Coal mine pressure ventilation pipeline monitoring and positioning method and system based on multi-modal fusion

Abstract

The invention provides a coal mine pressure ventilation pipeline monitoring and positioning method and system based on multi-modal fusion, and relates to the technical field of data processing. The method comprises the following steps: mapping sensors to pipe section space coordinates based on a pipe network topological structure diagram to form equipment space mapping distribution, and determining a distributed monitoring area according to the equipment space mapping distribution; the method comprises the following steps: acquiring monitoring data including pressure, flow, sound wave and voltage through each sensor, and after time alignment, inputting the monitoring data into a Transform model matched with a monitoring area to obtain a theoretical pressure loss value; the theoretical value is compared with the actual pressure loss value, and when the difference value exceeds a preset threshold value, it is judged that leakage exists, and a positioning process is started; a sound wave peak value and a peak value time difference are determined by using a first sound wave sensor and an adjacent second sound wave sensor in a distributed monitoring area, and a leakage point position is accurately calculated by combining a distance between the sensors and a sound wave propagation speed. Accurate identification and rapid positioning of leakage under complex dynamic working conditions are realized, the false alarm rate is effectively reduced, and the positioning precision is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for monitoring and locating coal mine compressed air pipelines based on multimodal fusion. Background Technology

[0002] The compressed air system in coal mines is a core facility providing power (for equipment such as pneumatic picks and drills) and breathing air for personnel during underground operations. Its pipeline network is widely distributed and operates under high pressure for extended periods. Due to pipeline corrosion, external damage, and loose joints, compressed air pipelines are prone to pressure loss and even leakage, leading to insufficient air supply, energy waste, and in severe cases, potentially causing safety accidents such as gas accumulation. Therefore, there is an urgent need for a technology capable of real-time monitoring of the compressed air pipeline's operating status and quickly and accurately locating leaks.

[0003] In existing technologies, pressure sensors and flow meters are typically installed at key nodes of the compressed air pipeline network. The theoretical pressure loss between pipe sections is calculated using real-time collected pressure and flow data. The theoretical value is then compared with the measured value. When the deviation between the two exceeds a preset fixed threshold, the system determines that there is an anomaly in the corresponding pipe section and roughly delineates the pipeline section where the fault occurred based on the sensor's installation location.

[0004] However, the aforementioned existing technologies mainly rely on static models based on a single physical dimension for discrimination, ignoring the strong nonlinear and time-varying characteristics of downhole compressed air systems under frequent start-up and shutdown conditions with multiple devices. This presents a technical problem in achieving accurate monitoring and positioning in complex dynamic environments. Specifically, when large downhole pneumatic equipment suddenly starts or stops, the airflow in the pipeline network will generate violent transient pressure fluctuations. Such normal operating condition fluctuations are easily misjudged as leakage faults under a single threshold discrimination logic, resulting in an extremely high false alarm rate for the system. Summary of the Invention

[0005] This application provides a method and system for monitoring and locating coal mine compressed air pipelines based on multimodal fusion, which can be used to accurately identify and quickly and precisely locate leaks in compressed air pipelines.

[0006] Firstly, this application provides a method for monitoring and locating coal mine compressed air pipelines based on multimodal fusion, applied to a compressed air pipeline monitoring system. The compressed air pipeline monitoring system includes a sensor set deployed along the compressed air pipeline. This sensor set includes multiple pressure sensors, multiple flow sensors, multiple acoustic sensors, and at least one voltage sensor connected to the compressed air equipment, as well as a data processing unit communicatively connected to each sensor. The method includes: mapping the sensors in the sensor set to corresponding pipe segment spatial coordinates based on a preset pipeline topology diagram, forming a spatial mapping distribution of the equipment; determining the area containing each sensor in the sensor set as a distributed monitoring area based on the spatial mapping distribution of the equipment; and acquiring monitoring data corresponding to the distributed monitoring area based on each sensor. The monitoring data includes pressure data, flow data, acoustic wave data, and voltage data. After time alignment, this monitoring data is input into a Transformer model matching the distributed monitoring area to obtain the theoretical pressure loss value. After obtaining the actual pressure loss value under the current operating conditions, the difference between the actual pressure loss value and the theoretical pressure loss value is calculated. If the current pressure loss difference is greater than a preset pressure loss threshold, a leak is identified, and the leak location process is initiated: based on the first acoustic wave sensor and at least one adjacent second acoustic wave sensor within the distributed monitoring area, the peak values ​​and peak time differences of the acoustic signals are determined. The leak location is then determined by combining the peak time difference, the distance between the first and second acoustic wave sensors, and the sound wave propagation speed.

[0007] By adopting the above technical solution, a multimodal fusion monitoring system was constructed. Pressure and flow sensors provide pipeline fluid dynamics characteristics, acoustic sensors capture acoustic signal characteristics generated by leaks, and voltage sensors reflect the real-time operating status of equipment. Data from the four types of sensors are time-aligned and input into a Transformer model for joint modeling. The Transformer model, through a self-attention mechanism, can learn the temporal correlations and interactions between different modal data, incorporating equipment start-up and shutdown information as contextual features into the pressure loss prediction process, thereby enabling the calculation of theoretical pressure loss values ​​to dynamically adapt to changes in actual operating conditions. When an anomaly is detected, the system immediately activates the acoustic sensor network for precise location, calculating the leak point coordinates using the time difference of sound wave arrival and the distance relationship between sensors. This closed-loop mechanism from anomaly detection to precise location enables accurate identification and rapid location of leaks under complex dynamic operating conditions, effectively reducing the false alarm rate and improving location accuracy.

[0008] In some embodiments of the first aspect, the method further includes: acquiring real-time pipe fluid pressure values ​​collected by each pressure sensor within the distributed monitoring area, as well as associated pipe temperature values; and calculating the actual sound wave propagation speed under the current operating conditions using the real-time pipe fluid pressure values ​​and the pipe temperature values ​​based on a preset gas state equation.

[0009] By adopting the above technical solution, the fluid pressure and temperature in the underground compressed air pipeline fluctuate with changes in operating conditions. Since the propagation speed of sound waves in a gaseous medium is closely related to gas density and temperature, using a fixed sound wave propagation speed would lead to positioning errors. By real-time acquisition of fluid pressure and temperature values ​​within the pipeline, and calculating the actual gas density and sound speed under the current operating conditions based on the gas state equation, the sound wave propagation speed parameter can dynamically adapt to the actual medium conditions. This real-time corrected sound wave propagation speed is applied throughout the entire process of leak point location calculation, ensuring the accuracy of distance estimation. Especially in deep well long-distance transport scenarios, where pressure and temperature variations are significant, using the actual sound wave propagation speed, compared to using a standard constant, can reduce positioning errors by several meters or even tens of meters, significantly improving the reliability of leak point location and the targeted nature of on-site maintenance.

[0010] In a second aspect, this application provides a compressed air duct monitoring system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the compressed air duct monitoring system to perform the method described in the first aspect and any possible implementation thereof.

[0011] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on a compressed air duct monitoring system, cause the compressed air duct monitoring system to perform the method described in the first aspect and any possible implementation thereof.

[0012] Fourthly, this application provides a computer program product that, when run on a compressed air duct monitoring system, causes the compressed air duct monitoring system to perform the method described in the first aspect and any possible implementation thereof.

[0013] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By employing multimodal sensor fusion monitoring and operating condition adaptive prediction technology based on the Transformer model, the data from four types of sensors—pressure, flow, acoustic waves, and voltage—are time-aligned before being input into the Transformer model. This allows the model to learn the temporal correlations and interactions between different modal data through a self-attention mechanism. Furthermore, operating condition information such as equipment start-up and shutdown is incorporated as contextual features into the pressure loss prediction process. Therefore, this effectively solves the technical problem in existing technologies where single-physical-dimensional static models struggle to distinguish between normal operating condition fluctuations and actual leaks, resulting in an extremely high false alarm rate. Consequently, it achieves accurate identification of pipeline leaks under complex dynamic operating conditions of frequent start-up and shutdown of large underground pneumatic equipment, and utilizes the time difference of arrival of acoustic wave sensor networks to quickly and accurately locate the leak point.

[0014] 2. By employing multi-condition training data construction techniques covering equipment startup, shutdown, and different load levels, and feature engineering techniques that encode voltage data as equipment operating status identifiers, the originally continuous voltage signal is transformed into operating condition features with clear physical meaning. The time-aligned multi-modal feature vector sequence eliminates the time mismatch problem of heterogeneous data. The Transformer model learns the temporal dependencies and cross-influence patterns of each modality feature through a self-attention mechanism. Through backpropagation optimization, the model parameters converge to the optimal state that can accurately characterize the mapping relationship between operating conditions and pressure loss. Therefore, it effectively solves the technical problems in existing technologies where fixed threshold discrimination logic cannot adapt to strong nonlinear and time-varying characteristics, and misjudges normal pressure fluctuations caused by equipment startup and shutdown as leakage. Thus, it achieves the technical effect that the theoretical pressure loss value can dynamically adapt to changes in actual operating conditions and accurately reflect the normal pressure loss under the current operating conditions.

[0015] 3. By employing a technique that dynamically calculates the sound wave propagation speed based on real-time pressure and temperature data using the gas state equation, the sound wave propagation speed parameter can adapt in real time to the fluctuating fluid pressure and temperature within the downhole compressed air pipeline, overcoming the influence of gas medium density and temperature changes on the sound speed. This real-time corrected sound wave propagation speed is applied to the entire process of leak point location calculation based on peak time difference. Therefore, it effectively solves the technical problem of existing technologies using fixed sound wave propagation speeds, which leads to positioning accuracy being affected by environmental factors, especially in deep well long-distance transportation scenarios, resulting in large positioning errors. Consequently, it achieves the technical effect of significantly reducing leak point positioning errors under conditions of large pressure and temperature fluctuations, and greatly improving positioning reliability and on-site maintenance targeting. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a coal mine compressed air pipeline monitoring and positioning method based on multimodal fusion in an embodiment of this application. Figure 2 This is a schematic diagram of a scenario for a coal mine compressed air pipeline monitoring and positioning method based on multimodal fusion in an embodiment of this application; Figure 3 This is a schematic diagram of time-refined extraction based on waveform analysis in an embodiment of this application; Figure 4 This is another flowchart illustrating the coal mine compressed air pipeline monitoring and positioning method based on multimodal fusion in the embodiments of this application; Figure 5 This application provides a schematic diagram of the physical device structure of a compressed air pipeline monitoring system. Detailed Implementation

[0017] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0018] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0019] To facilitate understanding, the following section will first introduce the relevant equipment information within the compressed air pipeline monitoring system of this application: System Composition: The compressed air duct monitoring system includes a sensor array deployed along the compressed air duct and a data processing unit. The sensor array includes multiple pressure sensors, multiple flow sensors, multiple acoustic sensors, and at least one voltage sensor connected to the compressed air equipment. Each sensor is communicatively connected to the data processing unit.

[0020] Sensor type and parameters: Pressure sensor: Collects pressure values ​​in the pipeline, in kPa, with a sampling frequency of 1 to 10 Hz, outputs low-frequency timing signals, with an accuracy of ±0.5% of full scale and a range of 0 to 200 kPa.

[0021] Flow sensor: Collects compressed air flow rate in the pipeline, in cubic meters per second, with a sampling frequency of 1 to 10 Hz, an accuracy of ±1% of full scale, and a range of 0 to 50 cubic meters per minute.

[0022] Acoustic wave sensor: Collects high-frequency acoustic wave signals generated by leakage, with the unit being volts corresponding to sound pressure. The sampling frequency is from 1 kHz to 10 kHz, and the output is a high-frequency timing signal with a sensitivity of -35 dB ± 2 dB.

[0023] Voltage sensor: Collects the power supply voltage of the air compressor in volts, with a sampling frequency of 0.1 to 1 Hz, outputs a low-frequency signal, has an accuracy of ±0.2% of full scale, and a range of 0 to 380 volts.

[0024] 3. Sensor deployment location: Pressure sensor: Installed at key nodes in the compressed air duct, including branch points, high-risk leakage areas such as joints and bends.

[0025] Flow sensor: Installed at the inlet of the main pipeline and the junction of branch pipelines, used to monitor overall and local flow.

[0026] Acoustic sensors: Distribute them evenly along the outside or inside of the pipe, with the spacing determined by the monitoring range, focusing on covering branch points and high-risk pipe sections.

[0027] Voltage sensor: Installed at the power input terminal of the air compressor to monitor power supply stability.

[0028] 4. Time synchronization mechanism: A high-precision time synchronization gateway is deployed in the underground central substation to connect all sensors via industrial Ethernet, ensuring that all sensor data have a unified timestamp with a timestamp synchronization accuracy of no more than 1 millisecond, providing a basis for subsequent multimodal data alignment and time difference of arrival calculation.

[0029] 5. Data transmission method: Sensor data is aggregated to the edge computing gateway via a bus with strong anti-interference capability. After the edge computing gateway preprocesses the data, it is uploaded to the ground center server or local edge server through the industrial network.

[0030] It should be noted that the specific types, quantities, performance parameters, and deployment locations of the aforementioned sensors are not fixed configurations, but can be flexibly adjusted according to the actual application scenario. For example, in simple pipe networks with short pipeline lengths and few branches, the number of sensors deployed can be appropriately reduced; in areas with complex geological conditions and high leakage risks, the sensor deployment density can be increased or the sensor accuracy level can be improved; in different underground coal mine environments, suitable sensor models can be selected based on factors such as pipeline material, transmission pressure, and ambient temperature and humidity. Therefore, this application does not impose restrictive requirements on the specific types, performance parameters, deployment locations, or quantities of sensors; the above embodiments are only described as preferred implementation methods.

[0031] The following describes the process of the method provided in this implementation. Please refer to [link / reference]. Figure 1This is a flowchart illustrating a coal mine compressed air pipeline monitoring and positioning method based on multimodal fusion in an embodiment of this application.

[0032] S101. Based on the preset pipeline topology diagram, map the sensors in the sensor set to the corresponding pipe segment spatial coordinates to form a spatial mapping distribution of equipment; The pipeline topology diagram refers to a pre-constructed spatial structure model of the underground compressed air pipeline network in a coal mine, including the coordinates of key locations such as the start point, end point, and branch nodes of the pipeline, as well as the connection relationships between each pipeline segment. The sensor set refers to all monitoring equipment deployed in the compressed air pipeline system, including pressure sensors, flow sensors, acoustic sensors, and voltage sensors. Pipeline segment spatial coordinates refer to the positional information of each pipeline segment in three-dimensional space, usually represented by the coordinates of the start and end points. The equipment spatial mapping distribution refers to the spatial distribution map formed after establishing a correspondence between the sensor equipment and the pipeline segments where they are installed.

[0033] During system initialization, the data processing unit loads a pre-stored pipeline topology diagram, which digitally describes the spatial layout of the entire compressed air pipeline network. The system parses the topology diagram into a graph structure of nodes and pipe segments, with each segment associated with its spatial coordinate range. Next, it acquires the installation location information of all sensors in the sensor set, including device number, sensor type, and installation coordinates. The system executes a spatial matching algorithm, comparing the coordinates of each sensor with the coordinates of the pipe segments, calculating the shortest distance from the sensor point to each pipe segment, selecting the segment with the smallest distance as the segment to which the sensor belongs, and recording the sensor's relative position on that segment. For example, if a pressure sensor is installed 30 meters from the starting point on the main pipe, the system records its mapping relationship as the segment number to which the sensor belongs, the relative position of 30 meters, and the corresponding three-dimensional coordinates. After all sensors are mapped, complete equipment spatial mapping distribution data is formed. This data is stored in tabular or database form, containing fields such as sensor number, pipe segment, relative position, and coordinates.

[0034] S102. Based on the spatial mapping distribution of the device, the area containing each sensor in the sensor set is determined as the distributed monitoring area; Among them, the distributed monitoring area refers to an independent monitoring unit consisting of a set of sensors and a specific pipe section, which has complete data acquisition and analysis capabilities.

[0035] After completing the sensor spatial mapping, the system begins the monitoring area division process. The data processing unit first parses the pipeline topology diagram, identifying all branch and confluence nodes. These nodes typically correspond to major branching points of the downhole compressed air pipeline, equipment access points, or boundaries between different work areas. The system defines the pipeline between two adjacent nodes as an independent logical segment and assigns a unique number to each logical segment. For example, the main pipeline from node A to node B is logical segment L01, from node B to node C is logical segment L02, and the branch pipeline from node B to node D is logical segment L03. For a specific logical segment, the system retrieves the sensors deployed near the starting and ending points of the segment from the equipment spatial mapping distribution data. The system identifies the pressure and flow sensors closest to the starting point of the segment as inlet boundary monitoring points and the pressure and flow sensors closest to the ending point as outlet boundary monitoring points. These four sensors constitute the complete set of boundary monitoring points for the logical segment.

[0036] The system further queries compressed air equipment deployment information, identifying all compressed air equipment connected to any location within the logical pipe segment, as well as compressed air equipment connected downstream of the outlet boundary monitoring point, and obtains a list of voltage sensors associated with these devices. Simultaneously, the system retrieves all acoustic sensors deployed along the path of the logical pipe segment, ensuring that the segment contains at least one acoustic sensor for leak detection. The system combines the identified boundary monitoring points, voltage sensors, acoustic sensors, and the logical pipe segment ranges they cover, defining a complete distributed monitoring area. The system creates a configuration record for this monitoring area, including the area number, the list of included sensors, the covered pipe segment range, the pipe segment length, and upstream and downstream connections. The system repeats this process until all logical pipe segments are divided into independent monitoring areas. This step achieves reasonable partitioning of the monitoring network, decomposing the complex pipeline system into multiple functionally independent monitoring units. Each unit is equipped with a complete sensor combination capable of independently performing pressure loss calculations and leak detection, improving the system's modularity and fault isolation capabilities, and providing a structured spatial framework for accurate monitoring and rapid location.

[0037] It should be noted that in some embodiments, the system can adopt a minimal configuration strategy when dividing the monitoring area. For a certain independent logical pipe segment, the system only needs to deploy one pressure sensor, one flow sensor, two acoustic sensors and one voltage sensor in the area to form a complete distributed monitoring area. These five sensors can be flexibly distributed at any location in the logical pipe segment, and do not have to be strictly located at both ends of the pipe segment. The pressure sensor and flow sensor can be deployed in a convenient installation location in the middle of the pipe segment to collect pressure and flow data in the area. The two acoustic sensors are deployed at an appropriate distance along the pipe segment path to form an acoustic monitoring array. The voltage sensor monitors the operating status of a certain air compressor in the area. This configuration provides monitoring data in four modes to meet the basic requirements of the Transformer model for multimodal input.

[0038] The subsequent system trains a dedicated Transformer model for this monitoring area. This model learns the correlation characteristics between various modal data and pressure loss within this small pipe segment, enabling it to precisely capture the fluid dynamics characteristics of this specific pipe segment. After collecting monitoring data, the system inputs it into the corresponding model for pressure loss prediction. The model output more closely matches the actual characteristics of the pipe segment, resulting in higher prediction accuracy. If a leak is detected, two acoustic sensors perform precise time-difference localization within a small area, with the localization error controllable within a few meters. The distributed monitoring area using a minimum sensor configuration has a small spatial coverage area, typically corresponding to pipe segments between 100 and 300 meters in length. When an abnormal pressure loss is detected, the system can immediately pinpoint the leak area to this small region, significantly narrowing the investigation scope. The dedicated Transformer model for this small area, optimized for specific pipe segments, can accurately identify subtle fluctuations in the normal operating conditions of the pipe segment, exhibiting higher sensitivity to abnormal pressure loss detection and enabling the discovery of smaller-scale leak events.

[0039] The zonal monitoring strategy system divides the entire pipeline network into dozens or even hundreds of independent logical pipe segments according to branch nodes. Each pipe segment is equipped with a minimum unit sensor combination to form an independent distributed monitoring area. Each area only needs to process the data within its own scope. The corresponding Transformer model is small in scale and has fast training and inference speed, which can complete pressure loss prediction in real time. It realizes the modular design and distributed computing of the monitoring system, improves the system's scalability, computing efficiency and fault location accuracy. Compared with the traditional large-scale monitoring method, the location error is reduced, and the emergency response efficiency and maintenance accuracy are significantly improved.

[0040] S103. Obtain monitoring data corresponding to the distributed monitoring area based on each sensor. The monitoring data includes pressure data, flow data, sound wave data, and voltage data. During normal system operation, the data processing unit continuously collects monitoring data from sensors in each distributed monitoring area. Each sensor periodically measures the corresponding physical quantity according to its set sampling frequency and generates a data packet, which includes the sensor number, measured value, timestamp, and data quality identifier. Pressure and flow sensors, as low-frequency sensors, collect 1 to 10 data points per second, reflecting the changing trends of pipeline pressure and flow. Acoustic sensors, as high-frequency sensors, collect thousands of data points per second, capable of capturing instantaneous high-frequency vibration signals generated by leaks. Voltage sensors monitor the power supply status of the compressed air equipment; although the sampling frequency is low, it can reflect the start-up, shutdown, and load changes of the equipment.

[0041] All sensors transmit data to the data processing unit via industrial bus or wireless network. The system employs a high-precision time synchronization protocol to ensure all data has a unified timestamp, with a time synchronization accuracy within 1 millisecond. After receiving the data, the data processing unit performs preliminary verification, eliminating outliers and communication errors, and storing valid data categorized by monitoring area. This step enables real-time acquisition and aggregation of multimodal sensor data, providing raw input for subsequent data fusion analysis and ensuring the system can comprehensively perceive the operational status of each monitoring area.

[0042] S104. After aligning the monitoring data with time, input it into the Transformer model that matches the distributed monitoring area to obtain the theoretical pressure loss value. Time alignment refers to the process of unifying multimodal sensor data with different sampling frequencies to the same time base and sampling interval. The Transformer model is a deep learning model based on a self-attention mechanism, capable of learning the complex relationships between multimodal data and predicting pressure loss. The theoretical pressure loss value refers to the expected pressure loss calculated based on parameters such as current flow rate, pressure, and equipment operating status under normal, leak-free pipeline conditions. This pressure loss is the energy loss that inevitably occurs during the normal operation of the compressed air pipeline system, mainly caused by pipeline friction resistance, local resistance, and equipment gas consumption, and serves as a benchmark reference value for judging whether abnormal leaks exist.

[0043] Before this step, the Transformer model can be trained in advance. The specific training process is as follows: Acquire monitoring data collected by various sensors within the distributed monitoring area under multiple operating conditions as initial training data. These multiple operating conditions include a first operating condition, a second operating condition, and a third operating condition. The first operating condition includes the first monitoring data and actual pressure loss value corresponding to the equipment operating status being "on". The second operating condition includes the second monitoring data and actual pressure loss value corresponding to the equipment operating status being "off". The third operating condition includes the third monitoring data and actual pressure loss value corresponding to different load levels of the equipment operating status. Time-align the initial training data to construct a multimodal feature vector sequence. The voltage data in this multimodal feature vector sequence is encoded as the equipment operating status identifier, which represents the start / stop status and load level of the compressed air equipment within the distributed monitoring area. Input this multimodal feature vector sequence into the Transformer model for training. Optimize the model parameters through backpropagation to obtain the trained Transformer model. This trained Transformer model is used to predict pressure loss values ​​based on the input monitoring data.

[0044] The initial training data refers to the raw time-series data collected from various sensors within the distributed monitoring area, including pressure sensor data, flow sensor data, acoustic sensor data, and voltage sensor data, along with the actual pressure loss value at the corresponding time point as a supervisory label. Multiple operating conditions refer to the working scenarios of the compressed air pipeline network under different operating states, covering typical operating modes such as equipment start-up and shutdown, and load changes, used to train the model to learn the pressure loss patterns under different scenarios. The first operating condition refers to the condition where the compressed air equipment within the distributed monitoring area is in the on-state, where the equipment normally consumes compressed air, resulting in a large flow rate and high pressure loss in the pipeline. The second operating condition refers to the condition where the compressed air equipment is in the off-state, where the equipment does not consume compressed air, resulting in a small flow rate and low pressure loss in the pipeline. The third operating condition refers to the condition where the compressed air equipment operates under different load levels, where the load level reflects the equipment's air consumption intensity, and different load levels correspond to different flow demands and pressure loss characteristics. The actual pressure loss value refers to the pressure difference between the inlet and outlet of the pipe section measured by the pressure sensors at the boundary monitoring points, serving as the true label value for model training. Time alignment refers to synchronizing and registering monitoring data from different sensors according to their acquisition timestamps, ensuring that multimodal data at the same moment correspond to the same feature vector. Multimodal feature vector sequence refers to combining time-aligned pressure, flow, sound, and voltage data into a unified vector representation, forming a sequence of data arranged in chronological order. Equipment operating status identification refers to encoding continuous voltage values ​​collected by voltage sensors into discrete status labels, such as on, off, high load, medium load, and low load, facilitating the model's understanding of the equipment's operating mode. The Transformer model is a deep learning model based on a self-attention mechanism, capable of capturing long-range dependencies and interactive relationships between multimodal features in sequential data. Backpropagation refers to the training process of updating model parameters using a gradient descent algorithm based on the error between the model's predicted values ​​and actual pressure loss values, gradually improving the model's prediction accuracy.

[0045] More specifically, the system first extracts all sensor monitoring data for a specified distributed monitoring area within a historical time period from the data acquisition system, including the pressure time series recorded by the pressure sensor, the flow time series recorded by the flow sensor, the sound wave signal time series recorded by the two sound wave sensors, and the voltage time series recorded by the voltage sensor. At the same time, the actual pressure loss time series measured by the monitoring points at the boundary of the area is extracted as tag data.

[0046] The system categorizes historical data by operating condition, identifying the periods when the equipment is fully operational as the first operating condition data. During these periods, voltage sensors show a stable supply voltage near the rated value, flow sensors show a large gas flow rate, and pressure loss is high. The system extracts all sensor data and corresponding actual pressure loss values ​​for these periods to form the first monitoring dataset. The system identifies the periods when the equipment is completely off as the second operating condition data. During these periods, voltage sensors show voltage close to zero or the equipment is in standby mode, flow sensors show extremely low or zero flow, and pressure loss is close to zero. The system extracts data for these periods to form the second monitoring dataset. The system further identifies the periods when the equipment operates under different load levels as the third operating condition data. By analyzing the fluctuation patterns of voltage sensor data and the magnitude of flow sensor data, the system classifies the equipment operating status into multiple levels such as high load, medium load, and low load. High load corresponds to high voltage and flow with large fluctuations; medium load corresponds to medium voltage and flow; and low load corresponds to low voltage and flow but the equipment is still operating. The system extracts the time period data corresponding to each load level to form the third monitoring dataset.

[0047] The system performs time alignment processing on all extracted monitoring data. Since different sensors may have different sampling frequencies or clock deviations, the system uses timestamp interpolation and resampling techniques to unify all sensor data onto the same time base, ensuring accurate correspondence between pressure, flow, sound, and voltage data at the same moment. The system performs feature encoding on voltage sensor data, converting it into equipment operating status identifiers based on the magnitude and change pattern of the voltage value. For example, voltage below threshold V1 is encoded as "off," voltage between V1 and V2 as "low load," voltage between V2 and V3 as "medium load," and voltage above V3 as "high load" or "on." This encoding process converts continuous voltage signals into discrete category labels, enabling the model to clearly identify the equipment's operating state. The system combines the time-aligned pressure data, flow data, sound characteristic data (such as sound energy and spectral characteristics), and equipment operating status identifiers into a unified feature vector. Each time step corresponds to one feature vector, and the feature vectors from multiple consecutive time steps constitute a multimodal feature vector sequence.

[0048] The system takes a pre-constructed multimodal feature vector sequence as input and the corresponding actual pressure loss value as the output label, feeding them into a Transformer model for training. The Transformer model learns the dependencies between different time steps in the feature vector sequence and the interaction patterns between different modal features through a multi-head self-attention mechanism. The encoder encodes the input sequence into a high-dimensional semantic representation, and the decoder or regression head outputs the predicted pressure loss value. The system calculates the mean squared error between the model's predicted value and the actual pressure loss value as the loss function, and uses the backpropagation algorithm to calculate the gradient of the loss function with respect to the model parameters. An optimizer such as Adam is used to update the model parameters. After multiple training epochs, the model parameters gradually converge, and the prediction error gradually decreases. After training, the system validates and tests the model, using an independent validation dataset to evaluate the model's generalization performance, ensuring that the model can accurately predict pressure loss values ​​on unseen data. Finally, the trained Transformer model is obtained and deployed to the corresponding distributed monitoring area for real-time pressure loss prediction. This model can quickly calculate the theoretical pressure loss value based on real-time collected monitoring data, providing an accurate benchmark reference for subsequent leak detection.

[0049] After determining the pre-trained Transformer model, during this step, the system collects real-time monitoring data from various sensors within the distributed monitoring area. This includes pressure values ​​measured by pressure sensors, flow values ​​measured by flow sensors, sound wave signals collected by acoustic sensors, and voltage values ​​measured by voltage sensors. Since the sampling frequencies of different sensors may differ (e.g., pressure sensors sample 10 times per second while flow sensors sample 5 times per second), or there may be slight clock deviations between sensors, the system needs to perform time alignment processing on this data. The system uses timestamp matching and interpolation techniques to resample all sensor data to a unified time base, for example, to 10 sampling points per second. For sensor data with lower sampling frequencies, linear interpolation or spline interpolation is used to supplement intermediate time values, ensuring that data at the same time index position comes from measurements taken by all sensors at the same physical moment. The system encodes the voltage sensor data into status codes, determining the equipment's operating status based on the current voltage value and encoding it as on, off, or a specific load level identifier.

[0050] The system combines time-aligned pressure data, flow data, acoustic feature data, and equipment operating status identifiers into a multimodal feature vector, which is then normalized and feature-engineered according to the input format used during training. The system inputs the constructed feature vector sequence into the Transformer model corresponding to the distributed monitoring area. The model's encoder extracts the spatiotemporal dependencies in the feature vector sequence through a multi-head self-attention mechanism, capturing the interaction patterns between different modal data and the influence of historical data on the current state. The decoder or regression head outputs a numerical value as the prediction result. This prediction result is the theoretical pressure loss value, representing the pressure loss that should occur under the current flow rate, pressure, and equipment operating status, assuming the pipeline system is completely normal and leak-free. This value comprehensively considers the influence of factors such as pipeline friction resistance, local resistance, and equipment gas consumption on pressure. For example, when the flow rate is 100 m³ / min and a certain equipment is operating at high load, the model predicts a theoretical pressure loss of 45 kPa, while when the flow rate drops to 50 m³ / min and the equipment is operating at low load, the theoretical pressure loss may only be 15 kPa. The system temporarily stores the theoretical pressure loss value for later comparison with the actual pressure loss value, thereby determining whether there is an anomaly in the pipeline system.

[0051] It should be noted that even when compressed air pipelines are operating under completely normal and leak-free conditions, pressure losses will still occur due to the physical characteristics of gas flow. Specifically, these losses include: the energy consumed by friction between compressed air and the pipe wall as it flows through the pipeline, leading to a decrease in pressure along the pipe; local pressure losses caused by changes in flow direction and turbulence when the airflow passes through fittings such as elbows, tees, and valves; and changes in flow rate and pressure reduction caused by compressed air consumption by air-using equipment such as pneumatic tools and cylinders within the distributed monitoring area. The magnitude of these normal pressure losses is affected by various factors such as pipeline length, pipe diameter, roughness, flow rate, and equipment load, and changes dynamically with operating conditions. For example, when the number of air-using equipment increases or the load increases, the flow rate increases, and the frictional resistance increases non-linearly, leading to a significant increase in pressure loss. Therefore, it is not possible to simply set a fixed normal pressure loss value. It is necessary to use a Transformer model to dynamically predict the theoretical pressure loss that should be present based on real-time operating parameters. Only by comparing the actual measured pressure loss with this theoretical value can abnormal losses exceeding the normal range be identified, thereby determining whether a leak exists.

[0052] S105. After obtaining the actual pressure loss value under the current operating conditions, calculate the current pressure loss difference between the actual pressure loss value and the theoretical pressure loss value. The actual pressure loss value refers to the pressure difference between the inlet and outlet of the pipe section measured by pressure sensors at the boundary monitoring points of the distributed monitoring area. It reflects the true pressure loss of the pipe section at the current moment. This actual pressure loss includes normal pipe friction loss and equipment gas consumption loss. If there is a leak, it will also include the additional pressure loss caused by the leak. The current pressure loss difference refers to the difference between the actual pressure loss value and the theoretical pressure loss value. This difference is used to characterize the degree of abnormality of the pipeline system. Under normal leak-free conditions, this difference should be close to zero or fluctuate within a small range, reflecting model prediction errors and measurement noise. If the difference is significantly large, it indicates that the actual pressure loss exceeds the level that should be under normal operating conditions, which may be due to a leak causing additional gas loss and increasing pressure drop, or other faults such as pipe blockage.

[0053] It is important to emphasize that the absolute value of the actual pressure loss alone cannot determine whether a leak exists. This is because normal pressure loss varies dynamically with operating conditions. For example, a pipe section with an actual pressure loss of 20 kPa under low load is considered normal, but under high load, the actual pressure loss may reach 60 kPa and still be considered normal. If a fixed threshold such as 40 kPa is simply set to judge a leak, a large number of false alarms will be generated under high load conditions, while leaks may be missed under low load conditions. Therefore, it is necessary to identify anomalies by comparing the difference between the actual value and the theoretical value. Only when the actual pressure loss significantly exceeds the theoretical value predicted by the model based on the current operating conditions can it be confirmed that there is an additional loss beyond the normal range, thereby accurately determining a leak.

[0054] The system acquires real-time pressure measurement data from boundary monitoring points within the distributed monitoring area. These boundary monitoring points are typically located at the inlet and outlet of the pipe section, each equipped with pressure sensors to measure the pressure at both ends. The system reads the pressure value Pin measured by the inlet pressure sensor and the pressure value Pout measured by the outlet pressure sensor, and calculates the difference between the two to obtain the actual pressure loss value ΔPactual = Pin - Pout. For example, if the inlet pressure is 0.7 MPa and the outlet pressure is 0.64 MPa, the actual pressure loss is 60 kPa. This value reflects the actual pressure drop that occurs when gas flows through the pipe section, including friction losses due to pipe friction, local resistance losses from elbows and tees, and pressure drops caused by gas consumption by equipment within the pipe section. Under normal, leak-free conditions, the actual pressure loss should be basically consistent with the theoretical pressure loss value; the difference between the two is only due to model prediction errors and sensor measurement noise, typically fluctuating within a few kPa. If there is a leak in the pipe section, the leak hole will consume additional compressed air, resulting in an increase in the total flow rate. At the same time, the pressure downstream of the leak point drops faster, making the pressure difference between the inlet and outlet, i.e., the actual pressure loss, significantly greater than the theoretical expected value. For example, if the theoretical pressure loss is 45 kPa but the actual measured pressure is 60 kPa, then the difference of 15 kPa indicates that there is an abnormal additional loss.

[0055] The system compares the actual pressure loss value with the theoretical pressure loss value predicted by the Transformer model in step S104, calculating the current pressure loss difference. For example, if the actual value is 60 kPa and the theoretical value is 45 kPa, the difference is 15 kPa. The magnitude and sign of this difference are of significant diagnostic importance. Under normal circumstances, the difference should fluctuate within a small range such as ±5 kPa, reflecting the impact of model prediction errors and measurement noise. If the difference is consistently positive and large, such as exceeding 10 kPa, it indicates that the actual pressure loss exceeds normal expectations, possibly due to leakage causing additional gas loss and increased pressure drop. If the difference is negative and has a large absolute value, it may indicate measurement abnormalities or other faults such as reduced flow and decreased pressure loss caused by partial blockage in the pipeline.

[0056] The system records and analyzes the calculated current pressure loss difference to provide a basis for subsequent leakage judgment. By comparing the actual loss with the dynamically predicted theoretical loss, the system can accurately identify anomalies under various operating conditions, avoiding the false alarms and missed detections of the fixed threshold method under different operating conditions.

[0057] S106. If the current pressure loss difference is greater than the preset pressure loss threshold, a leak is determined, and the leak location process is initiated: The preset pressure loss threshold is a discrimination standard determined based on the pipeline characteristics, equipment configuration, and historical operating data statistics of the distributed monitoring area. It is used to distinguish between normal pressure loss fluctuations and pressure loss increases caused by abnormal leakage. The setting of this threshold needs to take into account the model prediction accuracy, sensor measurement error, and the acceptable minimum leakage detection sensitivity.

[0058] The system compares the current pressure loss difference ΔPdiff calculated in step S105 with the preset pressure loss threshold ΔPthreshold. This threshold is determined through statistical analysis during system design or operation, and is typically set to three times the standard deviation of the normal pressure loss difference fluctuation range. Alternatively, it can be set according to safety and economic requirements to detect the pressure loss increment corresponding to the minimum amount of leakage of concern. For example, if the standard deviation of the pressure loss difference in the monitored area under normal leak-free conditions is 2 kPa, the threshold can be set to 6 kPa. Or, according to safety requirements, it can be set to detect the pressure loss change corresponding to 1% of the total flow rate of the pipe section, such as 8 kPa. It should be noted that since the theoretical pressure loss value already considers the influence of the current operating conditions, the pressure loss difference eliminates the interference of changes in operating conditions. Therefore, this threshold can be set to a fixed value and does not need to be adjusted with operating conditions. This is the fundamental difference from directly setting the threshold for the actual pressure loss value, ensuring that leakage detection maintains stable detection sensitivity and a low false alarm rate under various operating conditions.

[0059] The system performs conditional judgment. If ΔPdiff > ΔPthreshold, it indicates that the current actual pressure loss significantly exceeds the theoretical expectation, and the degree of excess has exceeded the normal fluctuation range. For example, if the difference reaches 15 kPa while the threshold is 8 kPa, the system determines that there may be a leakage fault in the distributed monitoring area.

[0060] To avoid false alarms caused by momentary interference or measurement noise, the system can employ a continuous judgment mechanism, requiring the pressure loss difference to exceed a threshold for multiple consecutive sampling periods before confirming a leak. For example, a leak alarm can only be triggered if the pressure loss difference exceeds the threshold for 5 consecutive seconds. Alternatively, a sliding window statistical approach can be used, confirming a leak when 8 out of the most recent 10 sampling points exceed the threshold. Once a leak is confirmed, the system immediately initiates the leak location process, which primarily relies on the leak acoustic signal collected by the acoustic sensor for location calculation. The system sends a leak alarm message to the operator, indicating the specific distributed monitoring area number, leak detection time, current pressure loss difference, and other key information. Simultaneously, it activates the acoustic sensors in that area to enter high-frequency sampling mode, increasing the sampling rate to obtain high-quality leak acoustic data, providing data support for subsequent precise location.

[0061] When a leak occurs in the compressed air duct, a specific high-frequency acoustic signal is generated, typically ranging from 500Hz to 5kHz. This acoustic signal propagates forward along the duct at a certain speed. Once a leak occurs at a point in the duct, the generated acoustic signal is simultaneously captured by multiple acoustic sensors arranged along the duct. Due to the different distances between each acoustic sensor and the leak point, the arrival time of the acoustic signal at each sensor will also differ; this time difference is denoted as time difference Δt. To accurately determine the location of the leak, a high-precision time synchronization protocol is used to precisely timestamp the acoustic signal captured by each acoustic sensor. By calculating the time difference Δt between the acoustic signals captured by different acoustic sensors and combining it with the known propagation speed v of sound waves within the duct, the location of the leak point in the duct can be accurately calculated using geometric relationships. The specific implementation method can be found in steps S107-S108, and will not be elaborated further here.

[0062] S107. Determine the peak values ​​and peak time differences of the acoustic signal based on the first acoustic sensor and at least one adjacent second acoustic sensor within the distributed monitoring area. The first acoustic sensor refers to one of the acoustic sensors deployed within the distributed monitoring area, used to receive acoustic signals generated by the leak point. The second acoustic sensor refers to one or more other acoustic sensors deployed adjacent to the first acoustic sensor. These sensors, together with the first acoustic sensor, constitute an acoustic monitoring array used to calculate the location of the leak point using a time-difference positioning method.

[0063] Upon detecting a pressure loss difference exceeding a threshold, the system immediately records this detection moment as the pressure anomaly trigger moment. A target time sliding window of 10 seconds is constructed, extending 5 seconds forward and 5 seconds backward from this moment. The system sends data extraction commands to all acoustic sensors within the monitoring area, retrieving raw acoustic waveform data segments within the target window range from each sensor's local buffer. Each sensor returns a sequence of amplitude values ​​collected at 10,000 sampling points per second during that time period. Each data point is accompanied by a timestamp with microsecond-level precision, and a precise clock synchronization protocol ensures a unified time stamp across different sensors.

[0064] The system preprocesses the acquired waveform data segments, applying a bandpass filter to retain leakage characteristic sound waves in the frequency range of 500 Hz to 5 kHz, while filtering out low-frequency mechanical vibration noise and high-frequency electromagnetic interference. The system then normalizes the filtered waveforms, unifying the signal amplitudes from different sensors to between 0 and 1, eliminating the effects of sensor sensitivity differences and sound wave propagation attenuation, and ensuring that waveform matching focuses on morphological characteristics rather than amplitude magnitude.

[0065] The system executes a timing comparison process, using the waveform data from the first acoustic sensor as a reference signal and sequentially calculating the similarity with the waveform data from the second acoustic sensor, etc. Since the sound waves from the same leakage source arrive at different sensors at different times, the waveforms are offset on the time axis. The system uses a sliding cross-correlation algorithm to gradually shift the waveforms to be compared, shifting by one sampling point interval each time. The correlation coefficient between the two waveforms after shifting is calculated, defined as the sum of the products of corresponding points of the two signals divided by the square root of their respective energy. The system finds the maximum value of the correlation coefficient and its corresponding time offset; the maximum correlation coefficient represents the morphological feature similarity. The system compares the calculated similarity with a preset threshold of 0.75. If the similarity is greater than the threshold, the two waveforms are determined to originate from the same leakage source, forming a correlated wave, and the time offset is recorded.

[0066] If the system fails to identify associated waves with a similarity exceeding the threshold, it indicates that the collected signals are mainly background noise or interference from different sources, making it impossible to confirm the same leak source. The system then terminates the localization process and generates an anomaly report indicating the need for manual inspection. If associated waves are successfully identified, for example, if the waveform similarity between the first and second sensors reaches 0.82, or the similarity between the first and third sensors reaches 0.78, both exceeding the threshold, the system determines that these sensors have captured valid sound waves from the same leak source, confirming the existence of a leak event.

[0067] The system performs multi-scale peak detection on the identified correlated waves, employing an automatic multi-scale peak monitoring algorithm to identify peaks in the correlated wave signals from each sensor. This algorithm sets a series of scale parameters, searches for local maxima in the signal at each scale, marks these points as 1 to form a local maxima vector for that scale, and accumulates the vectors across all scales. Since true peaks exhibit local maxima at multiple scales, the accumulated value is large, while noise and spurious peaks have small accumulated values. The system sets an accumulation threshold to select candidate peaks, filtering based on peak amplitude and spacing, ultimately identifying the main peak points. The system applies this algorithm to the correlated waves from the first sensor, identifying the peak with the largest amplitude as the correlated wave peak, and recording the timestamp of the corresponding data point as the arrival time of the sound wave from the first sensor. Peak detection is similarly performed on the second sensor and others to obtain the arrival times of the sound waves from each sensor, forming a timestamp sequence.

[0068] The system compares the timestamps of each sensor and identifies the sensor with the smallest timestamp, indicating that it detected the wave peak first. This sensor is closest to the leak point and is selected as the reference node. Using the arrival time of the reference node as a benchmark, the system calculates the time differences of other sensors relative to it. For example, the time difference for the second sensor is equal to its arrival time minus the arrival time of the reference node, resulting in a set of peak time difference data. The system then transmits these time differences to the subsequent location calculation module. Combining the distance between sensors and the speed of sound propagation, the system uses geometric relationships to accurately calculate the location of the leak point, achieving precise localization.

[0069] S108. The location of the leak point is determined by combining the peak time difference, the distance between the first acoustic sensor and the second acoustic sensor, and the sound wave propagation speed.

[0070] The system reads the actual length of the pipe section between the first and second acoustic sensors from the configuration database of the pipeline monitoring system. Assuming the first sensor is installed at a certain point in the pipeline and the second sensor is installed 200 meters downstream of the first sensor, the actual length of the pipe section is 200 meters. The system also reads the propagation speed of sound waves within the pipeline from the configuration database. The propagation speed of sound waves is affected by factors such as the pipe material, diameter, wall thickness, and pressure of the medium within the pipe. For steel compressed air ducts, the propagation speed of sound waves is typically between 3000 and 5000 meters per second. The system determines the propagation speed of sound waves based on the actual pipeline parameters and historical calibration data; assuming the propagation speed of sound waves in this pipeline is 4000 meters per second.

[0071] The system divides the actual length of the pipe section by the speed of sound propagation to calculate the standard sound path time. Specifically, 200 meters divided by 4000 meters per second equals 0.05 seconds, or 50 milliseconds. This standard sound path time represents the theoretical time required for a sound wave to travel from the location of the first sensor to the location of the second sensor, and is a key reference benchmark for determining the location of the leak source.

[0072] The system compares the acoustic arrival time difference calculated in the previous steps with the standard acoustic path time. If the acoustic arrival time difference equals the standard acoustic path time (e.g., a time difference of 50 milliseconds equals the standard acoustic path time of 50 milliseconds), it indicates that the time for the leaking acoustic wave to travel from the leak point to the first sensor is zero, meaning the leak point is exactly located at the first sensor. Alternatively, the path of the leaking acoustic wave from the leak point to the second sensor is exactly equal to the pipe length between the two sensors. In this case, the leak source is actually located in the pipe segment on the other side of the boundary with the first acoustic sensor. The system determines that the leak source is not in the current logical pipe segment between the first and second sensors, but rather in a new logical pipe segment adjacent to the current pipe segment with the first sensor as its endpoint. This new logical pipe segment may be a pipe segment upstream of the first sensor or a branch pipe segment.

[0073] The system identifies a new logical pipe segment adjacent to the first sensor based on the pipeline topology. This new pipe segment is defined by the boundary between the first sensor and another acoustic sensor. The system sets this other sensor as the new second acoustic sensor, re-extracts the acoustic data from the new second sensor and the first sensor, and repeats the associated wave identification, peak detection and time difference calculation process in the previous steps to obtain a new acoustic arrival time difference. The system then executes the judgment logic of this step again to determine the specific location of the leak source in the new pipe segment, thus achieving accurate location of the leak point.

[0074] If the arrival time difference of the sound wave is less than the standard sound path time, for example, a time difference of 30 milliseconds is less than the standard sound path time of 50 milliseconds, it indicates that the difference in path length of the leaking sound wave from the leak point to the first and second sensors is less than the length of the pipe section between the two sensors. According to the geometric relationship of sound wave propagation, this situation can only occur when the leak point is located within the pipe section between the first and second sensors. The system determines that the leak source is located within the target logical pipe section between the first and second sound wave sensors. The starting point of this target pipe section is the location of the first sensor, the ending point is the location of the second sensor, and the length is the actual length of the pipe section, 200 meters.

[0075] The system further checks for pipe branches within the target logical pipe segment. It reads the segment's structural information from the pipe topology database, checking for tees, crosses, or other branch connection points. If no branches exist within the target logical pipe segment, it indicates that the pipe in that section is a single straight line or a simple bend, and the leak point must be located on the main pipe between the first and second sensors. The system can directly apply the time-difference positioning formula to calculate the leak location. The method for calculating the leak location in this case can be found here. Figure 2 A schematic diagram, If a branch exists, proceed with steps S301-S307.

[0076] In some embodiments, during the actual operation of coal mine compressed air pipelines, the gas pressure and temperature inside the pipeline fluctuate with changes in operating conditions, and the propagation speed of sound waves in the gas medium is closely related to parameters such as gas pressure, temperature, and density. If a fixed sound velocity value is used for leak location calculation, a large location error will occur when the actual operating conditions deviate from the standard operating conditions. Therefore, the system acquires in real time the fluid pressure values ​​inside the pipe collected by each pressure sensor and the temperature values ​​inside the pipe collected by the temperature sensor within the distributed monitoring area. Based on a preset gas state equation, the real-time collected pressure and temperature values ​​are substituted into the equation, and combined with the physical characteristic parameters of the gas medium, the actual propagation speed of sound waves in the gas medium under the current operating conditions is calculated. The preset gas state equation is a mathematical model pre-established based on gas dynamics theory and acoustic propagation principles. Specifically, this gas state equation is derived based on the ideal gas law or the corrected equation for actual gases. For compressed air, the sound wave propagation speed has a definite functional relationship with physical parameters such as the absolute temperature, specific heat ratio, and gas constant of the gas. The system, based on the compositional characteristics of the fluid medium (mainly compressed air) in the coal mine compressed air pipeline, retrieves fundamental parameters such as the specific heat ratio and gas constant from standard gas physical parameter tables. Combining this with the theory of sound wave propagation in compressible fluids, it establishes a correlation equation between sound velocity and pressure and temperature. This preset equation is built into the calculation module during system initialization, serving as the theoretical basis for sound velocity calculation. Once the system acquires the real-time pressure and temperature values ​​of the fluid inside the pipeline, it directly substitutes these measured parameters into the preset gas state equation to quickly calculate the actual sound wave propagation speed under the current operating conditions.

[0077] By calculating the sound wave propagation speed under current operating conditions in real time, the system can adaptively adjust the sound velocity parameter in the positioning algorithm and use the actual sound wave propagation speed for subsequent time-difference positioning calculations, ensuring the accuracy of leak location calculations. This dynamic sound velocity correction mechanism effectively eliminates positioning errors caused by operating conditions such as pressure fluctuations and temperature changes, improving the adaptability and positioning accuracy of the leak detection system to complex downhole environments.

[0078] In the above embodiment, by monitoring the pressure data of each node in the pipeline in real time through a pressure sensor array and calculating the pressure loss difference, when the pressure loss difference exceeds a preset threshold, the acoustic sensor is immediately triggered to accurately locate the leak point. Combined with time-series comparison to identify associated waves, multi-scale peak detection to extract the arrival time difference of acoustic waves, and logical judgment based on standard sound path time to determine the pipe section where the leak source is located, the leak event can be quickly and accurately identified and the leak point can be precisely located in the early stage of the leak. This avoids the problems of traditional methods where pressure monitoring alone cannot accurately locate the leak or acoustic monitoring is easily affected by noise interference, leading to misjudgment. It effectively solves the technical problems of low positioning accuracy, slow response speed, large noise interference, and inability to distinguish between real leaks and equipment disturbances in coal mine compressed air pipeline leak detection. Thus, it realizes early detection, rapid response, and accurate location of coal mine compressed air pipeline leaks, significantly improving the safety and reliability of coal mine compressed air systems, reducing energy waste and safety hazards caused by leaks, and providing strong technical support for safe coal mine production.

[0079] Please refer to the following: Figure 2 This is a schematic diagram of a scenario for the monitoring and positioning method of coal mine compressed air pipeline based on multimodal fusion in the embodiments of this application: exist Figure 2 The diagram illustrates the basic principle of locating leaks in coal mine compressed air pipelines. The installation locations of two acoustic sensors, Sensor 1 and Sensor 2, are marked on the diagram. They are located at opposite ends of a pipeline segment, separated by the actual length L of the pipeline. When the leak point is located between Sensor 1 and Sensor 2, the sound waves generated by the leak propagate in both directions simultaneously and are captured by both sensors. Due to the different distances of the leak point from the two sensors, there is a time difference Δt between the arrival times of the sound waves at the two sensors, i.e., the peak time difference.

[0080] Assuming the distance from the leak point to sensor 1 is *a*, then the distance from the leak point to sensor 2 is *La*. The time required for the sound wave to travel from the leak point to sensor 1 is *a* / *v*, and the time required for the sound wave to travel from the leak point to sensor 2 is *(La)* / *v*, where *v* is the speed of sound in the pipe. The time difference Δt between the two sensors detecting the leaking sound wave can be expressed as: Δt=(La) / va / v=(L-2a) / v By solving for the distance 'a' from the leak point to sensor 1 using this equation, we can obtain the following formula: a=[L-(Δt×v)] / 2 Where L is the actual length of the pipe section, Δt is the peak time difference, and v is the speed of sound. For example, when the actual length of the pipe section L is 340 meters, the peak time difference Δt is 0.125 seconds, and the speed of sound v is 1500 meters per second, substituting these values ​​into the formula yields: a = [340 - (0.125 × 1500)] / 2 = (340 - 187.5) / 2 = 76.25 meters, meaning the distance from the leak point to sensor 1 is 76.25 meters.

[0081] The system uses this time-difference positioning algorithm, combined with the actual measured peak time difference, the actual length of the pipeline section, and the speed of sound propagation, to accurately calculate the specific location of the leak point on the pipeline. Here, sensor 1 and sensor 2 both refer to acoustic wave sensors.

[0082] exist Figure 2 During implementation, most acoustic sensors on the market send data packets at a fixed period. For example, if the sensor sends 8 data packets per second, the data acquisition interval is 125 milliseconds. This discrete sampling method introduces time errors, resulting in limited positioning accuracy. Under the condition that the speed of sound is about 340 meters per second, the spatial error caused by this time interval is about 0.125 × 340 = 42.5 meters, which is difficult to meet the requirements of high-precision positioning.

[0083] To overcome this limitation, this application employs a refined time extraction strategy based on waveform analysis. Specifically, a peak detection algorithm is used to detect the acoustic signal waveform in real time, accurately capturing the arrival time of the leaked acoustic wave peak. The offset of the peak point within the time window is superimposed on the data packet timestamp to achieve precise acoustic wave arrival time extraction. (See below.) Figure 3 As shown: Please see Figure 3 This is a schematic diagram of time-refined extraction based on waveform analysis in an embodiment of this application.

[0084] exist Figure 3 In the figure, the horizontal axis represents time (unit: ms), and the vertical axis represents the amplitude of the acoustic signal. The figure shows the acoustic waveforms corresponding to two consecutive data packets: the time window of the first data packet starts at 0ms, and the time window of the second data packet starts at 125ms (corresponding to a sampling interval of 8 data packets per second). Within the time window of the data packet starting at 0ms, the acoustic signal shows a leakage energy peak at 63ms (i.e., the "63ms" peak marked in the figure); this peak is not exactly at the timestamp node of the data packet (0ms or 125ms), but is located within the time window of 0ms-125ms. By capturing the specific time offset of this peak (63ms relative to the data packet start time of 0ms) using a peak detection algorithm, the arrival time of the acoustic wave can be accurately determined to be 63ms based on the 125ms discrete interval of the original data packet, instead of simply using 0ms or 125ms as the arrival time.

[0085] This method breaks through the limitation of the 125ms sampling interval: originally, time could only be estimated in units of 125ms, but now the precise peak moment in the data packet can be extracted, reducing the time error from 125ms to the millisecond level or even better. The corresponding spatial error is also greatly reduced from 42.5 meters (if the time accuracy is improved to 1ms, the spatial error is only 0.34 meters), thereby achieving high-precision positioning of the leak point.

[0086] Based on the above, the following is a more detailed description of the process provided in this implementation. Please refer to [link / reference]. Figure 4 This is another flowchart illustrating the coal mine compressed air pipeline monitoring and positioning method based on multimodal fusion in this application embodiment.

[0087] S401. If there is at least one branch point in the target logic tube segment, determine whether the location of the leak point is within the preset range of the branch point. After determining in step S108 that the leak source is located within the target logical pipe segment, the system immediately reads the structural information of the target pipe segment from the pipe topology database and checks whether there are any branch points within the pipe segment. The system traverses all pipe nodes of the target pipe segment, identifies branch connection points of tee or cross type, and counts the number of branch points. If there is at least one branch point within the target logical pipe segment, the system initiates the branch determination process, because the existence of branch points will cause multiple possible paths for sound wave propagation, and the simple time difference positioning formula cannot accurately calculate the location of the leak point.

[0088] The system reads the leak point location initially calculated in step S108. This location is calculated using a time-difference positioning algorithm under the assumption of no branches within the pipe section. The system compares this initially calculated location with the locations of each branch point within the target pipe section, calculating the distance between the leak point location and each branch point. The system sets a preset range of 8 meters before and after each branch point, forming a 16-meter-long determination area centered on the branch point. If the initially calculated leak point location falls within this preset range, for example, if the leak point is only 5 meters away from a branch point, the system determines that the leak point may be located near a branch point or on a branch pipeline, requiring further precise positioning using acoustic sensors on the branch pipeline. If the leak point location is not within the preset range of any branch point, for example, if it is more than 20 meters away from the nearest branch point, the system determines that the leak point is located in the non-branch area of ​​the main pipe section, and the calculation result of step S108 can be directly used as the final location result.

[0089] If it is within the preset range of the branch point, then proceed to step S402; If it is not within the preset range of the branch point, then proceed to step S407.

[0090] S402. Obtain the third acoustic sensor located on the branch pipe connected at the branch point, and calculate the second peak time difference of the third acoustic sensor relative to the reference node. The third acoustic sensor is a sensor installed on the branch pipeline to monitor the acoustic signal of the branch pipeline section. This sensor, together with the sensors on the main pipeline, forms a spatial monitoring network. The second peak time difference refers to the time difference between the moment when the third acoustic sensor detects the leaking acoustic wave and the moment when the reference node detects it. This time difference reflects the total time it takes for the acoustic wave to propagate along the main pipeline to the branch point and then along the branch pipeline.

[0091] After determining in step S401 that the leak point is within the preset range of the branch point, the system identifies the branch pipeline connected to that branch point based on the branch point connection relationships recorded in the pipeline topology database. Each branch point may connect to one or more branch pipelines. The system identifies each branch pipeline one by one and queries the acoustic wave sensor installed on that branch pipeline. The system obtains information such as the device number, installation location, and communication address of the third acoustic wave sensor on the branch pipeline from the sensor configuration table.

[0092] The system sends a data extraction command to the third acoustic sensor, retrieving raw acoustic waveform data from its local cache within the same time window range as in step S701. The system performs the same processing steps on the waveform data acquired by the third acoustic sensor as before, including bandpass filtering, normalization, and associated wave identification. The system performs a time-series comparison between the waveform data from the third acoustic sensor and the waveform data from the reference node, i.e., the first acoustic sensor, calculating the morphological similarity between the two. If the similarity is greater than a preset threshold, the system confirms that the third acoustic sensor has also captured valid acoustic waves generated by the same leakage source.

[0093] The system applies a multi-scale peak detection algorithm to the associated wave from the third acoustic sensor, identifying the associated peak with the largest amplitude. The timestamp of the data point corresponding to this peak is extracted and used as the arrival time of the sound wave from the third acoustic sensor. The system calculates the time difference between the arrival time of the sound wave from the third acoustic sensor and the arrival time of the sound wave from the reference node to obtain the second peak time difference. For example, if the reference node detects the peak at 100 milliseconds and the third acoustic sensor detects the peak at 145 milliseconds, the second peak time difference is 45 milliseconds.

[0094] S403. Obtain the first path length from the reference node to the branch point along the pipeline, and the second path length from the branch point to the third acoustic sensor along the branch pipeline. The first path length refers to the actual physical distance from the reference node along the pipe centerline to the branch point, which is measured from the pipe layout diagram. The second path length refers to the actual physical distance from the branch point along the branch pipe centerline to the third acoustic sensor.

[0095] The system reads the installation coordinates of reference nodes and the coordinates of branch points from the pipeline topology database. The pipeline layout diagram records the three-dimensional spatial coordinates of each pipeline node and the pipeline direction between nodes. Based on the pipeline path between the reference node and the branch point, the system calculates the actual distance along the pipeline centerline. If the pipeline segment between the reference node and the branch point is straight, the system directly calculates the Euclidean distance between the two points as the first path length. If the pipeline path contains bends or curves, the system decomposes the path into multiple straight and curved segments, calculates the length of each segment separately, and then sums them to obtain the total length along the pipeline centerline as the first path length.

[0096] The system also reads the pipeline path information between the branch point and the third acoustic sensor from the pipeline topology database. Based on the branch pipeline's direction, the system calculates the actual distance from the branch point along the branch pipeline's centerline to the third acoustic sensor, obtaining the second path length. For example, if the third acoustic sensor is installed 60 meters from the branch point, the second path length is 60 meters. The system stores the first and second path lengths and passes them to subsequent calculation modules; these two path lengths are key parameters for calculating the leak location.

[0097] S404. Based on the second peak time difference, the sound wave propagation speed, and the sum of the first path length and the second path length, calculate the total transmission distance of the leakage source relative to the reference node. The system multiplies the second peak time difference calculated in step S402 with the sound wave propagation speed to obtain the total propagation distance of the sound wave from the reference node to the third sound wave sensor.

[0098] The system adds the first path length to the second path length to obtain the total path length from the reference node through the main pipeline to the branch point and then through the branch pipeline to the third acoustic sensor. For example, if the first path length is 80 meters and the second path length is 60 meters, then the total path length is 140 meters.

[0099] The system multiplies the second peak time difference by the sound wave propagation speed and combines this with the total path length to calculate the total transmission distance of the leak source relative to the reference node. If the leak point is located between the reference node and a branch point on the main pipeline, the sound wave propagation path consists of the direct path from the leak point to the reference node and the path from the leak point through the main pipeline to the branch point and then through the branch pipeline to the third sound sensor. If the leak point is located on a branch pipeline, the sound wave propagation path consists of the path from the leak point through the branch pipeline to the branch point and then through the main pipeline to the reference node and the direct path from the leak point to the third sound sensor. By solving a system of equations, the system calculates the total transmission distance of the leak source relative to the reference node, providing a data foundation for subsequent localization.

[0100] S405. Subtract the first path length from the total transmission distance to obtain the branch offset of the leakage point relative to the branch point. After calculating the total transmission distance in step S404, the system assumes the leak point is located at a certain distance from the branch point on the branch pipeline. The sound wave propagating from the leak point to the third sound sensor requires traversing the distances from the leak point to the branch point and from the branch point to the third sound sensor. Similarly, the sound wave propagating from the leak point to the reference node requires traversing the distances from the leak point to the branch point and from the branch point to the reference node. Based on the second peak time difference and the sound wave propagation speed, the system calculates the distance difference corresponding to the time difference between the sound wave propagating from the reference node to the third sound sensor.

[0101] The system subtracts the first path length from the total transmission distance to obtain the branch offset of the leak point relative to the branch point. For example, if the total transmission distance is 180 meters and the first path length is 80 meters, then the branch offset is 100 meters. This branch offset indicates that the leak point is located 100 meters from the branch point in a certain direction. The system further determines the sign and value range of this offset. If the branch offset is positive and less than the second path length (e.g., a branch offset of 40 meters is less than the second path length of 60 meters), then the leak point is determined to be located 40 meters from the branch point on the branch pipeline. If the branch offset is negative or greater than the second path length, it indicates that the assumption is not valid, and the leak point is actually located on the main pipeline, not on the branch pipeline.

[0102] S406. Determine the specific location of the leak point on the branch pipeline based on the branch offset. After calculating the branch offset in step S405, the system determines whether the offset satisfies the condition that the leak point is located on the branch pipeline. If the branch offset is positive and within a reasonable range (i.e., the branch offset is greater than zero and less than or equal to the second path length), the system confirms that the leak point is located on the branch pipeline. The system reads the absolute position coordinates of the branch point from the pipeline topology database and calculates the absolute position coordinates of the leak point based on the branch offset along the direction of the branch pipeline.

[0103] For example, if a branch point is located at a coordinate position, and the branch pipeline extends eastward with a branch offset of 40 meters, then the leak point is located 40 meters east of the branch point. The system, considering the actual route of the branch pipeline (which may include bends or elevation changes), converts the branch offset into three-dimensional spatial coordinates, accurately marking the leak point's specific location on the branch pipeline. The system generates a leak point location report, which includes the branch pipeline number where the leak point is located, the distance from the branch point, the absolute position coordinates, and the location accuracy, providing maintenance personnel with accurate guidance to the leak point location.

[0104] S407. Determine that the leak point is located in a non-branch area of ​​the main section.

[0105] After calculating the branch offset in step S405, if the branch offset is negative or greater than the second path length, it indicates that the assumption that the leak point is located in the branch pipeline is invalid, and the leak point is actually located in the main pipeline. The system re-analyzes the acoustic wave propagation path to confirm that the leak point is located in the main pipeline section between the reference node and the branch point, and that this location is not within the preset range of the branch point, or although it is within the preset range, it is confirmed by the branch sensor data that it is not on the branch pipeline. Then, using the time difference positioning algorithm, combined with the peak time difference between the reference node and the second acoustic wave sensor, the actual length of the pipeline section, and the acoustic wave propagation speed, the system recalculates the location of the leak point in the main pipeline section. The system compares the calculation results with the branch point location to confirm that the leak point is located in the non-branch area of ​​the main pipeline section, that is, the leak point is located on the straight or curved section of the main pipeline between the two acoustic wave sensors, unaffected by the branch structure. The system generates a leak point location report, clearly marking the specific location of the leak point in the main pipeline section, providing maintenance personnel with accurate location information.

[0106] In this embodiment, the branch determination process is initiated by determining whether the leak point is within a preset range of the branch point. The second peak time difference is calculated by collecting acoustic data from a third acoustic sensor on the branch pipeline. Combined with the first path length from the reference node to the branch point and the second path length from the branch point to the third acoustic sensor, the branch offset is calculated using the difference between the total transmission distance and the first path length. Based on the positive or negative value and range of the branch offset, the leak point is accurately determined to be located in the non-branch area of ​​the branch pipeline or the main pipeline. Therefore, the specific location of the leak point can be accurately distinguished in complex pipe network structures with branch pipelines. This avoids the problem of inaccurate positioning due to multiple possible acoustic propagation paths in traditional methods when there are branch pipelines. It effectively solves the technical problems of low leak location accuracy near branch points in coal mine compressed air pipelines, confusion between branch and main pipeline leak determination, and algorithm failure in complex pipe network structures. This achieves accurate identification and positioning of leak points in branch pipe networks of coal mine compressed air pipelines, significantly improving the applicability and positioning reliability of the leak detection system under complex pipe network conditions, and providing an effective leak monitoring solution for coal mine compressed air systems with dense branch pipelines.

[0107] The compressed air pipeline monitoring system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference needed]. Figure 5 This is a schematic diagram of the physical device structure of a compressed air pipeline monitoring system in an embodiment of this application.

[0108] It should be noted that, Figure 5 The structure of the compressed air pipeline monitoring system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0109] like Figure 5As shown, the compressed air duct monitoring system includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 502 or programs loaded from storage section 508 into Random Access Memory (RAM) 503, such as performing the methods described in the above embodiments. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0110] The following components are connected to I / O interface 505: input section 506 including audio input devices, push-button switches, etc.; output section 507 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 508 including a hard disk, etc.; and communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 509 performs communication processing via a network such as the Internet. Drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0111] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the various functions defined in the present invention.

Claims

1. A method for monitoring and locating coal mine compressed air pipelines based on multimodal fusion, applied to a compressed air pipeline monitoring system, the compressed air pipeline monitoring system comprising a sensor set deployed along the compressed air pipeline, the sensor set including multiple pressure sensors, multiple flow sensors, multiple acoustic sensors, and at least one voltage sensor connected to the compressed air equipment, and a data processing unit communicatively connected to each sensor, characterized in that, The method includes: Based on the preset pipeline network topology diagram, the sensors in the sensor set are mapped to the corresponding pipe segment spatial coordinates to form a spatial mapping distribution of equipment; Based on the spatial mapping distribution of the devices, the area containing each sensor in the sensor set is determined as the distributed monitoring area; The monitoring data corresponding to the distributed monitoring area is obtained from each sensor, and the monitoring data includes pressure data, flow data, sound wave data and voltage data; The monitoring data is time-aligned and then input into a Transformer model that matches the distributed monitoring area to obtain the theoretical pressure loss value. After obtaining the actual pressure loss value under the current operating conditions, calculate the current pressure loss difference between the actual pressure loss value and the theoretical pressure loss value. If the current pressure loss difference is greater than the preset pressure loss threshold, a leak is determined, and the leak location process is initiated. The peak values ​​and peak time differences of the acoustic signal are determined based on the first acoustic sensor and at least one adjacent second acoustic sensor within the distributed monitoring area. The location of the leak is determined by combining the peak time difference, the distance between the first and second acoustic sensors, and the speed of sound propagation.

2. The method according to claim 1, characterized in that, The step of determining the area containing each sensor within the sensor set as a distributed monitoring area based on the device spatial mapping distribution specifically includes: Based on the branch nodes in the pipeline topology diagram, the compressed air pipeline is divided into several independent logical segments; For any given logic pipe segment, identify the pressure sensor and flow sensor located at both ends of the logic pipe segment as boundary monitoring points; Identify voltage sensors of all compressed air devices connected inside the logic pipe segment and downstream of the boundary monitoring point, as well as at least one acoustic sensor deployed along the path of the logic pipe segment; The logic segment covered by the boundary monitoring point and the voltage sensor is defined as the distributed monitoring area.

3. The method according to claim 1, characterized in that, Before the step of inputting the time-aligned monitoring data into a Transformer model matching the distributed monitoring area to obtain the theoretical pressure loss value, the method further includes: The monitoring data collected by each sensor in the distributed monitoring area under multiple operating conditions is used as initial training data. The multiple operating conditions include a first operating condition, a second operating condition, and a third operating condition. The first operating condition includes first monitoring data and actual pressure loss value corresponding to when the equipment operating status is identified as "on". The second operating condition includes second monitoring data and actual pressure loss value corresponding to when the equipment operating status is identified as "off". The third operating condition includes third monitoring data and actual pressure loss value corresponding to when the equipment operating status is identified as different load levels. The initial training data is time-aligned to construct a multimodal feature vector sequence. The voltage data in the multimodal feature vector sequence is encoded as a device operating status identifier, which represents the start-up and shutdown status and load level of the compressed air equipment in the distributed monitoring area. The multimodal feature vector sequence is input into the Transformer model for training. The model parameters are optimized through backpropagation to obtain the trained Transformer model. The trained Transformer model is used to predict pressure loss values ​​based on the input monitoring data.

4. The method according to claim 1, characterized in that, The steps of determining the peak values ​​and peak time differences of the acoustic signal based on the first acoustic sensor and at least one adjacent second acoustic sensor within the distributed monitoring area specifically include: Lock the target time sliding window after the current pressure loss difference is greater than the preset pressure loss threshold; Acquire multiple raw acoustic waveform data segments corresponding to the first and second acoustic sensors within a preset time period before and after the target time sliding window. The original waveform data segments of the sound wave are compared in time to identify whether there are associated waves with a similarity in morphological features greater than a preset similarity threshold. If the associated wave exists, it is determined that both the first acoustic sensor and the second acoustic sensor have captured the effective acoustic wave generated by the same leakage source. Multi-scale peak detection is performed on multiple associated waves to identify associated wave peaks; Extract multiple timestamps corresponding to the associated wave peaks as the arrival times of the sound waves from each acoustic sensor; The first acoustic sensor that first detects the associated peak is selected as the reference node, and the arrival time difference of the acoustic wave of the second acoustic sensor relative to the reference node is calculated to obtain the peak time difference.

5. The method according to claim 4, characterized in that, The step of determining the location of the leak point by combining the peak time difference, the distance between the first and second acoustic sensors, and the sound wave propagation speed specifically includes: Obtain the actual length of the pipe section between the first acoustic wave sensor and the second acoustic wave sensor; The actual physical length is converted into the standard acoustic path time for sound waves to travel between the first and second acoustic sensors by combining the speed of sound wave propagation. The arrival time difference of the sound waves is numerically compared with the standard sound path time: If the arrival time difference of the sound wave is equal to the standard sound path time, then the leakage source is determined to be located in a new logic segment on the other side adjacent to the current logic segment, with the first sound wave sensor as the endpoint. The location of the leak point is recalculated based on the new second acoustic sensor corresponding to the new logic tube segment. If the time difference of sound wave arrival is less than the standard sound path time, the leakage source is determined to be located in the target logic tube segment between the first sound wave sensor and the second sound wave sensor. If there are no branches within the target logical pipe segment, the location of the leak point is calculated by combining the peak time difference, the actual length of the pipe segment, and the sound wave transmission speed.

6. The method according to claim 5, characterized in that, If there are no branches within the target logical pipe segment, then after calculating the location of the leak point by combining the peak time difference, the actual length of the pipe segment, and the sound wave propagation speed, the method further includes: If there is at least one branch point in the target logic tube segment, then determine whether the location of the leak point is within the preset range of the branch point; If it is within the preset range, then the third acoustic sensor located on the branch pipe connected by the branch point is obtained, and the second peak time difference of the third acoustic sensor relative to the reference node is calculated. Obtain the first path length from the reference node along the pipeline to the branch point, and the second path length from the branch point along the branch pipeline to the third acoustic sensor; Based on the second peak time difference, the sound wave propagation speed, and the sum of the first path length and the second path length, the total transmission distance of the leakage source relative to the reference node is calculated; Subtracting the first path length from the total transmission distance yields the branch offset of the leak point relative to the branch point. The specific location of the leak point on the branch pipeline is determined based on the branch offset.

7. The method according to claim 1, characterized in that, Also includes: The system acquires real-time fluid pressure values ​​inside the pipe and associated temperature values ​​from each pressure sensor within the distributed monitoring area. Based on the preset gas state equation, the actual sound wave propagation speed under the current operating conditions is calculated using the real-time fluid pressure value and the temperature value inside the pipe.

8. A compressed air duct monitoring system, characterized in that, The compressed air duct monitoring system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the compressed air duct monitoring system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the compressed air duct monitoring system, the compressed air duct monitoring system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the compressed air duct monitoring system, it causes the compressed air duct monitoring system to perform the method as described in any one of claims 1-7.

Images