ​​adaptive class correction parameter to the anomaly calculation. T5 i The feature weight of the i-th frequency segment can be preset based on the frequency characteristics of the fault sound pattern. The low-frequency segment corresponding to the pipe blockage feature and the high-frequency segment corresponding to the leakage feature are given higher weights, while the weight of the other frequency segments is 1. The introduction of this parameter is used to amplify the signal contribution of the fault feature frequency segment and improve the sensitivity of anomaly identification. Specifically, this parameter is preset with a weight of 1.2 in the first and second frequency bands (0-1000Hz, the characteristic frequency band of pipe blockage), a weight of 1.5 in the 17th-20th frequency bands (8000-10000Hz, the characteristic frequency band of leakage), and a weight of 1 in the remaining frequency bands; Based on the above, this voiceprint feature unit is the core module at the entry level of the entire sound-based pipeline blockage and damage monitoring system. It is the first layer of implementation of the core principle of acoustic monitoring in the system. It transforms the raw, unstructured acoustic signals into quantifiable anomaly judgment indicators, achieving the first layer of screening between normal and abnormal operating conditions from the source. It filters out the vast majority of meaningless normal operating data, provides the basis for starting all subsequent fault analysis processes, and is the fundamental prerequisite for the stable operation of the entire system. It avoids the system from performing invalid calculations on all undifferentiated data. The output result, Voiceprint Anomaly Score (TA), is the sole entry point for triggering all subsequent fault identification, risk assessment, and location processes. The system will only initiate further calculations when this result reaches a preset trigger threshold; otherwise, it will be considered a normal operating condition without any further processing. Simultaneously, this result serves as the foundational weight for calculating subsequent fault severity and location reliability, determining the basis for the credibility of all subsequent calculations. The smaller the voiceprint anomaly TA value, the smaller the difference between the currently collected sound signal and the baseline voiceprint under normal operating conditions. The higher the probability that the current operating condition is in normal operation, the less likely the system will trigger the subsequent fault judgment process or generate any alarm information. Maintenance personnel do not need to pay special attention. When the calculated value of the voiceprint feature unit is larger, the greater the difference between the current sound signal and the baseline voiceprint, the higher the probability that there is a fault in the pipeline. After reaching the trigger threshold, the system will automatically start the subsequent full-process analysis. At the same time, this value itself will serve as the core reference for the subsequent fault severity judgment and directly affect the basic level of the subsequent risk value.

[0019] Please refer to Figure 1 , Figure 2 and Figure 3 The fault type unit's processing flow is as follows: Step B1: By introducing voiceprint anomaly degree as the basic input for fault type determination, the higher the anomaly degree, the higher the probability of fault occurrence. Step B2: By statistically analyzing the number of pipe blockages and leaks in the past 3 months, the proportion of pipe blockage and leak failures is analyzed to obtain the historical weighting coefficient of the failure. Step B3: Calculate the pump frequency coupling matching degree by analyzing the degree of matching between the abnormal sound signal fluctuation frequency and the operating frequency of the delivery pump; Step B4: Obtain the signal ratio weight by analyzing the intensity ratio of the air-conducted sound signal to the solid-conducted sound signal in the pipe wall; Step B5: By inputting the current acoustic signature spectrum features and the pressure time series features of the past 5 minutes into the Transformer model, the model outputs the matching degree with the standard pipe blockage and leakage failure modes to obtain the sound pressure cross-modal matching weights. Step B6: The pressure correlation coefficient, the difference between the current pressure and the reference pressure, and the matching degree between the current abnormal soundprint and the historical fault soundprint are combined and calculated. The global balance weight, the working condition correlation parameter weight, the signal feature parameter weight and the AI ​​multimodal parameter weight are combined and normalized to finally output the fault type attribution degree. Step B7, based on the fault type attribution degree, specifically obtains the probability of pipe blockage attribution and the probability of leakage attribution; Step B8: If the probability of pipe blockage is greater than the probability of leakage, then the current fault type is determined to be pipe blockage; if the probability of leakage is greater than the probability of pipe blockage, then the current fault type is determined to be leakage. The calculation formula for the fault type unit is as follows: ; in: PR j This refers to the degree of fault type attribution, specifically representing the probability that the current anomaly belongs to a pipe blockage fault or a leakage fault, with a value range of 0-1; PR1 represents the probability of pipe blockage attribution, and PR2 represents the probability of leakage attribution. This parameter serves as the basis for determining the fault type. The fault type with a higher probability is the fault type determined by the system, providing a basis for subsequent risk assessment. The introduction of voiceprint anomaly degree (TA) serves as the basic input for fault type determination; the higher the anomaly degree, the higher the probability of fault occurrence. The introduction of Sigmoid, in conjunction with classification weights, serves the purpose of normalization constraints, ensuring that the probability output always remains within the effective range of 0-1, and preventing the calculated probability from exceeding the range. P1 refers to the global balancing weight two. In this embodiment, the initial value is preset to 0.02. The introduction of this parameter is used to balance the output range after multi-parameter weighting, and to ensure that the probability value after Sigmoid normalization is stable in the 0-1 range. P2 refers to the weight of the operating condition related parameters, i.e., the operating condition related correction parameters (fault history weight coefficient P3). j Pump frequency coupling matching degree PRA j The weighting coefficient of ) is 0.35. It is a fixed value that can be preset based on the measured data of a large number of mine pipeline monitoring projects. It can be finely adjusted according to the on-site working conditions. The introduction of this parameter is the contribution ratio of the balanced working condition association correction parameter to the fault attribution probability calculation. P3 jThis refers to the historical fault weighting coefficient (j=1 corresponds to pipe blockage, j=2 corresponds to leakage). It is a correction coefficient based on the historical probability of pipeline faults, with a value range of 0.9-1.3. It can count the number of historical faults in the pipeline in the past 3 months, calculate the occurrence ratio of pipe blockage and leakage faults respectively, and calculate the corresponding coefficient based on the ratio. The higher the historical probability, the larger the coefficient value. The introduction of this parameter, combined with the historical fault characteristics of the pipeline, improves the sensitivity of identifying high-probability fault types. Specifically, the system automatically compiles historical fault records from the past three months, with N1 representing the number of pipe blockages, N2 representing the number of leaks, and the total number of faults NM = N1 + N2, which can be directly calculated: P31 = 0.9 + 0.4 × (N1 ÷ NM) (If NM = 0, then P31 = P32 = 1); P32 = 0.9 + 0.4 × (N2 ÷ NM); PRA j This refers to the pump-frequency coupling matching degree (j=1 corresponds to pipe blockage, j=2 corresponds to leakage), which is the degree of matching between the abnormal sound signal fluctuation frequency and the operating frequency of the delivery pump. The value is 0.8-1.5 for pipe blockage and 0.5-1.0 for leakage. The pump's operating frequency can be collected in real time by the variable frequency control system of the delivery pump and compared with the fluctuation frequency of the abnormal sound signal to calculate the frequency overlap. The matching degree is obtained by mapping the overlap degree. The higher the overlap degree, the higher the matching degree for pipe blockage and the lower the matching degree for leakage. The introduction of this parameter takes advantage of the high coupling between the pressure fluctuation of pipe blockage fault and pump frequency, and the fact that leakage fault is unrelated to pump frequency, to improve the accuracy of fault type classification. Specifically, the operating frequency fp of the transfer pump is read in real time through the inverter interface; the main frequency fs is extracted from the spectrum of the current abnormal sound signal, and the maximum overlap is calculated as imp = max(1 - |fs−kk×fp|÷fp) (kk is an integer multiple of 1-5, and the result with the highest overlap is taken). This is a direct calculation where fp is the real-time operating frequency of the transfer pump, i.e., the actual operating frequency of the motor of the plunger pump / centrifugal pump transporting the slurry, in Hz, which determines the pump's stroke per minute. The impeller speed is the reference frequency of the pressure pulsation generated by the pump. kk is the pump frequency harmonic multiple. In addition to the fundamental frequency component that is consistent with fp, the pressure pulsation generated by the industrial pump will also generate higher harmonic components that are 2-5 times the fundamental frequency. When a blockage occurs, the flow channel narrows, which will amplify these harmonic components. This will cause the main frequency of the abnormal sound signal to coincide with the fundamental frequency or any higher harmonic. Therefore, kk is set as the harmonic multiple of the traversal, and the fixed value range is a positive integer of 1, 2, 3, 4 and 5. PRA1 = 0.8 + 0.7 × simp (pipe matching degree); PRA2 = 1.0 - 0.5 × simp (leakage matching degree); Pipe blockage is caused by narrowing of the flow channel inside the pipe. Pressure fluctuations are completely synchronized with the stroke frequency of the delivery pump. The fluctuation frequency of abnormal sound signals is highly coupled with the operating frequency of the pump. Leakage, on the other hand, is a continuous high-frequency signal generated by the outward ejection of the medium, which is completely unrelated to the pump frequency. Technology almost never associates the operating parameters of power equipment with sound pattern fault classification. P4 refers to the signal feature class parameter weights, i.e., the signal feature class correction parameters (signal ratio weights PRB). j The weighting coefficient of the balanced signal feature class correction parameter is 0.3. The contribution of this parameter to the calculation of the fault attribution probability is the proportion of the balance signal feature class correction parameter. PRB j The signal ratio weight refers to the weight of the ratio between air-conducted and solid-conducted signals (j=1 corresponds to pipe blockage, j=2 corresponds to leakage). It is the correction weight corresponding to the intensity ratio of the air-conducted sound signal and the solid-conducted sound signal in the pipe wall. The value is 0.7-1 for pipe blockage and 1-1.8 for leakage. A wall-mounted acoustic sensor (collecting solid-conducted signals) and an industrial-grade explosion-proof microphone (collecting air-conducted signals) 10cm away from the wall can be installed simultaneously at each sensor point. The intensity ratio of the two types of signals is calculated in real time, and the corresponding weight is obtained by mapping the ratio. The higher the ratio, the higher the weight corresponding to leakage and the lower the weight corresponding to pipe blockage. The introduction of this parameter takes advantage of the characteristic that pipe blockage faults only generate solid-conducted signals and leakage faults generate both solid-conducted and air-conducted signals to distinguish the signal characteristics of the two types of faults. Specifically, the acquisition method involves synchronously installing one industrial-grade explosion-proof microphone at each DAS sensor location to collect airborne sound signals. The power integral PPs of the solid-conducting signal and the power integral PPa of the airborne signal (collected by the industrial-grade explosion-proof microphone) are calculated separately, yielding the ratio rPO = PPa ÷ PPs, which can then be directly calculated. PRB2 = 1 + 0.8 × min(rPO, 1) (Leakage weight) PRB1 = 1 - 0.3 × min(rPO,1) (pipe blockage weight); Abnormal sound signals generated by pipe blockage only have a component conducted by the solid (pipe wall), and almost no abnormal signals can be collected from the air outside the pipe; while when there is a leak, the slurry jet will generate a high-frequency signal conducted by the air. Existing monitoring solutions mostly only use solid conduction sensors attached to the pipe and do not collect additional air conduction signals for comparison. P5 refers to the AI ​​multimodal class parameter weights, namely the AI ​​multimodal class correction parameters (sound pressure cross-modal matching weights PRC). j The weighting coefficient of ) is 0.35. The introduction of this parameter balances the contribution of the AI ​​multimodal class correction parameter to the fault attribution probability calculation. PRC jThis refers to the cross-modal matching weight of sound pressure (j=1 corresponds to pipe blockage, j=2 corresponds to leakage), which is the degree of matching between acoustic features, pressure change features, and standard fault modes. The value ranges from 0.8 to 1.5. It can be based on a pre-trained cross-modal Transformer alignment model. The current acoustic spectrum features and the pressure time series features of the past 5 minutes are input in real time. The model automatically outputs the matching degree with the standard pipe blockage and leakage fault modes and maps them to the corresponding weights. The higher the matching degree, the higher the corresponding weight. The introduction of this parameter is achieved by using AI multimodal feature association to fuse the features of two heterogeneous data sources, acoustic and pressure, and reduce the classification errors caused by interference from a single data dimension. Specifically, the acquisition method is based on a pre-trained 2-layer cross-modal Transformer model (the pre-training dataset is a publicly available industrial pipeline fault dataset, containing 1000 sets of pipe blockage and 1000 sets of leakage acoustic and pressure samples; 5 sets of local fault samples need to be fine-tuned on-site for use). enter: ① 20-dimensional power spectral density vectors of the current voiceprint at each frequency band; ② A 10-dimensional vector of pressure differences per minute over the past 5 minutes; Intermediate process: The model performs self-attention encoding on the two feature vectors separately, then calculates the cosine similarity with the standard pipe blockage and leakage fault features, and outputs a matching score of 0-1. j ; Output: PRC j =0.8 + 0.7 × score j ; Because there is a strong coupling relationship between the acoustic signature characteristics and pressure change characteristics of pipe blockage and leakage faults, pipe blockage corresponds to low-frequency pulsating acoustic signature + slow pressure increase, while leakage corresponds to high-frequency continuous acoustic signature + rapid pressure decrease. Traditional methods only analyze the two types of data separately without cross-modal feature association, which can easily lead to misjudgment due to interference from single-dimensional data. This parameter automatically associates the features of the two heterogeneous data types of acoustic signature and pressure by an artificial intelligence multimodal model, and outputs the matching degree with the standard fault mode, further improving the classification accuracy of acoustic signature recognition and providing support for the accurate determination of health faults in industrial pipelines. K j The pressure correlation coefficient (j=1 corresponds to pipe blockage, j=2 corresponds to leakage) is the correlation coefficient between pressure change and fault type. The value is 0.8 for pipe blockage and -0.8 for leakage. It is a fixed value preset based on the physical characteristics of pressure increase during pipe blockage and pressure decrease during leakage. The introduction of this parameter correlates the direction of pressure change with fault type, strengthening the basis for fault type determination. DP refers to the difference between the current pressure and the reference pressure, that is, the difference between the current real-time pressure at the pipeline monitoring point and the reference pressure under normal operating conditions. It can be obtained by collecting the internal pressure of the pipeline in real time through pressure transmitters installed on the pipeline (the pressure transmitters are deployed at intervals of no more than 1 kilometer, and an additional one is deployed at each pump station and valve group to ensure the real-time performance of pressure data. Additional pressure transmitters are installed at elbows, valve groups, and diameter changes. When calculating DP, the pressure transmitter data closest to the fault point is used first). The difference is obtained by subtracting the pre-stored reference pressure value under the same operating conditions. The introduction of this parameter is used to provide a basis for fault judgment in the pressure dimension and to help distinguish between pipe blockage and leakage faults. It should be noted that when the difference DP between the current pressure and the reference pressure is obtained, it needs to be limited before being input into the fault type unit. The difference between the current pressure and the reference pressure DP needs to be divided by the maximum allowable pressure difference in the pipeline design, so as to constrain DP in the range of [-1,1], ensure the stability of the input range of the exponential term, and improve the gradient discrimination of the probability output. M j This refers to the matching degree between the current abnormal soundprint and the historical fault soundprint, that is, the degree of matching between the current abnormal soundprint and the historically stored pipe blockage and leakage fault soundprint samples. The value ranges from 0 to 1. The cosine similarity between the spectral features of the current abnormal soundprint and the features in the fault sample library can be calculated to obtain the corresponding matching degree. The introduction of this parameter utilizes the feature prior of historical faults to improve the accuracy of fault type identification. Based on the above, this fault type unit is the core fault classification module of the system. On the basis of the abnormality determined by the acoustic feature unit, it integrates multi-dimensional information such as acoustic signal features, pressure features, operating condition data and AI multimodal analysis to achieve accurate differentiation between pipe blockage and leakage. This solves the problem that a single signal dimension cannot accurately distinguish the fault category. Since the handling logic and potential consequences of pipe blockage and leakage are completely different, the judgment result of the fault type unit is the core basis for subsequent risk assessment and handling strategy formulation. It is the core support for the system to output targeted operation and maintenance guidance. The two types of fault attribution probabilities output by the fault type unit are the sole basis for the system to determine the fault type. The type with the higher value is the fault type finally confirmed by the system. At the same time, the maximum value of the two probabilities will serve as the core weight for subsequent risk assessment and location reliability calculation, directly determining the basic credibility of the subsequent results. This result will be directly synchronized to the fault tag library of the operation and maintenance system, automatically matching the standardized handling plan for the corresponding fault type, without requiring operation and maintenance personnel to manually determine the fault category. For the probability of attribution to a single type of fault, the smaller the value, the lower the match between the current abnormal signal and the characteristics of that type of fault, and the lower the probability of that type of fault occurring. The larger the value, the higher the match, and the higher the probability of that type of fault occurring. When the difference between the two probabilities is smaller, it indicates that the distinguishability of the fault characteristics is lower, and the system will automatically increase the caution of subsequent risk assessments and lower the initial benchmark of location reliability. When the difference between the two probabilities is larger, it indicates that the reliability of the fault type determination is higher, and the basic reliability of subsequent risk assessments and location results will also be improved in tandem.

[0020] Please refer to Figure 1 , Figure 2 and Figure 3 The processing flow for the comprehensive risk unit is as follows: Step C1: Input the maximum value of the two fault types, pipe blockage and leakage, obtained from the fault type unit into the comprehensive risk unit to obtain the maximum fault assignment probability, thereby quantifying the credibility basis of the risk assessment. Step C2: By combining the pH value and heavy metal content of the pipeline transport medium, we can obtain the risk amplification factors caused by the pollution and corrosiveness of the transported slurry, and thus calculate the risk coefficient of the transport medium. Step C3: By connecting to the mine's GIS system, the environmental attributes around the fault point are identified based on the spatial location of the pipeline, and then values ​​are assigned based on the sensitivity of the surrounding environment to obtain the regional sensitivity coefficient. Step C4: By analyzing the duration of the current abnormal sound signal from the first trigger threshold to the current moment, the duration of the abnormal signal is obtained, and combined with the static attribute parameter weights, AI inference parameter weights, inference risk correction coefficients and fault duration coefficients, a comprehensive risk value is output. Step C5: Analyze the mileage coordinates of all special structures in the pipeline to calculate the distance between the fault location and the nearest special structure point, so as to obtain the matching coefficient of the special pipe section structure. Step C6: By inputting the time difference of the abnormal acoustic signal arriving at two adjacent sensors, the number of historical faults in the pipe section, and the signal amplitude of the two adjacent sensors into the Bayesian network model, the location confidence correction coefficient is obtained. Step C7: Combine the time difference of abnormal sound wave signal reaching two adjacent sensors, the theoretical time difference of normal sound wave propagation at the same distance, and the maximum theoretical time difference of sound wave propagation between the two sensors, and combine the structural attribute class parameter weights and AI inference class weight coefficients to output the location confidence. The calculation formula for the comprehensive risk unit is as follows: ; ; in: RG stands for Comprehensive Risk Value, which represents the severity and risk of consequences of the current fault. The value ranges from 0 to 1. This calculation result serves as the basis for fault priority determination, and different risk levels correspond to different operation and maintenance handling strategies. CC stands for location confidence, which represents the credibility of the current fault location result. The value ranges from 0 to 1. This calculation result serves as the basis for determining the scope of troubleshooting by maintenance personnel. The higher the confidence level, the smaller the troubleshooting scope. The introduction of voiceprint anomaly TA is used to quantify the severity of faults based on the degree of voiceprint anomaly. max(PR1, PR2) represents the maximum fault attribution probability, which is the larger value among the pipe blockage and leakage attribution probabilities output by the fault type unit. It can be used to quantify the credibility of risk assessment based on the confidence level of fault type determination. RGA refers to the risk factor of the transported medium, which represents the risk amplification factor caused by the pollution and corrosiveness of the transported slurry. The value ranges from 1 to 1.2. It can be a fixed value preset according to the composition of the transported slurry. A higher value is used when transporting highly corrosive and highly polluting slurries, and a lower value is used when transporting ordinary slurries. The introduction of this parameter amplifies the risk weight of dangerous medium leakage and avoids underestimating the risk of high-hazard medium failure. Specifically, the acquisition method can be based on a fixed preset based on the slurry test report: pH < 4 or pH > 10, or heavy metal content exceeding the emission standard by more than 3 times: RGA = 1.2; pH 4-6 or 8-10, or heavy metals exceeding the standard by 1-3 times: RGA = 1.1; pH 6-8, heavy metals not exceeding the standard: RGA=1; RGB refers to the regional sensitivity coefficient, which represents the risk amplification factor brought about by the sensitivity of the surrounding environment of the fault point. The value range is 1-1.5. It can be connected to the mining GIS system. Based on the spatial location of the pipeline, the environmental attributes (drinking water source, residential area, uninhabited area, etc.) around the fault point are identified. The corresponding coefficient is obtained according to the sensitivity level. The higher the sensitivity level, the larger the coefficient value. The introduction of this parameter is used to amplify the risk weight of faults in sensitive areas and avoid the occurrence of environmental protection and personnel safety accidents. Specifically, the acquisition method involves connecting to the SHP format pipeline layer of the mining GIS system and determining the fault location based on the currently calculated coordinates: Distance from drinking water sources, basic farmland, and residential areas <50m: RGB = 1.5; At a distance of 50-200m from the aforementioned area, RGB = 1.25; If the distance is greater than 200m and the location is within the production area, RGB = 1; The environmental attributes of the failure point directly affect the severity of the accident's consequences. For example, the environmental risks of leak points near rivers and residential areas are much higher than those in wilderness areas. Existing risk assessment algorithms almost only consider process parameters and do not integrate GIS geographic information to adjust risk weights. RGC refers to the duration of the abnormal signal, which is the duration of the current abnormal sound signal from the first trigger threshold (voiceprint abnormality TA≥30) to the current moment, in minutes. That is, the system automatically counts the duration of the abnormal signal. The introduction of this parameter quantifies the degree of continuous development of the fault. The longer the duration, the more serious the fault. R1 refers to the weight of static attribute class parameters, that is, the weighting coefficient of static attribute class correction parameters (transport medium risk coefficient RGA and regional sensitivity coefficient RGB). In this embodiment, it is preset to 0.55. The introduction of this parameter is used to balance the contribution ratio of static attribute class correction parameters to risk value calculation. R2 refers to the weight of AI inference parameters, that is, the weighting coefficient of AI inference correction parameters (inference risk correction coefficient R3). In this embodiment, it is preset to 0.45. The introduction of this parameter is used to balance the contribution of AI inference correction parameters to the risk value calculation. R3 refers to the inference risk correction coefficient, which is the fault chain evolution risk correction coefficient output by the AI ​​model. Its value ranges from 0.9 to 1.8. Based on the pre-trained Bayesian network fault inference model, the model can automatically infer the chain evolution probability of the fault by inputting fault type, fault location, pipeline attributes, and environmental attribute data in real time and mapping the corresponding correction coefficient. The more severe the evolution consequences, the larger the coefficient value. The introduction of this parameter is used to assess the risk in combination with the potential evolution consequences of the fault, so as to avoid the underestimation of faults that appear minor but have serious consequences. Specifically, the acquisition method is based on a Bayesian network model with a fixed structure (the conditional probability table is pre-trained based on 100 sets of historical mine failure consequence data, requiring no secondary training on-site): enter: Four real-time parameters (fault type, transported medium risk factor, area sensitivity factor, and pipe section wear rate). Intermediate process: The model infers the severity of the fault chain evolution based on the pre-stored conditional probability table and directly outputs a continuous value of 0.9-1.8. Output: It is directly substituted into the calculation as the inference risk correction coefficient R3; R4 refers to the fault duration coefficient, which is the risk amplification coefficient of fault duration. In this embodiment, it is preset to 0.05 / minute, but can be preset according to the fault evolution rate. The introduction of this parameter is used to amplify the risk value with the fault duration, so as to encourage maintenance personnel to deal with faults that have not been resolved for a long time in a timely manner. CCA refers to the special pipe section structure matching coefficient, which is the correction coefficient corresponding to the degree of matching between the positioning result and the special structure point of the pipeline. The value ranges from 1 to 1.3. It can be calculated by using the coordinates of all special structure points (elbows, welds, tees and reducers) of the pipeline in advance, and the distance between the current positioning result and the nearest special structure point can be calculated. The corresponding coefficient is obtained according to the distance mapping. The closer the distance, the larger the coefficient value. The introduction of this parameter, combined with the prior knowledge of the structure location of the pipeline where the fault occurs frequently, improves the positioning reliability of the high-fault area. Specifically, the acquisition method involves pre-storing the mileage coordinates of all special structures (elbows, welds, tees, reducers) of the pipeline, the currently calculated fault location, calculating the distance dd between the fault location and the nearest special structure point, and mapping them according to rules. dd < 3m: CCA = 1.3; 3m≤dd<5m: CCA=1.15; dd≥5m: CCA=1; Since most pipe blockages and leaks occur at special structural locations such as elbows, welds, tees, and reducers (where the flow field is more complex, stress is more concentrated, and the probability of construction defects is higher), existing acoustic location algorithms rely solely on the time difference of sound wave propagation to calculate the fault location. They almost never combine pipeline design and as-built data to optimize the confidence of the location results. This parameter enables cross-domain linkage between the monitoring system and pipeline design asset data, demonstrating superior innovation. CCB stands for Location Confidence Correction Coefficient, which is a correction coefficient for the reasonableness of the positioning result output by the AI ​​model. The value ranges from 0.8 to 1.3. Based on a pre-trained Bayesian network fault inference model, the model can automatically output a reasonableness score for the positioning result by inputting the positioning result, the historical fault probability of the pipeline segment, and the signal time series matching data in real time. The higher the reasonableness, the larger the coefficient value. The introduction of this parameter combines the historical fault prior of the pipeline segment with the signal time series characteristics to verify the positioning result and reduce the positioning deviation caused by local signal interference. Specifically, the acquisition method uses the same Bayesian network model as the inference risk correction coefficient R3, serving as the model's second output node. enter: Three real-time parameters (DDT of abnormal acoustic signal arriving at two adjacent sensors, number of historical faults in this pipe section, and signal amplitude difference between two adjacent sensors); Intermediate process: The model infers the rationality of the location result based on the pre-stored conditional probability table and directly outputs a continuous value of 0.8-1.3; Output: It is directly substituted into the calculation as the location reliability correction coefficient CCB; The introduction of the inference risk correction coefficient R3 and the location confidence correction coefficient CCB is beneficial because traditional risk calculation only assesses the surface severity of the current fault and does not consider the chain evolution risk of the fault (for example, a small leak near a water source may cause a major environmental accident, and a minor blockage in a severely worn pipe section may quickly evolve into a pipe burst). It also does not combine the historical fault prior probability of the pipe section to correct the location results. Bayesian networks are a classic probabilistic inference model in the field of artificial intelligence. Based on historical fault prior knowledge and real-time operating data, they can automatically infer the potential evolutionary consequences of the fault and the rationality of the location results. When combined with voiceprint recognition technology, dynamic assessment of the health status of industrial pipelines throughout their entire life cycle can be achieved. DDT refers to the time difference between the arrival of an abnormal acoustic signal at two adjacent sensors. That is, the time difference between the arrival of an abnormal acoustic signal at two adjacent acoustic sensors upstream and downstream of the fault point. The time difference can be calculated by comparing the arrival time of the characteristic peak of the abnormal signal at the two sensors with the signals collected by the distributed acoustic sensors. This parameter is introduced as the core calculation parameter for fault location, and the location of the fault point is calculated based on the sound wave propagation speed and time difference. DDTA refers to the theoretical time difference of normal sound wave propagation at the same distance, which is the theoretical time difference obtained by dividing the distance between two adjacent sensors by the speed of sound wave propagation. It can be calculated as a fixed value based on the sensor installation spacing and the pre-calibrated speed of sound wave propagation in the current pipeline and slurry combination. This parameter serves as a benchmark reference for the time difference and is used to evaluate the rationality of the current time difference. DDTQ refers to the maximum theoretical time difference of sound wave propagation between two sensors, which is the threshold for the longest time that sound wave can propagate between the two sensors. It can be calculated as a fixed value based on the sensor installation spacing and the minimum propagation speed of sound wave in slurry. The introduction of this parameter serves as a filtering threshold for abnormal time differences to eliminate obviously unreasonable interference signals. CA refers to the weight of structural attribute class parameters, that is, the weighting coefficient of structural attribute class correction parameters (specific pipe segment structural matching coefficient CCA). In this embodiment, it is preset to 0.4. The introduction of this parameter is used to balance the contribution ratio of structural attribute class correction parameters to the location confidence calculation. CB refers to the AI ​​inference class weight coefficient, which is the weighting coefficient of the AI ​​inference class parameter (location reliability correction coefficient CCB). In this embodiment, it is preset to 0.6. The introduction of this parameter balances the contribution of the AI ​​inference class correction parameter to the location reliability calculation. Based on the above, this comprehensive risk unit is the final decision output module of the system. After completing the anomaly identification and fault type determination, it integrates multi-dimensional information such as pipeline attributes, environmental attributes, fault evolution logic, and location verification. At the same time, it outputs two types of decision information: comprehensive fault risk value and location reliability. This solves the problem that the previous process can only identify anomalies and determine fault types, but cannot clarify the severity of the fault, the scope of investigation, and the priority of handling. It is the core link of the system to realize a complete closed loop from anomaly identification to implementable operation and maintenance guidance. The comprehensive risk value output by the comprehensive risk unit is the sole basis for the system to classify fault alarm levels and determine the priority of operation and maintenance. Different levels of risk values ​​correspond to different alarm push methods and response time limits, which directly guide the work priority ranking of operation and maintenance personnel. The output location reliability is the core basis for the system to delineate the fault investigation scope, which directly corresponds to the recommended investigation area on the operation and maintenance map, eliminating the need for operation and maintenance personnel to inspect the entire pipeline. Regarding the overall risk value, a smaller value indicates a lower severity and potential consequences of the fault, corresponding to a lower priority for handling, meaning maintenance personnel can address it during routine inspections. A larger value indicates a higher severity and potential consequences of the fault, corresponding to a higher priority for handling, requiring immediate on-site intervention. Regarding location reliability, a smaller value indicates lower reliability of the location results, and the system will automatically expand the recommended investigation scope to avoid missing fault points. A larger value indicates higher reliability of the location results, and the system will automatically narrow the recommended investigation scope, reducing the workload for maintenance personnel.

[0021] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of monitoring for blockages and breaks in a pipe based on sound, characterized in that, Includes the following steps: Step S1: Obtain acoustic signature data, fault signature data, and risk correlation data of the pipeline transporting medium through a multi-source real-time data acquisition module; Step S2: Input the voiceprint feature data, fault feature data and risk association data into the data processing module. The data processing module cleans the input data and then inputs the cleaned data into the calculation management module. Step S3: The calculation and management module receives the data output in step S2 and uses the voiceprint feature unit, fault type unit, and comprehensive risk unit to realize the analysis and location of anomaly identification, fault classification, and risk rating. The processing flow of the computing management module is as follows: Step A: Based on the pipe wall wear compensation coefficient, support resonance compensation coefficient, slurry concentration correction coefficient, valve opening linkage correction coefficient, dynamic benchmark adaptive update coefficient, total number of frequency segments of the acoustic signal, and the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, output the acoustic anomaly degree. Step B: Based on the fault history weight coefficient, pump frequency coupling matching degree, signal ratio weight, sound pressure cross-modal matching weight, pressure correlation coefficient, difference between current pressure and reference pressure, and matching degree between current abnormal soundprint and historical fault soundprint in the fault feature data, and combined with soundprint anomaly degree, output the fault type attribution degree. Step C: Based on the risk coefficient of the transport medium, the regional sensitivity coefficient, the duration of the abnormal signal, the matching coefficient of the special pipe section structure, the location confidence correction coefficient, and the time difference of the abnormal acoustic signal arriving at two adjacent sensors in the risk association data, and combined with the acoustic anomaly degree and the fault type attribution degree, output the comprehensive risk value and the location confidence. Step S4: Input the voiceprint anomaly degree, fault type attribution degree, comprehensive risk value and location reliability into the handling analysis module. The handling analysis module classifies and executes risk levels based on the input data.

2. The sound-based method for monitoring pipe blockage and pipe damage according to claim 1, characterized in that: The processing procedure for the voiceprint feature unit is as follows: Step A1: By combining the measured pipe wall thickness with the initial pipe wall thickness, the wear rate is obtained, and the attenuation deviation of the acoustic signal propagation caused by pipe wall wear is analyzed, thereby obtaining the pipe wall wear compensation coefficient. Step A2: By combining the resonant frequency of the pipeline with the reference resonant frequency, the frequency offset rate is obtained to analyze the inherent resonant offset of the pipeline caused by the loosening of the support and hanger, and then the resonant compensation coefficient of the support and hanger is obtained. Step A3: By combining the slurry concentration and the reference delivery concentration, the concentration deviation rate is obtained to analyze the flow acoustic signal intensity deviation caused by the change in slurry concentration, and then the slurry concentration correction coefficient is obtained. Step A4: By combining the opening degree of the pipeline valve with the reference opening degree, the maximum opening degree deviation rate is obtained to analyze the flow field noise deviation caused by valve opening adjustment, and then the valve opening degree linkage correction coefficient is obtained. Step A5: By inputting the fault-free normal sound signals from the past 7 days into the lightweight adaptive learning model, the dynamic baseline adaptive update coefficients are obtained, and then the matching degree between the current normal voiceprint and the original baseline voiceprint is calculated. Step A6: Combine the total number of frequency segments of the acoustic signal, the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, and the reference power spectral density of the i-th frequency segment when the pipeline is running normally under the same working conditions, and combine the global balance weight, hardware attribute parameter weight, working condition fluctuation parameter weight, AI adaptive parameter weight, and feature weight of the i-th frequency segment to finally output the acoustic anomaly degree. Step A7: If the voiceprint anomaly score is <30, it is determined to be a normal working condition. All collected data will only be cached for a short period of time and will not trigger subsequent processes. The process will directly enter the next collection cycle. If the voiceprint anomaly degree is ≥30, it is determined that there is an abnormal operating condition, triggering the subsequent fault analysis process. The voiceprint anomaly degree value, voiceprint feature data, fault feature data, and risk association data are simultaneously input into the fault type unit and the comprehensive risk unit.

3. The sound-based method for monitoring pipe blockage and pipe damage according to claim 2, characterized in that: The processing procedure for the fault type unit is as follows: Step B1: By introducing voiceprint anomaly degree as the basic input for fault type determination, the higher the anomaly degree, the higher the probability of fault occurrence. Step B2: By statistically analyzing the number of pipe blockages and leaks in the past three months, the proportion of pipe blockage and leak failures is analyzed to obtain the historical weighting coefficient of the failures. Step B3: Calculate the pump frequency coupling matching degree by analyzing the degree of matching between the abnormal sound signal fluctuation frequency and the operating frequency of the delivery pump; Step B4: Obtain the signal ratio weight by analyzing the intensity ratio of the air-conducted sound signal to the solid-conducted sound signal in the pipe wall; Step B5: By inputting the current acoustic signature spectrum features and the pressure time series features of the past 5 minutes into the Transformer model, the model outputs the matching degree with the standard pipe blockage and leakage failure modes to obtain the sound pressure cross-modal matching weights. Step B6: The pressure correlation coefficient, the difference between the current pressure and the reference pressure, and the matching degree between the current abnormal soundprint and the historical fault soundprint are combined and calculated. The global balance weight, the working condition correlation parameter weight, the signal feature parameter weight and the AI ​​multimodal parameter weight are combined and normalized to finally output the fault type attribution degree. Step B7, based on the fault type attribution degree, specifically obtains the probability of pipe blockage attribution and the probability of leakage attribution; Step B8: If the probability of pipe blockage is greater than the probability of leakage, then the current fault type is determined to be pipe blockage; if the probability of leakage is greater than the probability of pipe blockage, then the current fault type is determined to be leakage.

4. The sound-based method for monitoring pipe blockage and pipe damage according to claim 3, characterized in that: The processing procedure for the integrated risk unit is as follows: Step C1: Input the maximum value of the two fault types, pipe blockage and leakage, obtained from the fault type unit into the comprehensive risk unit to obtain the maximum fault assignment probability, thereby quantifying the credibility basis of the risk assessment. Step C2: By combining the pH value and heavy metal content of the pipeline transport medium, we can obtain the risk amplification factors caused by the pollution and corrosiveness of the transported slurry, and thus calculate the risk coefficient of the transport medium. Step C3: By connecting to the mine's GIS system, the environmental attributes around the fault point are identified based on the spatial location of the pipeline, and then values ​​are assigned based on the sensitivity of the surrounding environment to obtain the regional sensitivity coefficient. Step C4: By analyzing the duration of the current abnormal sound signal from the first trigger threshold to the current moment, the duration of the abnormal signal is obtained, and combined with the static attribute parameter weights, AI inference parameter weights, inference risk correction coefficients and fault duration coefficients, a comprehensive risk value is output. Step C5: Analyze the mileage coordinates of all special structures in the pipeline to calculate the distance between the fault location and the nearest special structure point, so as to obtain the matching coefficient of the special pipe section structure. Step C6: By inputting the time difference of the abnormal acoustic signal arriving at two adjacent sensors, the number of historical faults in the pipe section, and the signal amplitude of the two adjacent sensors into the Bayesian network model, the location confidence correction coefficient is obtained. Step C7: Combine the time difference of the abnormal sound wave signal reaching the two adjacent sensors, the theoretical time difference of normal sound wave propagation at the same distance, and the maximum theoretical time difference of sound wave propagation between the two sensors, and combine the structural attribute class parameter weights and AI inference class weight coefficients to output the location confidence.

5. The sound-based method for monitoring pipe blockage and pipe damage according to claim 1, characterized in that: The processing analysis module includes a division unit and an action execution unit.

6. The sound-based method for monitoring pipe blockage and pipe damage according to claim 5, characterized in that: The specific division unit is: When the overall risk value is less than 0.3, it is considered a low-risk level. When 0.3 ≤ comprehensive risk value < 0.7, it is classified as medium risk. When the comprehensive risk value is ≥0.7, it is classified as a high-risk level; When the location reliability is ≥0.8, the investigation range is 3 meters before and after the fault location point; When 0.5 ≤ location confidence level < 0.8, the investigation range is 10 meters before and after the fault location point; When the location confidence level is less than 0.5, the investigation scope is the entire 200-meter section of pipeline covered by the two adjacent sensors upstream and downstream of the fault location point.

7. The sound-based method for monitoring pipe blockage and pipe damage according to claim 6, characterized in that: The action execution unit is specifically: When the risk level is low, abnormal events are recorded in the system background without pushing alarm information. Maintenance personnel can simply perform routine inspections on the corresponding pipe sections during regular inspection periods. When the risk level is medium, push fault warning messages to the mobile terminal of the operation and maintenance team, simultaneously display the fault type and the scope of investigation, and require the handling to be completed within 24 hours; When the risk level is high, an audible and visual alarm is sent to the dispatch and monitoring room, and a text message is sent to the operation and maintenance manager, requiring them to go to the site immediately for handling.

8. A monitoring system for implementing the sound-based pipe blockage and pipe damage monitoring method of claim 1, characterized in that: It includes a multi-source data real-time acquisition module, a data processing module, a computing management module, and a disposal and analysis module; The multi-source data real-time acquisition module: Used to acquire acoustic signature data, fault signature data, and risk correlation data when transporting media through pipelines; The data processing module: It is used to receive voiceprint feature data, fault feature data and risk association data, and clean the received data so that the cleaned data can be input into the calculation management module. The computing management module: It is used to receive voiceprint feature data, fault feature data and risk correlation data output by the data processing module, so as to perform analysis and location of anomaly identification, fault classification and risk rating; The computing management module includes: Voiceprint feature unit, fault type unit, and comprehensive risk unit; The voiceprint feature unit: Based on the pipe wall wear compensation coefficient, support and hanger resonance compensation coefficient, slurry concentration correction coefficient, valve opening linkage correction coefficient, dynamic reference adaptive update coefficient, the total number of frequency segments of the acoustic signal and the power spectral density of the real-time acquired acoustic signal in the i-th frequency segment, the acoustic texture anomaly is output. The fault type unit: Based on the fault history weight coefficient, pump frequency coupling matching degree, signal ratio weight, sound pressure cross-modal matching weight, pressure correlation coefficient, the difference between current pressure and reference pressure, and the matching degree between current abnormal soundprint and historical fault soundprint, combined with soundprint anomaly degree, the fault type attribution degree is output. The comprehensive risk unit: Based on the risk coefficient of the transport medium, the regional sensitivity coefficient, the duration of the abnormal signal, the matching coefficient of the special pipe section structure, the location reliability correction coefficient, and the time difference of the abnormal acoustic signal arriving at two adjacent sensors, combined with the acoustic anomaly degree and the fault type attribution degree, a comprehensive risk value and location reliability are output. The disposal analysis module: The system receives the voiceprint anomaly level, fault type attribution, comprehensive risk value, and location confidence level output by the calculation and management module, and the handling and analysis module classifies and executes the risk level accordingly.

Images