Unmanned aerial vehicle inspection control system and method based on acoustic imaging gas pipeline leak detection

Abstract

The present application belongs to the technical field of unmanned aerial vehicle leak detection of gas pipeline, and discloses an unmanned aerial vehicle inspection control system and method for leak detection of gas pipeline based on acoustic imaging, wherein an optical imaging auxiliary unit collects environmental images in real time, a scene self-adaptive switching unit identifies the scene type through image feature recognition and loads a corresponding interference feature template; an interference feature acquisition subunit synchronously collects three-dimensional data of pipeline vibration, body vibration and environmental noise, interference types, intensity levels and spectrum masks are generated through modeling of a CNN-LSTM interference feature learning unit, and are transmitted to a composite noise reduction preprocessing unit to support a three-level noise reduction process, finally generating a pure leak point signal, effectively eliminating residual interference and reducing the missed detection rate; a collaboration module establishes three types of channels of control instructions, detection data and optimization parameters, a collaborative instruction priority channel unit ensures priority transmission of control instructions, and a blockchain data notarization unit generates hash values for all transmission data, ensuring that the data of the whole link is reliable.

Description

Technical Field

[0001] The present invention belongs to the technical field of gas pipeline drone leak detection, and specifically relates to an unmanned aerial vehicle (UAV) patrol control system and method for gas pipeline leak detection based on acoustic imaging. Background Technique

[0002] Gas pipelines are important infrastructure for the transportation of energy such as petroleum and natural gas. Their leak detection is directly related to operational safety and energy loss control. Traditional methods such as manual patrol and fixed-point detection have problems such as low efficiency, limited coverage, and poor adaptability to complex terrains. The combination of drones and acoustic leak detection technology has become an important development direction in this field due to its advantages such as high flexibility and wide patrol range. In the field of the integration of gas pipeline acoustic leak detection and UAV patrol, although the existing technology has initially combined the two, there are still the following technical problems:

[0003] Lack of dynamic coordination between acoustic imaging and UAV flight. In the existing technology, the acoustic perception module and the UAV flight control work in an independent mode. The UAV flies according to a preset path, and the acoustic module passively collects signals. There is no real-time linkage feedback between the two. When a suspected leak point is detected, there is a significant delay from when the UAV receives the signal to when it adjusts its attitude and hovers. During this period, the acoustic signal of the leak point generates a Doppler frequency shift due to the change in the flight attitude, resulting in the subsequent imaging algorithm being unable to accurately match the signal source position, and the leak point positioning error is significantly enlarged.

[0004] Insufficient adaptive suppression ability for multi-source interference in complex scenarios. In the pipeline patrol scenario, the acoustic wave of the leak point is easily affected by the superposition of multiple interferences, including the structural noise generated by the vibration of the pipeline itself, the vibration noise of the UAV fuselage caused by the high-altitude airflow, and the industrial noise in the surrounding environment. The existing technology only uses fixed-frequency filtering or a single noise reduction algorithm, and cannot dynamically identify the characteristic spectra of different interferences, resulting in a significantly low signal-to-noise ratio of the leak point signal. The leak point and the interference area in the acoustic imaging map are blurred and overlapped, and the missed detection rate is relatively high, making it difficult to adapt to complex patrol environments such as mountainous areas and industrial areas. Summary of the Invention

[0005] The purpose of the present invention is to provide an unmanned aerial vehicle patrol control system and method for gas pipeline leak detection based on acoustic imaging to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An unmanned aerial vehicle patrol control system for gas pipeline leak detection based on acoustic imaging, the system includes:

[0007] Perception module: Collect environmental images and identify the scene type, and at the same time collect interference characteristics and model them. After three-level noise reduction processing, a pure leak point signal is generated;

[0008] Control module: Based on the pure leak signal, predict the signal location and generate a lead command by combining attitude data; when a leak signal is detected, summon the UAV to form a triangular array, adjust the flight trajectory, and collect signals from multiple angles;

[0009] Processing module: Removes residual interference, corrects signal phase difference to generate acoustic imaging, acquires structured light data and optical images to generate a 3D point cloud model, calculates leak point coordinates and determines the leak level;

[0010] Collaboration module: Establishes three channels for control commands, detection data, and optimization parameters; ensures control commands are transmitted first; compresses data and buffers it locally when the signal is weak; generates data hash values ​​for all transmitted data.

[0011] Control module: Integrates historical data to determine the confidence level of leaks and visualizes the distribution of interference; uploads data after inspection, optimizes the interference model and prediction model, generates inspection reports and archives them.

[0012] Preferably, the sensing module is specifically as follows:

[0013] It includes a MEMS array microphone, a composite noise reduction preprocessing unit, an interference feature acquisition subunit, an optical imaging auxiliary unit, a scene adaptive switching unit, a CNN-LSTM interference feature learning unit, and an interference source tracing subunit;

[0014] The scene adaptive switching unit is equipped with the MobileNetV3 lightweight image recognition model, the CNN-LSTM interference feature learning unit is equipped with an embedded edge AI processor, and the interference tracing subunit is associated with the pipeline GIS and the historical interference database.

[0015] The optical imaging auxiliary unit acquires environmental images in real time, and the scene adaptive switching unit identifies scene types through image features and loads corresponding interference feature templates, and outputs scene type data synchronously.

[0016] The interference feature acquisition subunit synchronously acquires multi-dimensional feature data of pipeline vibration (displacement / acceleration), fuselage vibration (attitude angle change rate), and environmental noise (sound pressure level-frequency spectrum). After the interference feature learning unit of CNN-LSTM models and generates interference type, intensity level, and spectral mask, it is transmitted to the composite noise reduction preprocessing unit to support the three-level noise reduction process of CNN-LSTM feature mask pre-filtering → adaptive wavelet threshold noise reduction → dynamic cancellation of interference features. On the other hand, it is synchronized to the interference feature recognition unit of the processing module through the collaboration module. Finally, the generated clean leak signal is output to the control module.

[0017] Preferably, the control module is as follows:

[0018] It includes a GPS positioning unit, a six-axis attitude sensor, an acoustic-flight cooperative control unit, a dynamic path planning unit, an actuator, a signal trend prediction unit, a multi-aircraft cooperative communication unit, and a distributed path optimization unit;

[0019] The signal trend prediction unit is equipped with a GRU time series prediction model, the multi-machine collaborative communication unit is equipped with an industrial-grade 5G Mesh self-organizing network module, and the distributed path optimization unit is equipped with an improved ant colony algorithm processor.

[0020] Based on the clean leak signal from the sensing module, after the acoustic-flight cooperative control unit receives the signal strength and direction angle data, the signal trend prediction unit predicts the signal position, and combines the current attitude data collected by the six-axis attitude sensor to generate a lead attitude command and control the actuator to move. At the same time, the attitude adjustment data is synchronized to the processing module through the cooperative module.

[0021] When a leak signal is detected, the multi-machine collaborative communication unit summons the drone. The distributed path optimization unit combines the coordinate data of the three drones from the GPS positioning unit with the leak signal strength to allocate triangular surround array parameters. At the same time, the array data is fed back to the dynamic path planning unit to adjust the flight trajectory to avoid interference sources and ensure that the array microphone is always facing the direction of the pipeline. The multi-angle signals collected are synchronized to the processing module through the collaborative module.

[0022] Preferably, the processing module is as follows:

[0023] It includes an embedded heterogeneous processor, an interference feature recognition unit, a delay-compensated acoustic imaging algorithm unit, a multi-source data fusion unit, a point cloud data generation unit, a joint Kalman filter compensation unit, and a leakage level intelligent judgment unit;

[0024] The point cloud data generation unit is equipped with an optical camera and a structured light sensor; the joint Kalman filter compensation unit integrates IMU and GPS data; and the leakage level intelligent judgment unit is equipped with a transfer learning classification model.

[0025] The interference feature identification unit receives the interference feature template from the sensing module and the attitude adjustment data from the control module to eliminate residual interference; the joint Kalman filter compensation unit fuses the attitude data from the control module and the signal transmission delay data from the coordination module to correct the signal phase difference and generate an acoustic imaging map by combining the optimized beamforming algorithm.

[0026] The structured light data and optical images collected by the point cloud data generation unit, together with the acoustic imaging image, are used by the multi-source data fusion unit to generate a three-dimensional point cloud model through the ICP algorithm, and the three-dimensional coordinates of the leak point are calculated. The intelligent leakage level assessment unit combines the leak point signal amplitude, the measured area of ​​the three-dimensional model, and the type of pipeline medium to determine the leakage level and generate handling suggestions. Finally, the leak point coordinates, leakage level, and three-dimensional model data are uploaded to the control module through the collaboration module.

[0027] Preferably, the collaborative module is as follows:

[0028] It includes a 5G industrial communication module, an anti-interference transmission unit, a data caching unit, a collaborative instruction priority channel unit, a blockchain data storage unit, a heterogeneous network switching unit, and an edge computing preprocessing unit.

[0029] The blockchain data storage unit is equipped with a lightweight Fabric consortium blockchain node. The heterogeneous network switching unit is compatible with 5G / 4G / LoRa, where 5G / 4G is used for high-speed transmission of imaging data and control commands within the inspection area, and LoRa is used for low-speed data retransmission and node communication in weak signal areas such as mountainous areas. The edge computing preprocessing unit is integrated into the main node of the UAV cluster and is used for detection data compression.

[0030] Three types of transmission channels are established: control commands, detection data, and optimization parameters. The collaborative command priority channel unit ensures the priority of attitude adjustment commands and multi-machine collaborative commands from the control module. The detection data channel synchronously transmits interference feature data from the sensing module, GPS data and attitude data from the control module, and imaging data and coordinate data from the processing module. The edge computing preprocessing unit compresses the data and transmits it to the ground terminal. When the signal is weak, local caching is activated and the data is retransmitted after the signal is restored.

[0031] The optimized parameter channel receives model update parameters from the control module and feeds them back to the sensing and control modules; the blockchain data storage unit generates hash values ​​for all transmitted data to ensure the trustworthiness of the data throughout the entire chain from collection to control.

[0032] Preferably, the control module is specifically as follows:

[0033] It includes an industrial-grade control computer, a multi-dimensional human-machine interface, a data storage and analysis unit, an interference spectrum visualization sub-unit, an AI-assisted interpretation unit, a self-learning update unit, and a multi-dimensional report generation unit;

[0034] The AI-assisted interpretation unit has a built-in Transformer leak detection model, and the self-learning update unit and the multi-dimensional report generation unit are both associated with a cloud-based big data platform.

[0035] The AI-assisted interpretation unit receives the leakage data from the processing module and the hash verification results from the collaboration module, integrates the historical leakage data, and outputs the leakage confidence judgment. The result is displayed in real time on the multi-dimensional human-computer interaction interface. The interference spectrum visualization subunit receives the interference feature data from the sensing module and displays the interference distribution.

[0036] After each inspection, the self-learning update unit uploads the leak data from the processing module, the interference characteristics from the perception module, and the control parameters from the control module to the cloud big data platform. It then optimizes the CNN-LSTM interference model and the GRU prediction model using the gradient descent algorithm. The generated optimization parameters are fed back to the perception module and the control module through the collaboration module. The multi-dimensional report generation unit combines the data from all modules to automatically generate an inspection report containing leak details, interference source tracing, and handling suggestions. The report is then archived to the data storage and analysis unit.

[0037] This invention also provides a UAV inspection control method based on acoustic imaging for gas pipeline leak detection. Based on the above system, the specific steps of this method are as follows:

[0038] Step S1: The ground terminal imports pipeline parameters and GIS data through the control module, generates the initial interference model of the scene, and completes the initialization of the noise reduction algorithm; the master UAV and the slave UAV synchronize through 5G Mesh, calibrate the signal and attitude mapping relationship, generate the initial flight path, and archive parameters; providing basic data for step S2.

[0039] Step S2: The UAV flies along the initial path described in Step S1, loads the initial interference model and performs a three-level noise reduction process. When a signal is missed, the UAV simultaneously predicts the signal location, adjusts its attitude, and calls in a follower UAV. The three UAVs collect multi-angle data in a triangular array, compress the data, and transmit it to the control module for caching.

[0040] Step S3: Receive the data from step S2, remove interference, correct phase difference, establish a three-dimensional model to determine the level of leakage, and generate an inspection report; if verification is required, control the UAV to re-collect data, optimize the CNN-LSTM interference model and GRU prediction model after inspection, and feed back the optimized model to step S1 to update the initial interference model and parameter calibration basis.

[0041] Preferably, step S1 is as follows:

[0042] The ground terminal imports the parameters of the pipeline to be inspected, the pipeline GIS data and the historical interference database through the control module. The system automatically generates an initial interference model for typical scenarios based on data mining and completes the initialization of the composite noise reduction algorithm parameters of the sensing module.

[0043] The master and slave drones synchronize time through the 5G Mesh network of the collaboration module. The control module and the perception module calibrate parameters and set the mapping relationship between signal strength and turning angular velocity, signal direction angle and deceleration acceleration, and signal distance and array spacing. The blockchain data storage unit completes node registration and encryption key distribution. The dynamic path planning unit generates the initial partition flight path in combination with the pipeline GIS map and synchronizes it to the flight control unit of all drones through the collaboration module.

[0044] Finally, the UAV GPS positioning calibration, attitude self-check, and communication link test are performed to establish a stable connection between the ground terminal and the UAV cluster. All calibration parameters and path data are archived to the data storage and analysis unit of the control module.

[0045] Preferably, step S2 is as follows:

[0046] The drone swarm takes off according to the initial partition flight path generated in step S1. The scene adaptive switching unit of the perception module collects environmental images in real time, loads the corresponding scene initial interference model generated in step S1, and starts the interference monitoring, feature learning and composite noise reduction linkage process. The output clean leakage signal is directly transmitted to the control module through the collaboration module.

[0047] When the main UAV detects that the signal strength of the leak point exceeds the preset threshold, the signal trend prediction unit generates a signal position prediction value based on the mapping relationship calibrated in step S1; the acoustic-flight cooperative control unit generates attitude adjustment commands to control the main UAV to predict the turning and hovering; the multi-UAV cooperative communication unit summons the slave UAV, and after the slave UAV flies to the designated area and obtains the coordinates through the GPS positioning unit, the distributed path optimization unit allocates triangular surround parameters according to the array parameters set in step S1.

[0048] The three units synchronously acquire multi-angle acoustic signals, optical images, and structured light data in an array. After the edge computing preprocessing node compresses the data, it is transmitted to the control module of the ground terminal on one hand, and cached in the local data cache unit on the other hand. After the signal stabilizes, it is completely uploaded to the processing module.

[0049] Preferably, step S3 is as follows:

[0050] The processing module receives the multi-machine acquisition data uploaded in step S2. The interference feature identification unit calls the initial interference model generated in step S1 and compares it with the real-time interference spectrum to eliminate residual interference. The Kalman filter compensation unit fuses the flight attitude data transmitted in step S2, corrects the signal phase difference, and generates an acoustic imaging map and a three-dimensional point cloud model.

[0051] The AI-assisted interpretation unit integrates the 3D model with historical leak data and outputs the leak verification results and leakage level. After the ground terminal verifies the consistency of the data hash, the multi-dimensional report generation unit automatically generates an integrated inspection report. The ground terminal issues a secondary review or return command through the human-machine interface. If a review is required, the coordination module sends an adjustment command to the control module to control the drone to re-collect data.

[0052] After the inspection is completed, the self-learning update unit uploads the leak data, interference characteristics, and control parameters to the cloud, optimizes the CNN-LSTM interference model and GRU prediction model, archives the generated updated model and parameters to the historical interference database, and feeds back to step S1.

[0053] The beneficial effects of this invention are as follows:

[0054] 1. This invention is based on the pure leak signal output by the sensing module. After the acoustic-flight cooperative control unit receives the signal strength and direction angle data, the signal trend prediction unit predicts the signal position in advance. Combined with the current attitude data collected by the six-axis attitude sensor, it generates a lead attitude command and controls the actuator to move. When a leak signal is detected, the multi-machine cooperative communication unit summons the slave drone. The distributed path optimization unit combines the three-machine coordinates of the GPS positioning unit with the leak signal strength to allocate triangular orbital array parameters. The dynamic path planning unit adjusts the flight trajectory to avoid interference sources and ensures that the array microphone is always facing the pipe direction. Finally, the error of the three-dimensional coordinate calculation of the leak is controlled within ±0.2m, improving the positioning accuracy.

[0055] 2. The optical imaging auxiliary unit of this invention acquires environmental images in real time, and the scene adaptive switching unit identifies the scene type through image features and loads the corresponding interference feature template. The interference feature acquisition subunit simultaneously acquires multi-dimensional feature data of pipeline vibration, fuselage vibration and environmental noise. The interference type, intensity level and spectrum mask are generated by the CNN-LSTM interference feature learning unit and transmitted to the composite noise reduction preprocessing unit to support the three-level noise reduction process of "CNN-LSTM feature mask pre-filtering → adaptive wavelet threshold noise reduction → dynamic cancellation of interference features". Finally, a clean leak signal with a signal-to-noise ratio ≥30dB is generated, which effectively removes residual interference, reduces the false negative rate and is suitable for complex scenes such as mountainous areas and industrial areas.

[0056] 3. The collaborative module of this invention establishes three channels: control commands, detection data, and optimization parameters. The collaborative command priority channel unit ensures the priority transmission of control commands, and the blockchain data storage unit generates hash values ​​for all transmitted data to ensure the reliability of data throughout the entire chain. When the signal is weak, the data caching unit starts local caching and retransmits the data after the signal is restored to avoid data loss. After inspection, the self-learning update unit of the control module uploads the leakage data of the processing module, the interference characteristics of the perception module, and the control parameters of the control module to the cloud big data platform. The gradient descent algorithm is used to optimize the CNN-LSTM interference model and the GRU prediction model, and the optimized parameters are fed back to the perception module and the control module. The multi-dimensional report generation unit automatically generates and archives an inspection report containing leakage details, interference source tracing, and handling suggestions, reducing manual intervention, improving current inspection efficiency, and ensuring long-term detection accuracy. Attached Figure Description

[0057] Figure 1 This is a flowchart of the UAV inspection and control system for gas pipeline leak detection based on acoustic imaging, as described in this invention.

[0058] Figure 2 This is a flowchart of the UAV inspection and control method for gas pipeline leak detection based on acoustic imaging, as described in this invention.

[0059] Figure 3 This is a flowchart of the three-level noise reduction core of the sensing module of the present invention;

[0060] Figure 4 This is a flowchart of the multi-drone collaborative control process for leak detection in UAVs according to the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] like Figures 1 to 4 As shown, this embodiment of the invention provides a drone inspection and control system based on acoustic imaging for gas pipeline leak detection. The system includes:

[0063] The perception module includes a 32-channel MEMS array microphone, a composite noise reduction preprocessing unit, an interference feature acquisition subunit, an optical imaging auxiliary unit, an interference tracing subunit, a scene adaptive switching unit equipped with a MobileNetV3 lightweight image recognition model, and a CNN-LSTM interference feature learning unit equipped with an embedded edge AI processor.

[0064] The interference source tracing subunit is associated with the pipeline GIS and the historical interference database. It combines the real-time interference spectrum data of the interference feature collection subunit, matches the feature templates of historical interference sources (such as industrial fans and pipeline weld vibration), and generates source tracing information fragments containing the type of interference source and the estimated location. This information is then synchronized to the control module through the detection data channel of the collaborative module.

[0065] The optical imaging auxiliary unit acquires environmental images in real time, and the scene adaptive switching unit identifies scene types through image features and loads corresponding interference feature templates. Simultaneously, the scene type data is fed back to the control module through the collaboration module, providing scene labels for subsequent model optimization.

[0066] The interference feature acquisition subunit synchronously acquires multi-dimensional feature data of pipeline vibration, fuselage vibration, and environmental noise. After the interference type, intensity level, and spectral mask are generated by the CNN-LSTM interference feature learning unit, the data is transmitted to the composite noise reduction preprocessing unit to support the three-level noise reduction process of CNN-LSTM feature mask pre-filtering → adaptive wavelet threshold noise reduction → dynamic cancellation of interference features. On the other hand, the data is synchronized to the interference feature recognition unit of the processing module through the collaboration module to provide a benchmark for subsequent residual interference removal. Finally, a clean leak signal with a signal-to-noise ratio of ≥30dB is generated and output to the control module through a high-speed data link to provide a precise signal source for its attitude adjustment and multi-machine collaboration.

[0067] The control module includes a high-precision GPS positioning unit with a positioning accuracy of ±0.1m, a six-axis attitude sensor, an acoustic-flight cooperative control unit, a dynamic path planning unit, an actuator, a signal trend prediction unit based on a GRU time-series prediction model, a multi-machine cooperative communication unit equipped with an industrial-grade 5G Mesh self-organizing network module, and a distributed path optimization unit equipped with an improved ant colony algorithm processor; the actuator includes a motor and a servo motor.

[0068] Based on the clean leak signal from the sensing module, after receiving the signal strength and direction angle data, the acoustic-flight cooperative control unit predicts the signal position 0.2s later using the GRU model. Combined with the current attitude data (pitch angle, roll angle) collected by the six-axis attitude sensor, it generates a lead attitude command and controls the actuator to move. It predicts steering at 0.4rad / s and decelerates and hovers at 0.6m / s². At the same time, it synchronizes the attitude adjustment data to the processing module through the cooperative module to provide real-time attitude parameters for its attitude-delay joint compensation.

[0069] When a leak signal is detected, the multi-machine collaborative communication unit summons the drone through the 5G Mesh self-organizing network. The distributed path optimization unit combines the coordinate data of the three drones from the high-precision GPS positioning unit with the leak signal strength to allocate triangular surround array parameters. At the same time, the array data is fed back to the dynamic path planning unit to adjust the flight trajectory to avoid interference sources, such as industrial fans, and to ensure that the array microphones are always facing the direction of the pipe. The collected multi-angle signals are synchronized to the processing module through the collaborative module.

[0070] The processing module includes an embedded heterogeneous processor, an interference feature recognition unit, a delay-compensated acoustic imaging algorithm unit, a multi-source data fusion unit, a point cloud data generation unit equipped with an optical camera and a structured light sensor, a joint Kalman filter compensation unit that fuses IMU and GPS data, and a leakage level intelligent judgment unit based on a transfer learning classification model.

[0071] The interference feature identification unit receives the interference feature template from the sensing module and the attitude adjustment data from the control module, and compares the real-time interference spectrum with the template to eliminate residual interference; the joint Kalman filter compensation unit fuses the attitude data from the control module and the signal transmission delay data from the coordination module, corrects the signal phase difference, and combines the optimized beamforming algorithm to generate an acoustic imaging map with a resolution of 0.08m×0.08m.

[0072] The structured light data and optical images collected by the point cloud data generation unit, together with the acoustic imaging image, are used by the multi-source data fusion unit to generate a three-dimensional point cloud model through the ICP algorithm, and the three-dimensional coordinates of the leak point are calculated with an error of ±0.2m. The intelligent leakage level assessment unit combines the leak point signal amplitude, the measured area of ​​the three-dimensional model, and the pipeline medium type (pre-modeling data from the control module) to determine the leakage level and generate handling suggestions. Finally, the leak point coordinates, leakage level, and three-dimensional model data are uploaded to the control module through the collaboration module.

[0073] Joint Kalman filter phase compensation formula:

[0074] ;

[0075] In the formula: It is the signal phase difference after compensation at time k, with a value range of 0~2π, which is directly used to optimize the beamforming algorithm and generate a high-resolution acoustic image.

[0076] It is the predicted phase difference value at time k, and its value range is also 0~2π. It is calculated based on the compensation results at time k-1 and the UAV attitude change rate.

[0077] It is the Kalman gain at time k, with a value ranging from 0 to 1 (dimensionless). Its function is to balance the phase difference prediction error and the measurement error, and to dynamically optimize the compensation accuracy.

[0078] It is the phase difference of the measured signal at time k, with a value of 0~2π, which is calculated from the multi-channel leak signal collected by the MEMS array microphone;

[0079] This is the prediction error covariance matrix at time k, with dimensions 3×3. The diagonal elements correspond to the prediction error variances of phase difference, attitude angle, and delay time, respectively, with units of 1, 2, 3, and 4. , used to quantify and predict uncertainty;

[0080] yes Measurement noise covariance at time, range of values The measurement noise variance of a default 32-channel array is determined by the hardware measurement accuracy of the MEMS microphone. ;

[0081] It is an identity matrix with dimensions 3×3, used to update the compensated error covariance matrix. ;

[0082] yes The observation matrix at time t, with dimension t. The corresponding observation mapping relationship for phase difference, attitude angle, and delay time is represented by the matrix elements. , It is the partial derivative of the phase difference with respect to the signal transmission delay time. The combined partial derivative of the attitude angle with respect to the phase difference is given by the formula: ; where the roll angle partial derivative Weights are set to 0.6, pitch angle partial derivatives The weight is set to 0.4. Based on historical data of the impact of UAV flight attitude on phase difference, the calibration ensures that the impact of attitude angle on phase difference is accurately quantified, which is used to establish the correlation between observation value and state value.

[0083] It refers to the drone's flight attitude angles, including the roll angle. and pitch angle The unit is radians, and the data is collected in real time by a six-axis attitude sensor at a frequency of 100Hz.

[0084] It is the signal transmission delay time, ranging from 0 to 100 ms, which is monitored and fed back in real time by the anti-interference transmission unit of the collaborative module.

[0085] Leakage level determination formula:

[0086] ;

[0087] In the formula: The leak level is represented by a value of 0, 1, 2, 3, or 4, corresponding to no leak, minor leak, moderate leak, severe leak, and emergency leak, respectively. This information is used to generate targeted handling recommendations, such as recommending periodic retesting for minor leaks and immediate shutdown and repair for emergency leaks.

[0088] The weighting coefficient for the average strength of the leak signal is fixed at 0.4 and is calibrated by the control module based on historical maintenance data to ensure its correlation with the actual leak risk.

[0089] The weighting coefficient for the leak area is fixed at 0.3. The leak area is measured by fitting the 3D point cloud model using the ICP algorithm. The weighting value is calibrated based on the correlation data between historical leak area and leakage hazard.

[0090] The weighting factor for the medium type coefficient is fixed at 0.3. It is set according to pipeline safety specifications and is used to reflect the difference in the degree of danger of leakage of different media. The weighting value matches the results of the media hazard assessment.

[0091] The average strength of the leaked signal ranges from 0 to 10. After normalization, it is the average value of all signal strengths during the multi-machine triangular surround acquisition process to ensure data representativeness.

[0092] The actual area of ​​the leak point, ranging from 0 to 1, is the normalized area corresponding to the measured leak point area. The data is obtained by the point cloud data generation unit by fitting optical images and structured light data, reflecting the physical size of the leak point.

[0093] The pipeline medium type coefficient is set with specific values ​​according to the type of medium: 3 for natural gas, 5 for liquefied petroleum gas, and 1 for industrial inert gases (such as nitrogen). It is determined by the pipeline parameters imported by the control module in the pre-modeling stage of step S1.

[0094] when The leakage level is determined when the calculation results fall within the following ranges:

[0095] When the calculation result is <0.5, the corresponding leakage level is 0;

[0096] The calculation results fall within the range [0.5, 1.5), corresponding to a leakage level of 1.

[0097] The calculation result is within the range [1.5, 2.5), corresponding to a leakage level of 2;

[0098] The calculation result is within the range of [2.5, 3.5), corresponding to a leakage level of 3;

[0099] The calculation result is in the range of [3.5, 4.5], corresponding to a leakage level of 4.

[0100] The collaborative module includes a 5G industrial communication module, an anti-interference transmission unit, a data caching unit, a collaborative instruction priority channel unit, an edge computing preprocessing unit, a blockchain data storage unit equipped with a lightweight Fabric consortium chain node, and a heterogeneous network switching unit compatible with 5G / 4G / LoRa; the edge computing preprocessing unit is integrated into the main node of the drone cluster and is responsible for detecting data compression.

[0101] Three transmission channels are established: control commands, detection data, and optimization parameters. The priority channel unit for collaborative commands ensures the priority of attitude adjustment commands and multi-machine collaborative commands from the control module. The detection data channel synchronously transmits interference feature data from the sensing module, GPS data and attitude data from the control module, and imaging data and coordinate data from the processing module. The edge computing preprocessing unit compresses the data (1 / 10 of the original data volume) and transmits it to the ground terminal. When the signal is weak, the local cache is activated and the data is retransmitted after the signal is restored.

[0102] The optimized parameter channel receives and controls the model update parameters of the control module, feeds them back to the perception module to update the interference feature template, and outputs them to the control module to update the array parameters; the blockchain data storage unit generates hash values ​​for all transmitted data to ensure the trustworthiness of the data throughout the entire chain from acquisition to control.

[0103] The control module includes an industrial-grade control computer, a multi-dimensional human-machine interface, a data storage and analysis unit, an interference spectrum visualization sub-unit, an AI-assisted interpretation unit based on the Transformer leak detection model, a self-learning update unit connected to a cloud-based big data platform, and a multi-dimensional report generation unit.

[0104] The AI-assisted interpretation unit receives the leakage data from the processing module and the hash verification results from the collaboration module. After integrating the historical leakage data (retrieved by the data storage and analysis unit), it outputs the leakage confidence judgment, and the result is displayed in real time on the multi-dimensional human-computer interaction interface. The interference spectrum visualization subunit receives the interference feature data from the sensing module and intuitively presents the interference distribution to assist staff in adjusting the detection parameters.

[0105] After each inspection, the self-learning update unit uploads the leak data from the processing module, the interference features from the perception module, and the control parameters from the control module to the cloud big data platform. It optimizes the CNN-LSTM interference model and the GRU prediction model using the gradient descent algorithm. The generated optimization parameters are fed back to the perception module through the collaboration module to update the initial interference model and output to the control module to update the mapping relationship. The multi-dimensional report generation unit combines the data from all modules to automatically generate an inspection report containing leak details, interference source tracing, and handling suggestions. At the same time, the report is archived to the data storage and analysis unit to provide historical data support for the pre-modeling of subsequent inspections.

[0106] This invention also provides a UAV inspection control method based on acoustic imaging for gas pipeline leak detection. Based on the aforementioned system, the specific steps of this method are as follows:

[0107] The ground terminal imports the parameters of the pipeline to be inspected, the pipeline GIS data, and the historical interference database (including the optimized model data from previous inspections) through the control module. The system automatically generates an initial interference model for typical scenarios based on data mining and completes the initialization of the composite noise reduction algorithm parameters of the sensing module. The pipeline parameters include pipe diameter, medium type, and pressure level.

[0108] The master and slave drones synchronize time through the 5G Mesh network of the collaboration module. The control module and the perception module calibrate parameters and set the mapping relationship between signal strength and turning angular velocity, signal direction angle and deceleration acceleration, and signal distance and array spacing to ensure that the multi-drone collaborative response latency is ≤80ms. The blockchain data storage unit completes node registration and encryption key distribution. The key is used for encrypted signature and hash value verification of transmitted data to ensure that the data is not tampered with during the data transmission process. The dynamic path planning unit generates the initial partition flight path in combination with the pipeline GIS map, with an altitude of 5-10m above the pipeline and a speed of 5-8m / s, and synchronizes it to the flight control unit of all drones through the collaboration module.

[0109] Finally, the UAV GPS positioning calibration, attitude self-check, and communication link test are performed to establish a stable connection between the ground terminal and the UAV cluster. All calibration parameters and path data are archived to the data storage and analysis unit of the control module.

[0110] The drone swarm takes off according to the initial partition flight path generated in step S1. The scene adaptive switching unit of the perception module collects environmental images in real time, loads the corresponding scene initial interference model generated in step S1, and starts the interference monitoring, feature learning and composite noise reduction linkage process. The output clean leakage signal is transmitted to the control module through the collaboration module.

[0111] When the master UAV detects that the signal strength of the leak point exceeds the preset threshold, the signal trend prediction unit generates a predicted signal position value after 0.2 seconds based on the mapping relationship calibrated in step S1; the acoustic-flight cooperative control unit generates an attitude adjustment command to control the master UAV to predict the turn and hover; the multi-UAV cooperative communication unit summons the slave UAV. After the slave UAV flies to the designated area and obtains the coordinates through the GPS positioning unit, the distributed path optimization unit allocates triangular surround parameters according to the array parameters set in step S1, with a radius of 1.5m for the master UAV, a radius of 2.0m for the slave UAV, and a pitch of 0.5m; the preset threshold is set based on the pipeline medium safety specifications, and the normalized threshold is 1.2, corresponding to an actual sound pressure level of 12dB.

[0112] The three devices collect data synchronously in an array. After the edge computing preprocesses the data and compresses it, it is preferentially transmitted to the management and control module through the 5G channel. If the anti-interference transmission unit of the collaborative module detects that the 5G signal-to-noise ratio is <10dB (weak signal threshold), it will automatically switch to the LoRa network for low-speed data transmission. At the same time, it will start the local data cache unit to store the complete data. After the signal is restored (5G signal-to-noise ratio ≥15dB), the cached data will be retransmitted through the 5G channel to avoid data loss.

[0113] The processing module receives the multi-machine acquisition data uploaded in step S2. The interference feature identification unit calls the initial interference model generated in step S1 and compares it with the real-time interference spectrum to eliminate residual interference. The Kalman filter compensation unit fuses the flight attitude data transmitted in step S2, corrects the signal phase difference, and generates a high-resolution acoustic imaging map and a three-dimensional point cloud model.

[0114] The AI-assisted interpretation unit integrates the 3D model with historical leak data and outputs the leak verification results and leakage level. After the ground terminal verifies the consistency of the data hash, the multi-dimensional report generation unit automatically generates an integrated inspection report. The ground terminal issues a secondary review or return command through the human-machine interface. If a review is required, the coordination module sends an adjustment command to the control module to control the drone to re-collect data.

[0115] After the inspection is completed, the self-learning update unit uploads the leak data, interference characteristics, and control parameters to the cloud, optimizes the CNN-LSTM interference model and GRU prediction model, archives the generated updated model and parameters to the historical interference database, and feeds back to step S1.

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0117] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A UAV inspection and control system based on acoustic imaging for gas pipeline leak detection, characterized in that: The system includes: Perception module: Acquires environmental images and identifies scene types, while also acquiring and modeling interference features, and generating a clean leak signal through three levels of noise reduction processing; Control module: Based on the pure leak signal, predict the signal position and generate flight lead command by combining the UAV attitude data; when a leak signal is detected, summon the UAV to form a triangular array and adjust the flight trajectory to collect leak signals from multiple angles; Processing module: Removes residual interference after noise reduction by the sensing module, corrects the phase difference of the leak signal to generate an acoustic imaging map, collects structured light data and optical images to generate a three-dimensional point cloud model, calculates the three-dimensional coordinates of the leak point and determines the leakage level. Collaboration module: Establishes three channels for control commands, detection data, and optimization parameters; ensures priority transmission of control commands; compresses detection data and caches it locally when the signal is weak; generates data hash values ​​for all transmitted data. Control module: Integrates historical leak data to determine the confidence level of leaks, visualizes the distribution of interference in the environment; uploads the inspection data after inspection, optimizes the interference model and prediction model, generates inspection reports and archives them to the database.

2. The UAV inspection and control system based on acoustic imaging gas pipeline leak detection according to claim 1, characterized in that: The perception module includes a MEMS array microphone, a composite noise reduction preprocessing unit, an interference feature acquisition subunit, an optical imaging auxiliary unit, a scene adaptive switching unit, a CNN-LSTM interference feature learning unit, and an interference source tracing subunit. The scene adaptive switching unit is equipped with the MobileNetV3 lightweight image recognition model, and the CNN-LSTM interference feature learning unit is equipped with an embedded edge AI processor. The optical imaging auxiliary unit acquires environmental images in real time, and the scene adaptive switching unit identifies scene types through image features and loads corresponding interference feature templates, and outputs scene type data synchronously. The interference feature acquisition subunit synchronously acquires multi-dimensional feature data of pipeline vibration, fuselage vibration, and environmental noise. After the interference type, intensity level, and spectral mask are generated by the CNN-LSTM interference feature learning unit, the data is transmitted to the composite noise reduction preprocessing unit to support the three-level noise reduction process, and simultaneously transmitted to the processing module. Finally, the generated clean leak signal is output to the control module.

3. The UAV inspection and control system based on acoustic imaging gas pipeline leak detection according to claim 2, characterized in that: The control module includes a GPS positioning unit, a six-axis attitude sensor, an acoustic-flight cooperative control unit, a dynamic path planning unit, an actuator, a signal trend prediction unit, a multi-aircraft cooperative communication unit, and a distributed path optimization unit. The signal trend prediction unit is equipped with a GRU time-series prediction model, and the multi-machine collaborative communication unit is equipped with a 5G Mesh self-organizing network module. Based on the clean leak signal from the sensing module, after the acoustic-flight cooperative control unit receives the signal strength and direction angle data, the signal trend prediction unit predicts the signal position, combines the current attitude data collected by the six-axis attitude sensor to generate a lead attitude command, and controls the actuator to move, while synchronizing the attitude adjustment data to the processing module. When a leak signal is detected, the multi-machine collaborative communication unit summons the drone, the distributed path optimization unit combines the coordinate data of the three drones from the GPS positioning unit with the leak signal strength, allocates triangular surround array parameters, and the dynamic path planning unit adjusts the flight trajectory to avoid interference sources, ensuring that the array microphone is always facing the direction of the pipeline, and the collected multi-angle signals are synchronized to the processing module.

4. The UAV inspection and control system based on acoustic imaging gas pipeline leak detection according to claim 3, characterized in that: The processing module includes an embedded heterogeneous processor, an interference feature recognition unit, a delay-compensated acoustic imaging algorithm unit, a multi-source data fusion unit, a point cloud data generation unit, a joint Kalman filter compensation unit, and a leakage level intelligent judgment unit. The point cloud data generation unit is equipped with an optical camera and a structured light sensor, and the leakage level intelligent judgment unit is equipped with a transfer learning classification model. The interference feature identification unit receives the interference feature template from the sensing module and the attitude adjustment data from the control module to eliminate residual interference; the joint Kalman filter compensation unit fuses the attitude data and signal transmission delay data, corrects the signal phase difference, and combines the optimized beamforming algorithm to generate an acoustic imaging map. The structured light data and optical images collected by the point cloud data generation unit, together with the acoustic imaging image, are used by the multi-source data fusion unit to generate a three-dimensional point cloud model through the ICP algorithm, and the three-dimensional coordinates of the leak point are calculated. The intelligent leakage level assessment unit combines the leak point signal amplitude, the measured area of ​​the three-dimensional model, and the type of pipeline medium to determine the leakage level and generate handling suggestions, and finally uploads the data to the control module.

5. The UAV inspection and control system based on acoustic imaging gas pipeline leak detection according to claim 4, characterized in that: The collaborative module includes a 5G industrial communication module, an anti-interference transmission unit, a data caching unit, a collaborative instruction priority channel unit, a blockchain data storage unit, a heterogeneous network switching unit, and an edge computing preprocessing unit. The blockchain data storage unit is equipped with a lightweight Fabric consortium blockchain node, and the heterogeneous network switching unit is compatible with 5G / 4G / LoRa. The edge computing preprocessing unit is integrated into the main node of the UAV cluster and is used to detect data compression; Three types of transmission channels are established: the collaborative instruction priority channel unit ensures the priority of attitude adjustment instructions and multi-machine collaborative instructions transmission from the control module; the detection data channel synchronously transmits interference feature data from the sensing module, GPS data and attitude data from the control module, and imaging data and coordinate data from the processing module. The data is compressed by the edge computing preprocessing unit before transmission. When the signal is weak, the local buffer is activated and the data is retransmitted after the signal is restored. The model update parameters of the parameter channel receiving and control module are optimized and fed back to the sensing module and control module; the blockchain data storage unit generates hash values ​​for all transmitted data to ensure data trustworthiness.

6. The UAV inspection and control system based on acoustic imaging gas pipeline leak detection according to claim 5, characterized in that: The control module includes an industrial-grade control computer, a multi-dimensional human-machine interface, a data storage and analysis unit, an interference spectrum visualization subunit, an AI-assisted interpretation unit, a self-learning update unit, and a multi-dimensional report generation unit. The AI-assisted interpretation unit has a built-in Transformer leak detection model, and the self-learning update unit and the multi-dimensional report generation unit are both associated with a cloud-based big data platform. The AI-assisted interpretation unit receives the leakage data from the processing module and the hash verification result from the collaboration module, and outputs the leakage confidence judgment after fusing historical leakage data; the interference spectrum visualization subunit displays the interference distribution. After the inspection is completed, the self-learning update unit uploads the leak data from the processing module, the interference characteristics from the perception module, and the control parameters from the control module to the cloud big data platform. It optimizes the CNN-LSTM interference model and the GRU prediction model through the gradient descent algorithm, and feeds back the generated optimization parameters to the perception module and the control module. The multi-dimensional report generation unit generates an inspection report containing leak details, interference source tracing, and handling suggestions, and archives it.

7. A UAV inspection and control method based on acoustic imaging for gas pipeline leak detection, based on the system described in claim 6, characterized in that: The specific steps of this method are as follows: Step S1: The ground terminal imports pipeline parameters and GIS data through the control module, generates the initial interference model of the scene, and completes the initialization of the noise reduction algorithm; the master and slave UAVs synchronize through 5G Mesh, calibrate the signal and attitude mapping relationship, generate the initial flight path and archive parameters; Step S2: The UAV flies along the initial path described in Step S1, loads the initial interference model and performs a three-level noise reduction process. When a leak is detected, it simultaneously predicts the signal position, adjusts its attitude, and summons slave UAVs. The three UAVs collect multi-angle data in a triangular array, compress the data, and transmit it to the control module for caching. Step S3: After receiving the data from step S2, remove interference, correct phase difference, establish a three-dimensional model to determine the level of leakage, and generate an inspection report; If verification is required, the drone is controlled to collect data again. After inspection, the CNN-LSTM interference model and GRU prediction model are optimized, and the optimized model is fed back to step S1 to update the initial parameters.

8. The UAV inspection control method for gas pipeline leak detection based on acoustic imaging according to claim 7, characterized in that: The specific steps of S1 are as follows: The ground terminal imports the parameters of the pipeline to be inspected, the pipeline GIS data and the historical interference database through the control module. The system generates an initial interference model for typical scenarios based on data mining and completes the initialization of the composite noise reduction algorithm parameters of the sensing module. The master and slave drones synchronize time through the 5G Mesh network of the collaboration module, and set the mapping relationship between signal strength and turning angular velocity, signal direction angle and deceleration acceleration, and signal distance and array spacing; the blockchain data storage unit completes node registration and key distribution, and the dynamic path planning unit combines the pipeline GIS map to generate the initial partition flight path and synchronize it to all drones; Perform GPS positioning calibration, attitude self-test and communication link test for the UAV, establish a stable connection, and archive all calibration parameters and path data to the management module.

9. The UAV inspection control method based on acoustic imaging gas pipeline leak detection according to claim 8, characterized in that: Step S2 is as follows: The drone swarm takes off according to the initial partition flight path. The scene adaptive switching unit of the perception module collects environmental images in real time, loads the corresponding scene initial interference model, starts the interference monitoring, feature learning and composite noise reduction linkage process, and outputs a clean leak signal. When the signal strength at the leak point exceeds a preset threshold, the signal trend prediction unit generates a predicted signal location value based on the calibrated mapping relationship. The acoustic-flight cooperative control unit generates attitude adjustment commands to control the master UAV to predict, turn, and hover; after the multi-UAV cooperative communication unit summons the slave UAVs, the distributed path optimization unit allocates triangular orbiting parameters based on the array parameters. The three machines synchronously acquire multi-angle acoustic signals, optical images, and structured light data in an array. After being compressed by the edge computing preprocessing unit, the data is transmitted to the control module and simultaneously cached in the local data cache unit.

10. The UAV inspection and control method for gas pipeline leak detection based on acoustic imaging according to claim 9, characterized in that: Step S3 is as follows: The processing module receives data collected from multiple machines, the interference feature identification unit calls the initial interference model, compares it with the real-time interference spectrum to eliminate residual interference, and the Kalman filter compensation unit fuses the flight attitude data to correct the signal phase difference and generate an acoustic imaging map and a three-dimensional point cloud model. The AI-assisted interpretation unit integrates the 3D model with historical leak data, outputs leak verification results and leakage level, and after the ground terminal verifies the consistency of the data hash, it generates an integrated inspection report and issues a second review or return command. After the inspection is completed, the self-learning update unit uploads the leak data, interference characteristics, and control parameters to the cloud, optimizes the CNN-LSTM interference model and GRU prediction model, archives the generated updated model and parameters to the historical interference database, and feeds back to step S1.

Images