Variable array piping detection radar device, piping monitoring device and identification method

Abstract

The present invention discloses a variable array piping detection radar device, piping monitoring device, and identification method. The device comprises a radar bracket with three levels of vertical and horizontal adjustment, with 16 radar units arranged in a 4×4 matrix. An integrated edge computing module implements local signal noise reduction and preprocessing. The device is mounted on a self-propelled platform and combines a visible infrared dual-spectrum probe, a voiceprint recognition module, and a Beidou positioning system with an integrated acoustic, optical, and electrical mast to construct a multimodal monitoring system. The monitoring method dynamically adjusts daytime and nighttime monitoring weights, utilizes a bidirectional LSTM combined with an attention mechanism to extract features, and optimizes noise robustness through adversarial training to achieve fusion calculation of piping risk factors. The present invention constructs a closed-loop monitoring system with environmental self-adaptation, multi-source signal perception, and intelligent decision-making, enabling efficient and high-precision monitoring and early warning of piping, and achieving comprehensive breakthroughs in core indicators such as positioning accuracy, recognition accuracy, and environmental adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of embankment engineering safety monitoring, and in particular to a variable array piping detection radar device, a piping monitoring device and an identification method. Background Art

[0002] In recent years, extreme weather has become increasingly frequent worldwide. Piping, a typical form of seepage damage in levees, has become a primary threat to flood control due to its suddenness, concealment, and rapid breach. Traditional monitoring methods have significant drawbacks: manual inspections rely on a large workforce and suffer from delayed response times, low positioning accuracy, and insufficient continuous monitoring capabilities. Furthermore, due to changes in the workforce structure, personnel with professional identification experience are increasingly scarce. Existing single-point detection technologies, such as piezometers, have a false alarm rate as high as 35% under complex conditions such as turbid water and vegetation cover. They are also incapable of capturing early abnormal signals such as micro-vibrations and low seepage flow, making it difficult to achieve full-scale real-time monitoring. Early warning models based on static geological parameters fail to effectively integrate real-time hydrological data with intelligent algorithms, resulting in prediction accuracy dropping by over 40% under extreme conditions, making them incapable of meeting the needs of dynamic risk assessment.

[0003] The core bottlenecks of existing technologies are concentrated in three aspects: First, the global perception capability is insufficient. Single-point monitoring equipment can only obtain local seepage data and cannot build a dynamic model of a complex seepage field. The risk of missed detection increases significantly with changes in water conditions; second, the efficiency of multi-source data fusion is low. Multi-dimensional data such as hydrology, geology, and environment have not formed an effective correlation. The early warning model lacks the ability to analyze the evolution process of piping nonlinearly, making it difficult to predict the development path of dangerous situations; third, the project adaptability is limited. Traditional monitoring equipment is large in size, high in power consumption, high in deployment cost, and lacks modular design. It is difficult to apply it on a large scale in long-distance embankments and remote areas, and it is also unable to adapt to mobile loading platforms such as drones and patrol vehicles.

[0004] To address the above-mentioned issues, the present invention relies on technological breakthroughs such as the Internet of Things, edge computing, and machine learning to develop an active variable array detection radar device and an intelligent recognition method. High-sensitivity array sensors are used to achieve real-time monitoring of the seepage field with decimeter-level spatial resolution. By integrating BeiDou high-precision positioning technology with deep neural network algorithms, a full-chain technology system for data acquisition, intelligent analysis, precise positioning, and dynamic early warning is constructed, compressing the early warning response time from hours to minutes and increasing the prediction reliability to over 95%. The device adopts a miniaturized and modular design, supporting flexible deployment in various embankment scenarios, effectively filling the technical gap in all-weather intelligent monitoring and providing innovative solutions for flood control safety in extreme climates. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a variable array piping detection radar device, a piping monitoring device and an identification method. The device combines the variable array radar structure design, multimodal sensor fusion and dynamic weight deep learning to construct a closed-loop monitoring system with environmental adaptation, multi-source signal perception and intelligent decision-making, thereby realizing efficient and high-precision monitoring and early warning of piping, and achieving comprehensive breakthroughs in core indicators such as positioning accuracy, recognition accuracy and environmental adaptability.

[0006] In order to achieve the above-mentioned object, the present invention discloses an active variable array pipe burst detection radar device, comprising a radar bracket and a plurality of radar units arranged on the radar bracket;

[0007] The radar bracket includes a bracket base plate, a transverse telescopic bracket and a longitudinal telescopic bracket, wherein the longitudinal telescopic bracket is mounted on the bracket base plate, and the longitudinal telescopic bracket includes a longitudinal guide rail and a longitudinal linear bearing slidably connected to the longitudinal guide rail, and a longitudinal limit buckle is provided on the longitudinal linear bearing, and the longitudinal linear bearing is relatively fixed to the longitudinal guide rail by the longitudinal limit buckle;

[0008] The transverse telescopic bracket is mounted on the longitudinal linear bearing of the longitudinal telescopic bracket, the transverse telescopic bracket includes a transverse guide rail and a transverse linear bearing slidably connected to the transverse guide rail, and a transverse limit buckle is provided on the transverse linear bearing, and the transverse linear bearing is fixed relative to the transverse guide rail through the transverse limit buckle;

[0009] Each radar unit is mounted on a transverse linear bearing. The radar units are slidably adjusted in the longitudinal direction by longitudinal telescopic brackets and fixed by longitudinal limit buckles. The radar units are then adjusted in multiple levels within the longitudinal unit spacing.

[0010] Each radar unit is slidably adjusted in the lateral direction by a lateral telescopic bracket and fixedly limited by a lateral limit buckle, and then the radar unit is multi-level adjusted within the unit spacing in the lateral direction.

[0011] Furthermore, there are 16 radar units, which are distributed in a regular 4×4 matrix on the radar bracket; the radar units are set up with different transmitting and receiving arrays in the longitudinal and transverse directions to adapt to the complex embankment detection environment with different soil types and heights; the radar units are deployed in multiple levels in the longitudinal and transverse directions to adapt to different embankment widths and improve the monitoring range and accuracy.

[0012] Furthermore, the radar units are divided into four groups, A1, A2, A3, and A4, in the vertical direction. The four groups of radar units use the same frequency, or A1 and A3 are combined, and A2 and A4 are combined, so as to synchronously transmit and collect two groups of signals with different frequency characteristics. The collected signals are grouped and analyzed through digital filtering technology.

[0013] Furthermore, the radar units are divided into four groups, B1, B2, B3, and B4, in the horizontal direction. The four groups of radar units use the same frequency, or B1 and B3 are combined, and B2 and B4 are combined, so as to synchronously transmit and collect two groups of signals with different frequency characteristics. The collected signals are grouped and analyzed through digital filtering technology.

[0014] Furthermore, each of the radar units includes a DA power amplifier component, a signal acquisition AD component, an edge computing control and filtering component, and an integrated antenna module; the DA power amplifier component is used to convert the baseband digital signal into an analog RF signal and amplify the signal power; the signal acquisition AD component is used to collect the echo analog signal received by the antenna and convert it into a digital signal after conditioning; the edge computing control and filtering component is used for real-time signal processing, system control and noise filtering, and removes noise and clutter through digital filtering to improve the signal-to-noise ratio; and completes data processing locally to reduce back-end computing pressure and transmission delay;

[0015] The integrated antenna module is assembled from a built-in three-channel antenna transceiver module, a magnetic flux protection head and a transmitting antenna protection cover. The built-in three-channel antenna transceiver module is composed of a radar signal transmitting antenna, a radar reflection signal micro receiving antenna and a micro laser displacement receiving and transmitting sensor arranged at equal intervals along the longitudinal direction; the radar signal transmitting antenna, the radar reflection signal micro receiving antenna and the micro laser displacement receiving and transmitting sensor are connected by shielded cable I.

[0016] The present invention also provides a real-time piping monitoring device, comprising a self-propelled platform, an acoustic-optical integrated mast mounted on the self-propelled platform, and the piping detection radar device; the piping detection radar device is fixedly connected to one end of the self-propelled platform via a flange connector, and a roller is mounted on the bottom of the piping detection radar device;

[0017] A mast connection seat is fixed on the top wall of the self-propelled platform, and the acoustic, optical and electrical integrated mast is fixedly installed on the mast connection seat. The bottom of the acoustic, optical and electrical integrated mast is provided with an acoustic, optical and electrical signal front-end processing module and its three-proof cover, which can remove noise and pre-process the monitoring data to meet the data analysis preparation requirements of the back-end early warning platform.

[0018] Furthermore, the acousto-optical integrated mast is provided with a transverse probe module, a longitudinal probe module, a voiceprint recognition module and a Beidou satellite guidance device;

[0019] The transverse probe module is composed of a visible infrared dual spectrum probe on the back slope side and a visible light probe on the water side; the visible infrared dual spectrum probe and the visible light probe are respectively arranged on both sides of the transverse outer wall of the top of the acoustic and optical integrated mast;

[0020] The longitudinal probe module is composed of a forward video probe and a rearward video probe, which are respectively arranged on both sides of the longitudinal outer wall of the top of the acoustic and optical integrated mast;

[0021] The probe of the voiceprint recognition module is placed at the top of the acoustic, optical and electrical integrated mast; the Beidou satellite guidance device is installed on the top of the front outer side wall of the three-proof cover to provide a reference for the positioning of pipe bursts.

[0022] The present invention also provides an intelligent piping recognition method, using the piping real-time monitoring device according to claim 7, the recognition method comprising the following steps:

[0023] S1: Adjust the longitudinal extension width of the variable array piping detection radar device according to the height of the dam to be detected and the detection depth;

[0024] S2: Based on the width of the dike top to be detected, adjust the lateral extension width of the variable array piping detection radar device;

[0025] S3: Adjust the ground clearance of the variable array pipe burst detection radar device according to the surface flatness of the detection target;

[0026] S4: According to the depth and accuracy of piping detection and the inspection speed requirements, set the number of piping radar units to be started and determine the spatial layout of the front-end radar units;

[0027] S5: The high-frequency electromagnetic signal transmitting and receiving array of the variable array piping detection radar device is set to adapt to the complex monitoring environment of different soil types and altitudes. The piping detection radar device can set the shock wave frequency according to the above array matching;

[0028] S6: Adjust the height and angle of the integrated acoustic, optical, and optical mast based on the bottom width and detection range of the target dike to cover the required surface area. This ensures that the visible and infrared dual-spectrum probe covers the potential piping outburst area, acquires visible and infrared image data of the backslope and a certain range beyond it, and enables the audio probe of the voiceprint recognition module to effectively receive audio signals.

[0029] S7: The data collected by the variable array piping detection radar device is preprocessed to form a clear ripple data graph. The signal preprocessing frame is set to a frame length of 20-40ms. Time-frequency analysis can generate a Mel spectrum graph, which is then subjected to Z-score normalization.

[0030] S8: A bidirectional LSTM deep model analysis is performed on the preprocessed data acquired by the variable array pipe burst detection radar device to capture the contextual correlation features. To improve analysis efficiency, an attention mechanism is introduced to focus on key frequency bands, depthwise separable convolution is used to assist feature extraction, and an improved Sigmoid classification is used in the output layer.

[0031] S9: After the ripple data graph is identified by the deep learning algorithm, piping identification data within the dike is generated in real time. The model quickly determines the piping type and penetration based on the piping characteristics, and preliminarily assesses the current status and potential risks of the piping, and assigns a piping risk factor A1.

[0032] S10: The visible-infrared dual-spectrum probe collects visible light image information from the backwater surface in real time, introduces an environmental adaptive module, dynamically adjusts the suppression parameters, and identifies the piping outlet morphology based on the pre-trained deep learning model B. A denoising algorithm removes environmental interference and provides a piping risk factor A2.

[0033] S11: Based on the difference in piping water temperature and ambient temperature on the backwater side, the visible infrared dual-spectrum probe perceives the infrared image within a set range on the backwater side in real time. Simultaneously, the pre-trained deep learning model C performs piping infrared morphology recognition, assesses the likelihood of piping, and provides a piping risk factor A3.

[0034] S12: The acoustic wave probe of the voiceprint recognition module captures audio signals in real time. Noise suppression uses a deep learning method for noise reduction, combined with adversarial training for a noise sample library. During feature engineering, a 50-100 frame sliding window is selected as the feature sequence. Data enhancement is performed by adding random background noise, time shift / speed change processing, and dynamic mixed piping features. Noise robustness of the analysis method is optimized by adding noise perturbations through adversarial training, multi-task learning, and a dual verification mechanism in the time and frequency domains. Identification and judgment are performed by combining a noise removal algorithm with a pre-trained deep learning voiceprint recognition model D, and a piping risk factor A4 is determined.

[0035] S13: Based on the pre-trained model and the hydrological and geomorphological features around the dam, the environmental risk factor A5 is given;

[0036] S14: Based on the set fusion algorithm, the above five risk factors are weighted and analyzed to obtain the risk value of piping. Combined with spatial positioning and correction technology, the piping risk probability of the spatial positioning point of the dam is obtained, and warning information is automatically broadcast for the points where the risk probability exceeds the set value.

[0037] Furthermore, in step S14, the weighted calculation formula of the risk value is:

[0038]

[0039] Among them, k is the risk value, A i is the risk factor, ω i is the weight coefficient corresponding to the risk factor.

[0040] Furthermore, in step S14, when it is during the daytime with good visual conditions:

[0041] The weight coefficients corresponding to each risk factor are: weight A1(ω1)=0.253, weight A2(ω2)=0.382, weight A3(ω3)=0.185, weight A4(ω4)=0.136, weight A5(ω5)=0.044;

[0042] During night time:

[0043] The weight coefficients corresponding to each risk factor are: weight A1(ω1)=0.286, weight A2(ω2)=0.019, weight A3(ω3)=0.425, weight A4(ω4)=169, and weight A5(ω5)=0.101.

[0044] Beneficial effects of the present invention:

[0045] 1. The variable array piping detection radar device of the present invention utilizes a multi-stage, longitudinally and transversely adjustable array layout to achieve coverage adaptation for various dam widths, soil types, and complex terrains, resolving the difficulty of achieving a balanced monitoring range and accuracy experienced by traditional fixed array radars.

[0046] 2. Each radar unit of the variable array pipe burst detection radar device of the present invention has built-in edge computing control and filtering components, which can perform signal noise reduction and data compression in real time, shortening the delay of traditional back-end centralized processing from seconds to milliseconds, meeting the real-time warning needs; dynamic weight allocation based on deep learning can realize adaptive adjustment of monitoring strategies and avoid response lag caused by human intervention.

[0047] 3. The real-time pipe burst monitoring device of the present invention is equipped with a self-propelled platform and the Beidou satellite positioning system, combined with an offline map interpolation algorithm. It can still maintain a positioning accuracy of less than 5 cm in areas without satellite signals, solving the problems of high missed detection rate and fuzzy positioning in traditional manual inspections.

[0048] 4. The real-time pipe burst monitoring device of the present invention adopts a modular design and architecture. The front-end radar unit of the pipe burst detection radar device is miniaturized and arranged in an array. The pipe burst audio detector is placed on the top of the self-propelled platform, and an omnidirectional video probe is provided. It can collect real-time panoramic sound, light, and electrical monitoring information of complex outdoor sites, and can monitor multi-physical intuitive data related to pipe bursts to the greatest extent. Based on the data fusion algorithm, it completes the identification and comparison of pipe bursts, significantly improving the identification speed and accuracy of pipe bursts under complex working conditions.

[0049] 5. The real-time pipe burst monitoring device of the present invention integrates radar electromagnetic signals, acoustic wave signals, infrared thermal imaging and visual images to construct a multimodal monitoring system, which solves the defect that a single sensor is susceptible to environmental interference and improves the recognition reliability under complex working conditions.

[0050] 6. The present invention combines variable array mechanical design, multimodal sensor fusion, and dynamic weight deep learning to construct a closed-loop monitoring system with environmental self-adaptation, multi-source signal perception, and intelligent decision-making, thereby achieving efficient and high-precision monitoring and early warning of pipe bursts; and achieving comprehensive breakthroughs in core indicators such as positioning accuracy, recognition accuracy, and environmental adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of the overall structure of a real-time monitoring device for piping according to the present invention;

[0052] Figure 2 This is a schematic diagram of a partial structure of a longitudinal telescopic bracket in the present invention;

[0053] Figure 3 Schematic diagram of the structure of the variable array piping detection radar device of the present invention;

[0054] Figure 4 Schematic diagram of the structure of the radar unit in the present invention;

[0055] Figure 5 Schematic diagram of horizontal or vertical test grouping of piping detection radar units in the present invention;

[0056] Figure 6 This is a schematic diagram of the structure of the acoustic, optical and electrical integrated mast in the present invention;

[0057] Figure 7 This is a schematic diagram of the overall control logic of the modular monitoring information fusion and early warning based on deep learning of the present invention;

[0058] Figure 8 This is a schematic diagram of the LSTM recognition process of the piping soundprint feature of the present invention;

[0059] Figure 9 Schematic diagram of the visible light and infrared light identification algorithm for piping according to the present invention;

[0060] Figure 10 Schematic diagram of the comprehensive prediction algorithm for piping probability of the present invention.

[0061] Figure 1: 1. Self-propelled platform; 2. Mast connector; 3. Tri-proof cover; 4. Audio-visual integrated mast; 41. Flange seat; 42. Protective cover; 43. Shielded cable II; 5. Horizontal probe module; 51. Visible-infrared dual-spectrum probe; 52. Visible light probe; 6. Voiceprint recognition module; 7. Longitudinal probe module; 71. Forward video probe; 72. Rear video probe; 8. Satellite guidance device; 9. 5G transmitting antenna; 10. Piping detection radar device; 11. Radar unit; 111. Magnetic flux protection head; 112. Radar signal transmitting antenna; 113. Radar reflection signal miniature receiving antenna; 114. Laser displacement transceiver sensor; 115. Signal transmission optical cable. 150. Transmitting antenna protection cover; 151. DA power amplifier component; 152. Signal acquisition AD component; 153. Edge computing control and filtering component; 154. Shielded cable I; 155. Sealing gasket; 156. Connecting screws; 101. Radar bracket; 102. Flange connector; 103. Roller; 105. Horizontal telescopic bracket; 106. Horizontal guide rail; 107. Horizontal linear bearing; 108. Horizontal limit buckle; 116. Longitudinal telescopic bracket; 117. Longitudinal guide rail; 118. Longitudinal linear bearing; 119. Longitudinal limit buckle; 120. Bracket bottom plate; 12. Data storage module; 13. Integrated computing power cabinet; 14. Protective base. DETAILED DESCRIPTION

[0062] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0063] The invention discloses an active variable array piping detection radar device.

[0064] Reference Figures 1 to 3 An active variable array pipe burst detection radar device includes a radar bracket 101 and a plurality of radar units 11 arranged on the radar bracket 101; the radar bracket 101 includes a bracket base plate 120, a transverse telescopic bracket 105 and a longitudinal telescopic bracket 116, the longitudinal telescopic bracket 116 is installed on the bracket base plate 120, the longitudinal telescopic bracket 116 includes a longitudinal guide rail 117 and a longitudinal linear bearing 118 slidably connected to the longitudinal guide rail 117, and a longitudinal limit buckle 119 is provided on the longitudinal linear bearing 118, and the longitudinal linear bearing 118 is relatively fixed to the longitudinal guide rail 117 through the longitudinal limit buckle 119.

[0065] Reference Figure 2The transverse telescopic bracket 105 is installed on the longitudinal linear bearing 118 of the longitudinal telescopic bracket 116. The transverse telescopic bracket 105 includes a transverse guide rail 106 and a transverse linear bearing 107 slidably connected to the transverse guide rail 106, and a transverse limit buckle 108 is provided on the transverse linear bearing 107. The transverse linear bearing 107 is relatively fixed to the transverse guide rail 106 through the transverse limit buckle 108.

[0066] Reference Figure 2 Each radar unit 11 is mounted on a transverse linear bearing 107. The radar units 11 are slidably adjusted longitudinally by longitudinal telescopic brackets 116 and fixedly limited by longitudinal limit buckles 119. In this embodiment, the radar unit 11 can be adjusted longitudinally in three levels: 0.4m, 0.6m, and 0.8m. The radar unit 11 can be slidably adjusted transversely by transverse telescopic brackets 105 and fixedly limited by transverse limit buckles 108. In this embodiment, the radar unit 11 can be adjusted transversely in three levels: 0.5m, 0.75m, and 1m. In this embodiment, the longitudinal limit buckles 119 and transverse limit buckles 108 are locked and secured using locking bolts.

[0067] Reference Figure 3 and Figure 5 In this embodiment, 16 radar units 11 are provided and distributed in a regular 4×4 matrix. A shielding cover is provided outside the radar unit 11 to improve the signal-to-noise ratio. The 16 units are connected by a signal transmission optical cable 115. The radar unit 11 is configured with a longitudinal and transverse transmitting and receiving array to adapt to complex dam detection environments such as different soil types and heights. The radar unit 11 is deployed in multiple stages in the longitudinal and transverse directions to adapt to different dam widths and improve the monitoring range and accuracy. The 16 radar units 11 are divided vertically into four groups: A1, A2, A3, and A4. These four groups can use the same frequency, or they can combine rows A1 and A3, and rows A2 and A4, to simultaneously transmit and collect two sets of signals with different frequency characteristics. These collected signals are filtered and processed using digital filtering techniques before being grouped and analyzed. Alternatively, they can be divided horizontally into four groups: B1, B2, B3, and B4. These four groups can use the same frequency, or they can combine rows B1 and B3, and rows B2 and B4, to simultaneously transmit and collect two sets of signals with different frequency characteristics. These collected signals are filtered and processed using digital filtering techniques before being grouped and analyzed. Each group of radar units 11 selects 2-4 groups for operation based on the complexity of the actual operating environment and the duration of use. The piping detection radar device 10 can set the shock wave frequency based on the aforementioned formation matching. The variable frequency range is 20 kHz to 600 MHz and is controlled by the main control panel.

[0068] Reference Figure 4, each radar unit 11 includes a DA power amplifier component 151, a signal acquisition AD component 152, an edge computing control and filtering component 153 and an integrated antenna module;

[0069] The DA amplifier assembly 151 converts baseband digital signals into analog RF signals and amplifies them. The signal acquisition AD assembly 152 collects the analog echo signals received by the antenna and converts them into digital signals after conditioning. The edge computing control and filtering assembly 153 is responsible for real-time signal processing, system control, and noise filtering. It removes noise and clutter through digital filtering to improve the signal-to-noise ratio. It also performs data processing locally, reducing back-end computing pressure and transmission latency. A sealing gasket 155 is installed between the edge computing control and filtering assembly and the integrated antenna module for waterproofing. Connecting screws 156 securely connect the assembly.

[0070] The integrated antenna module consists of a built-in three-channel antenna transceiver module, a magnetic flux protection head 111, and a transmitting antenna protective cover 150. The built-in three-channel antenna transceiver module consists of a radar signal transmitting antenna 112, a radar reflection signal miniature receiving antenna 113, and a miniature laser displacement transceiver sensor 114, arranged evenly spaced longitudinally. Both the radar signal transmitting antenna 112 and the radar reflection signal miniature receiving antenna 113 are made of gallium oxide. The miniature laser displacement transceiver sensor 114 has a measurement range of 1 meter and an accuracy of 0.1 mm. The radar signal transmitting antenna 112, the radar reflection signal miniature receiving antenna 113, and the miniature laser displacement transceiver sensor 114 are connected by a shielded cable I 154.

[0071] Reference Figure 1 The present invention also discloses a real-time piping monitoring device, comprising a self-propelled platform 1, an integrated acoustic, optical, and electrical mast 4 mounted on the platform, and the aforementioned piping detection radar device 10. The piping detection radar device 10 is fixedly connected to one end of the self-propelled platform 1 via a flange connector 102. To prevent the piping detection radar device 10 from contacting the ground due to uneven road surfaces while the self-propelled platform 1 is in motion, a silent roller 103 is installed at the bottom of the piping detection radar device 10, with adjustable ground clearance to accommodate varying road conditions.

[0072] Data acquired by the piping radar device 10 is transmitted via signal transmission optical cable 115 to the data storage module 12, and then to the integrated computing power cabinet 13 for data processing and analysis. The data storage module 12 and the integrated computing power cabinet 13 are connected to the interior of the self-propelled platform 1 for vibration isolation via a protective base 14, and are powered by the self-propelled platform 1.

[0073] Reference Figure 1 and Figure 6The top wall of the self-propelled platform 1 is fixed with a mast connection base 2, to which the integrated acoustic, optical, and optic mast 4 is fixedly mounted. The bottom of the integrated acoustic, optical, and optic mast 4 is equipped with an acoustic, optical, and optic signal front-end processing module and its three-proof cover 3, which can remove noise and pre-process monitoring data to meet the data analysis and preparation requirements of the back-end early warning platform. The integrated acoustic, optical, and optic mast 4 is connected to the top wall of the three-proof cover 3 via a flange base. The top of the integrated acoustic, optical, and optic mast 4 is equipped with a protective cover 42 to protect its precision sensors and electrical circuits in complex and extreme environments.

[0074] Reference Figure 1 and Figure 6 The height and angle of the integrated acoustic, photoelectric, and electric mast 4 are adjustable to ensure effective coverage of the required detection range. The integrated acoustic, photoelectric, and electric mast 4 is equipped with a transverse probe module 5, a longitudinal probe module 7, a voiceprint recognition module 6, and a Beidou satellite guidance device 8. The transverse probe module 5 consists of a visible infrared dual-spectrum probe 51 on the slope away from the water and a visible light probe 52 on the waterside. The visible infrared dual-spectrum probe 51 and the visible light probe 52 are located on either side of the outer wall of the top of the integrated acoustic, photoelectric, and electric mast 4. Given that infrared recognition and discrimination on the waterside is almost impossible, no infrared probe is installed on this side. The longitudinal probe module 7 consists of a forward video probe 71 and a rearward video probe 72. These probes are located on either side of the outer wall of the top of the integrated acoustic, photoelectric, and electric mast 4. These probes facilitate auxiliary monitoring of the surrounding environment and spatial obstacle avoidance information for the autonomous platform 1, preventing the platform from straying into water or dangerous areas. Combined with the vehicle's own autonomous driving function, this further enhances the platform's own safety and self-defense capabilities.

[0075] Reference Figure 1 and Figure 6 The probe of the voiceprint recognition module 6 is placed at the top of the integrated acoustic, optical, and optic mast 4 and is waterproofed. The voiceprint recognition probe can identify the specific acoustic fluctuations of micro-piping surges with high resolution. After training, the module is sensitive to acoustic signals in the relevant bands for piping surges. After noise removal, it can identify the soundprint characteristics of piping surges with high clarity.

[0076] Reference Figure 1 and Figure 6 The Beidou satellite guidance device 8 is mounted on the top of the front outer wall of the three-proof cover 3. Using a whip antenna, it can achieve high-precision real-time positioning in the field where satellite signals are available, providing a reference for locating piping. In areas with weak or no satellite signals, the system automatically determines the position based on its built-in offline map and travel speed, and performs interpolation evaluation based on signal segments.

[0077] The signals detected by each of the aforementioned probes are pre-processed by the acoustic, optical, and optic signal front-end processing module before being transmitted via shielded cable II 43 to the data storage module 12. They are then transferred to the integrated computing power cabinet 13 for algorithm prediction. A 5G transmitting antenna 9 is installed at the rear end of the top wall of the triple-protection cover 3, which is used to transmit real-time warning signals to the flood control headquarters or command center. The monitoring warning signals and prediction results generated by intelligent piping identification are intuitively displayed on the vehicle's integrated display system and can be rapidly transmitted to the regional warning center via the 5G transmitting antenna 9.

[0078] The invention also discloses a method for intelligently identifying piping.

[0079] Reference Figures 7 to 10 A method for intelligently identifying piping bursts employs the above-mentioned real-time piping burst monitoring device, comprising the following steps:

[0080] S1: adjusting the longitudinal extension width of the variable array piping detection radar device 10 according to the height and detection depth of the dam to be detected;

[0081] S2: Based on the width of the dike top to be detected, adjust the lateral extension width of the variable array piping detection radar device;

[0082] S3: Adjusting the ground clearance of the variable array piping detection radar device 10 according to the surface flatness of the detection target;

[0083] S4: According to the depth and accuracy of piping detection and the inspection speed requirements, the number of radar units 11 to be started is set, and the spatial layout of the front-end radar units 11 is determined;

[0084] S5: The high-frequency electromagnetic signal transmitting and receiving array of the variable array piping detection radar device 10 is set to adapt to complex monitoring environments such as different soil types and altitudes. The piping detection radar device 10 can set the shock wave frequency according to the above array matching. The variable frequency range is 20 kHz to 600 MHz. The monitoring depth is inversely proportional to the frequency selection. The deeper the intended monitoring depth, the lower the detection frequency range.

[0085] S6: Based on the bottom width and detection range of the target dike, adjust the vertical height and angle of the integrated acoustic, optical and electrical mast 4 to cover the required specific monitoring surface range, so that the monitoring range of the visible infrared dual spectrum probe 51 covers the potential pipe burst area, obtain visible and infrared image data of the backwater slope and a certain range of extension, and enable the audio probe of the voiceprint recognition module 6 to effectively receive audio signals;

[0086] S7: The data information collected by the variable array piping detection radar device 10 is pre-processed to form a clear ripple data graph. The frame length of the signal pre-processing is set to 20-40ms. The time-frequency analysis can generate a Mel spectrum graph, which is then subjected to Z-score normalization processing.

[0087] S8: Figure 7 and Figure 9 A bidirectional LSTM deep model analysis A is performed on the data obtained and preprocessed by the variable array pipe burst detection radar device 10 to capture the previous and next correlation features. To improve the analysis efficiency, the analysis introduces an attention mechanism to focus on the key frequency bands, uses depthwise separable convolution to assist feature extraction, and uses an improved Sigmoid classification in the output layer.

[0088] S9: After the ripple data graph is identified by the deep learning algorithm, piping identification data within the dike is generated in real time. The model quickly determines the piping type and penetration based on the piping characteristics, and preliminarily assesses the current status and potential risks of the piping, and assigns a piping risk factor A1.

[0089] S10: Figure 7 and Figure 9 The visible infrared dual spectrum probe 51 collects visible light image information from the back water surface in real time, introduces an environmental adaptive module, dynamically adjusts the suppression parameters, and identifies the piping outlet morphology based on the pre-trained deep learning model B. The denoising algorithm removes the influence of environmental interference and gives the piping risk factor A2.

[0090] S11: If Figure 7 and Figure 9 Based on the difference between piping water temperature and the ambient temperature on the back side, the visible infrared dual-spectrum probe 51 perceives the infrared image within the set range of the back side in real time. At the same time, the pre-trained deep learning model C performs piping infrared morphology recognition, judges the possibility of piping, and gives the piping risk factor A3;

[0091] S12: Figure 8 The acoustic wave probe of voiceprint recognition module 6 captures audio signals in real time. Noise suppression uses a deep learning method for noise reduction and adversarial training combined with a noise sample library. During feature engineering, a 50-100 frame sliding window is selected as the feature sequence. Data enhancement is performed by adding random background noise, time shift / speed change processing, and dynamic mixed piping features. Noise perturbations are added through adversarial training. Multi-task learning and a time-domain-frequency domain dual verification mechanism are used to optimize the noise robustness of the analysis method. Identification and judgment are performed by combining a noise removal algorithm with a pre-trained deep learning voiceprint recognition model D, and a piping risk factor A4 is given.

[0092] S13: Based on the pre-trained model and the hydrological and geomorphological features around the dam (the probability of piping is significantly increased for certain geomorphological features), an environmental risk factor A5 is further provided. The levels and values ​​of the environmental risk factors at this stage of the present invention are as follows:

[0093]

[0094] S14: Figure 7 Based on the set fusion algorithm, the above five risk factors are weighted and analyzed to obtain the risk value k of piping. The weighted calculation formula is:

[0095] A i is the risk factor, ω i is the weight coefficient corresponding to the risk factor.

[0096] In the above analysis, during the daytime when visibility conditions are good:

[0097] The weight coefficients corresponding to each risk factor are: weight A1(ω1)=0.253, weight A2(ω2)=0.382, weight A3(ω3)=0.185, weight A4(ω4)=0.136, weight A5(ω5)=0.044;

[0098] During night time:

[0099] The weight coefficients corresponding to each risk factor are: weight A1(ω1)=0.286, weight A2(ω2)=0.019, weight A3(ω3)=0.425, weight A4(ω4)=169, and weight A5(ω5)=0.101.

[0100] Combining spatial positioning and correction technology, the piping risk probability P of the dam's spatial positioning points is obtained, and early warning information is automatically broadcast for points with a probability greater than 75% (adjustable).

[0101] Through the above steps, the temporal and spatial location of piping can be verified and early warning can be issued automatically, in real time and efficiently. By automatically comparing multiple monitoring data in parallel, the accuracy of monitoring results can be further improved, efficiency can be significantly improved and the original workload can be reduced.

[0102] The present invention can autonomously inspect and locate the spatial location of piping and further issue early warnings. Its accuracy and effectiveness are qualitatively improved compared to previous manual inspections. It can also greatly save dike inspection funds and personnel investment, significantly reducing the workload of large-scale dike inspections.

[0103] Through modular design and architecture, the front-end radar unit 11 of the piping detection radar device 10 is miniaturized and arranged in an array. The piping audio detector is placed on the top of the self-propelled platform, and an omnidirectional video probe is installed. This can collect real-time panoramic sound, light, and electrical monitoring information of complex field sites, and can monitor multi-physical and intuitive data related to piping to the greatest extent. Based on the data fusion algorithm, it can complete the identification and comparison of piping, significantly improving the speed and accuracy of piping identification under complex working conditions.

[0104] Based on the detected piping data, signal preprocessing is carried out using a deep learning algorithm. Adversarial training is performed in conjunction with a noise sample library to identify and suppress noise. Time series feature sequences are constructed based on feature engineering. Through data enhancement strategies and LSTM model design, combined with noise robustness optimization, high-quality single signal discrimination results can be obtained.

[0105] Multi-signal fusion analysis significantly improves the assessment of the spatial distribution and connectivity of piping, achieving a positioning error of less than 5cm. Deep feature learning and active noise mitigation enable high-precision piping identification under complex working conditions. Experiments have shown that accuracy exceeds 90% when the signal-to-noise ratio is ≥5dB.

[0106] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. An active variable array piping detection radar device, characterized by: It comprises a radar bracket (101) and a plurality of radar units (11) arranged on the radar bracket (101); The radar bracket (101) includes a bracket base plate (120), a transverse telescopic bracket (105) and a longitudinal telescopic bracket (116), wherein the longitudinal telescopic bracket (116) is mounted on the bracket base plate (120), and the longitudinal telescopic bracket (116) includes a longitudinal guide rail (117) and a longitudinal linear bearing (118) slidably connected to the longitudinal guide rail (117), and a longitudinal limiting buckle (119) is provided on the longitudinal linear bearing (118), and the longitudinal linear bearing (118) is fixed relative to the longitudinal guide rail (117) through the longitudinal limiting buckle (119); The transverse telescopic bracket (105) is mounted on a longitudinal linear bearing (118) of the longitudinal telescopic bracket (116). The transverse telescopic bracket (105) includes a transverse guide rail (106) and a transverse linear bearing (107) slidably connected to the transverse guide rail (106). A transverse limiting buckle (108) is provided on the transverse linear bearing (107). The transverse linear bearing (107) is relatively fixed to the transverse guide rail (106) via the transverse limiting buckle (108). Each of the radar units (11) is mounted on a transverse linear bearing (107), and the radar units (11) are slidably adjusted in the longitudinal direction by a longitudinal telescopic bracket (116), and are fixed and limited by a longitudinal limit buckle (119), so that the radar units (11) are multi-level adjusted within the unit spacing in the longitudinal direction; Each radar unit (11) is slidably adjusted in a lateral position by a lateral telescopic bracket (105), and is fixed and limited by a lateral limit buckle (108), and then the radar unit (11) is multi-level adjusted within the unit spacing in the lateral direction; The radar unit (11) is adapted to a complex embankment detection environment with different soil types and heights by setting different transmitting and receiving formations in the longitudinal and transverse directions; the radar unit (11) is deployed in multiple stages in the longitudinal and transverse directions to adapt to different embankment widths and improve the monitoring range and accuracy; Each radar unit (11) includes a DA power amplifier component (151), a signal acquisition AD component (152), an edge computing control and filtering component (153) and an integrated antenna module; the DA power amplifier component (151) is used to convert a baseband digital signal into an analog radio frequency signal and perform power amplification on the signal; the signal acquisition AD component (152) is used to collect the echo analog signal received by the antenna and convert it into a digital signal after conditioning; the edge computing control and filtering component (153) is used for real-time signal processing, system control and noise filtering, and removes noise and clutter through digital filtering to improve the signal-to-noise ratio; and data processing is completed locally, reducing back-end computing pressure and transmission delay. The integrated antenna module is assembled from a built-in three-channel antenna transceiver module, a magnetic flux protection head (111) and a transmitting antenna protection cover (150); the built-in three-channel antenna transceiver module is composed of a radar signal transmitting antenna (112), a radar reflection signal micro receiving antenna (113) and a micro laser displacement transceiver sensor (114) arranged at equal intervals along the longitudinal direction; the radar signal transmitting antenna (112), the radar reflection signal micro receiving antenna (113) and the micro laser displacement transceiver sensor (114) are connected by a shielded cable I (154).

2. The active variable array piping detection radar device according to claim 1, characterized in that: There are 16 radar units (11) distributed on the radar bracket (101) in a regular 4×4 matrix.

3. The active variable array piping detection radar device according to claim 2, characterized in that: The radar units (11) are divided into four groups A1, A2, A3 and A4 in the longitudinal direction. The four groups of radar units (11) use the same frequency, or A1 and A3 are combined, and A2 and A4 are combined, so as to synchronously transmit and collect two groups of signals with different frequency characteristics. The collected signals are grouped and analyzed through digital filtering technology.

4. The active variable array piping detection radar device according to claim 2, characterized in that: The radar units (11) are divided into four groups, B1, B2, B3 and B4, in the horizontal direction. The four groups of radar units (11) use the same frequency, or B1 and B3 are combined, and B2 and B4 are combined, so as to synchronously transmit and collect two groups of signals with different frequency characteristics. The collected signals are grouped and analyzed through digital filtering technology.

5. A real-time piping monitoring device, characterized by: The invention comprises a self-propelled platform (1), an acoustic-optical integrated mast (4) installed on the self-propelled platform (1), and a pipe burst detection radar device (10) according to claim 4; the pipe burst detection radar device (10) is fixedly connected to one end of the self-propelled platform (1) via a flange connector (102), and a roller (103) is installed at the bottom of the pipe burst detection radar device (10); A mast connection seat (2) is fixed to the top wall of the self-propelled platform (1), an acoustic, optical and electrical integrated mast (4) is fixedly mounted on the mast connection seat (2), and an acoustic, optical and electrical signal front-end processing module and a three-proof cover (3) thereof are provided at the bottom of the acoustic, optical and electrical integrated mast (4), which can remove noise and pre-process monitoring data to meet the data analysis preparation requirements of a back-end early warning platform.

6. The real-time piping monitoring device according to claim 5, characterized in that: The acousto-optical integrated mast (4) is provided with a transverse probe module (5), a longitudinal probe module (7), a voiceprint recognition module (6) and a Beidou satellite guidance device (8); The transverse probe module (5) is composed of a visible infrared dual spectrum probe (51) on the back slope side and a visible light probe (52) on the water side; the visible infrared dual spectrum probe (51) and the visible light probe (52) are respectively arranged on the transverse sides of the outer side wall of the top of the acoustic, optical and electrical integrated mast (4); The longitudinal probe module (7) is composed of a forward video probe (71) and a backward video probe (72), and the forward video probe (71) and the backward video probe (72) are respectively arranged on both longitudinal sides of the outer side wall of the top of the acoustic, optical and electrical integrated mast (4); The probe of the voiceprint recognition module (6) is placed at the top of the acoustic-optical-electrical integrated mast (4); the Beidou satellite guidance device (8) is installed on the top of the front outer side wall of the three-proof cover (3) to provide a reference for the positioning of the piping.

7. A method for intelligently identifying piping, characterized by: Using the piping real-time monitoring device according to claim 6, the identification method includes the following steps: S1: adjusting the longitudinal extension width of the variable array piping detection radar device (10) according to the height of the dam to be detected and the detection depth; S2: Based on the width of the dike top to be detected, adjusting the lateral extension width of the variable array piping detection radar device (10); S3: adjusting the ground clearance of the variable array pipe burst detection radar device (10) according to the surface flatness of the detection target; S4: according to the depth and accuracy of pipe burst detection and the inspection speed requirements, the number of operating groups of pipe burst radar units (11) is set, and the spatial layout of the front-end radar units (11) is determined; S5: setting the high-frequency electromagnetic signal transmitting and receiving array of the variable array piping detection radar device (10) to adapt to the complex monitoring environment of different soil types and heights. The piping detection radar device (10) can set the shock wave frequency according to the above array matching; S6: According to the bottom width and detection range of the target dike, the height and angle of the integrated acoustic, optical and electrical mast (4) are adjusted to cover the required specific monitoring surface range, so that the monitoring range of the visible infrared dual spectrum probe (51) covers the potential pipe burst area, so as to obtain visible and infrared image data of the back slope and a certain range of extension, and the audio probe of the voiceprint recognition module (6) can effectively receive the audio signal; S7: The data information collected by the variable array pipe burst detection radar device (10) is pre-processed to form a clear ripple data graph. The frame processing of the signal pre-processing is set to a frame length of 20-40ms. The time-frequency analysis can generate a Mel spectrum graph, and then Z-score normalization processing is performed; S8: A bidirectional LSTM deep model analysis A is performed on the data obtained and pre-processed by the variable array pipe burst detection radar device (10) to capture the previous and next correlation features; in order to improve the analysis efficiency, the analysis introduces an attention mechanism to focus on the key frequency band, uses deep separable convolution to assist feature extraction, and uses improved Sigmoid classification in the output layer; S9: After the ripple data graph is identified by the deep learning algorithm, piping identification data within the dike is generated in real time. The model quickly determines the piping type and penetration based on the piping characteristics, and preliminarily assesses the current status and potential risks of the piping, and assigns a piping risk factor A1. S10: The visible infrared dual spectrum probe (51) collects visible light image information from the back water surface in real time, introduces an environmental adaptive module, dynamically adjusts the suppression parameters, and identifies the piping outlet morphology based on the pre-trained deep learning model B. The denoising algorithm removes the influence of environmental interference and gives the piping risk factor A2; S11: Based on the difference between piping water temperature and the ambient temperature on the back side, the visible infrared dual spectrum probe (51) senses the infrared image within the set range of the back side in real time, and simultaneously performs piping infrared morphology recognition through the pre-trained deep learning model C, judges the possibility of piping, and gives a piping risk factor A3; S12: The acoustic wave probe of the voiceprint recognition module (6) captures the audio signal in real time, and the noise suppression adopts the noise reduction deep learning method, combined with the noise sample library for adversarial training; in the feature engineering construction, the feature sequence selects a 50-100 frame sliding window, and data enhancement is carried out by adding random background noise, time shift / speed change processing and dynamic mixed piping features; noise perturbation is added through adversarial training, multi-task learning and time domain-frequency domain dual verification mechanism to optimize the noise robustness of the analysis method; recognition judgment is carried out by combining the noise removal algorithm with the pre-trained deep learning voiceprint recognition model D, and the piping risk factor A4 is given; S13: Based on the pre-trained model and the hydrological and geomorphological features around the dam, the environmental risk factor A5 is given; S14: Based on the set fusion algorithm, the above five risk factors are weighted and analyzed to obtain the risk value of piping. Combined with spatial positioning and correction technology, the piping risk probability of the spatial positioning point of the dam is obtained, and warning information is automatically broadcast for the points where the risk probability exceeds the set value.

8. The method for intelligently identifying piping according to claim 7, characterized in that: In step S14, the weighted calculation formula of the risk value is: Among them, k is the risk value, A i is the risk factor, ω i is the weight coefficient corresponding to the risk factor.

9. The method for intelligently identifying piping according to claim 8, characterized in that: In step S14, when it is daytime with good visual conditions: The weight coefficients corresponding to each risk factor are: weight A1(ω1)=0.253, weight A2(ω2)=0.382, weight A3(ω3)=0.185, weight A4(ω4)=0.136, weight A5(ω5)=0.044; During night time: The weight coefficients corresponding to each risk factor are: weight A1(ω1)=0.286, weight A2(ω2)=0.019, weight A3(ω3)=0.425, weight A4(ω4)=169, and weight A5(ω5)=0.101.

Images