Pipeline water leakage detection method and system based on multi-mode composite sensing and storage medium

Abstract

The invention discloses a pipeline water leakage detection method and system based on multi-modal composite sensing and a storage medium, and belongs to the technical field of water leakage detection.The pipeline water leakage detection method comprises the steps that increment changes of temperature, pressure and conductance signals arranged on the periphery of a pipeline in an array mode are collected on the basis of a sensor array, and a standardized multi-dimensional feature data set is formed; extracting time sequence signal features of the multi-dimensional feature data by using a long short-term memory (LSTM) network, and extracting spatial distribution features by using a convolutional neural network (CNN); and performing multi-modal fusion on the time sequence signal features and the spatial distribution features through a feature fusion layer, and finally outputting a water leakage state, a water leakage position and a water leakage severity based on a multi-stage classifier or a multi-layer full-connection neural network. According to the method, the fused high-dimensional features integrate the multi-physical field information of water leakage, and the identification capability of tiny leakage and complex environment change is remarkably improved; the method has high robustness, and can effectively suppress the interference of abnormal signals and environmental noise on the judgment result.

Description

Technical Field

[0001] This invention belongs to the technical field of water leakage detection, specifically relating to a method, system, and storage medium for detecting pipeline water leakage based on multimodal composite sensing. Background Technology

[0002] Currently, the detection techniques for leaks in water supply systems include the following: (1) Leakage detection method based on water-absorbing and heat-releasing materials and temperature sensing: By setting a detection layer containing water-absorbing and heat-releasing materials on the outside of the pipe, and using temperature sensors to monitor local temperature changes caused by leakage, leakage detection is achieved. This type of solution usually analyzes temperature changes in time or space to infer the location and status of leakage.

[0003] For example, existing Chinese patent CN101472543B discloses a moisture monitoring system for monitoring moisture in one or more absorbent items, including an input unit for receiving one or more sensor signals indicating the presence of moisture in the absorbent items; a processor for processing the one or more sensor signals and for performing analysis on the signals to describe the characteristics of a moisture event occurring in the absorbent items; and a user interface for communicating with a system user.

[0004] (2) A leakage detection method based on distributed optical fiber temperature sensing, which uses a temperature-sensing optical fiber fixed to the outer layer of a pipe and utilizes the signal change when the built-in water-absorbing and heat-releasing material changes temperature to achieve real-time monitoring of pipe leakage. For example, existing Chinese patent CN223105855U discloses a buried water supply pipeline leakage detection device based on distributed optical fiber sensing, including a temperature-sensing optical fiber, a laying device, and a reaction block. The laying device is a double-layer sleeve structure, with an inner layer being a water-permeable layer and an outer layer being a waterproof layer. The inner diameter of the water-permeable layer is equal to the outer diameter of the water supply pipeline. Equally spaced spiral optical fiber grooves are provided between the water-permeable layer and the waterproof layer, and the temperature-sensing optical fiber is fixed in the spiral optical fiber grooves. The reaction block is made of a material that reacts with water and releases heat, and is fixed in the water-permeable layer, with one reaction block provided between every two adjacent optical fiber grooves. The optical fiber is fixed to the surface of the water supply pipeline to be tested by the laying device, and the other end is connected to the DTS temperature sensing host. When the water supply pipeline leaks, it reacts with the reaction block in the seepage layer, releasing heat. The temperature-sensing optical fiber transmits the detected signal to the DTS temperature measurement host, thereby monitoring the leakage of the water supply pipeline in real time.

[0005] (3) A temperature-sensing leak monitoring platform based on a signal amplification layer is used to enhance the temperature change signal by setting a reaction layer outside the detection pipe to improve the sensitivity and accuracy of leak detection. For example, the existing Chinese patent CN107167263B discloses a water pipe leak detection experimental platform based on a fiber optic Raman temperature sensor. It includes a water circulation system, a signal amplification system and a fiber optic Raman temperature sensor. The water circulation system simulates the leakage of urban water pipes in a laboratory environment, and the temperature signal is further amplified by the conversion layer set outside the detection pipe to facilitate detection by the fiber optic Raman temperature sensor.

[0006] (4) Water pressure sensor method: Pressure sensors are installed on the main pipeline or at key nodes to monitor pressure changes in the pipeline in real time. Pressure anomalies caused by water leakage are captured by the data acquisition and analysis module for leakage alarm. This solution is suitable for overall system anomaly monitoring.

[0007] (5) Humidity sensor method: Install a humidity sensor array outside the pipe or at potential leak points to detect increases in local ambient humidity in real time to determine if there is a leak. The signal acquisition and alarm unit is responsible for processing the data and triggering the alarm.

[0008] However, the above method has the following drawbacks: 1) The single sensing mode limits the accuracy of detection and has low sensitivity to detect minor leaks. Water absorption and heat release temperature monitoring, pressure monitoring, or humidity monitoring are all single physical quantities. If leak detection is based solely on a single feature such as temperature change, pressure fluctuation, or humidity increase, the detection accuracy will be insufficient.

[0009] On the one hand, single sensor signals are easily affected by fluctuations in ambient temperature, water pressure, and natural changes in humidity, leading to frequent false alarms and missed alarms. On the other hand, these technologies struggle to capture subtle signal changes caused by minute leaks, especially in cases of concealed or early-stage leaks, making timely and accurate detection difficult. For example, pressure sensors struggle to capture the slight pressure changes caused by small-flow leaks, and the pressure signal is delayed, making it impossible to accurately pinpoint the leak location. While pressure sensor methods are suitable for overall pressure anomaly alarms, their leak location capability is weak. Secondly, humidity sensors are densely deployed and have high maintenance costs: humidity sensors need to be directly exposed to humid environments, have limited coverage areas, and are easily damaged by prolonged immersion, resulting in complex and costly system maintenance. 2) Lack of multimodal feature fusion and intelligent analysis: Existing technologies do not combine multi-dimensional features such as temperature, pressure and conductivity for joint intelligent analysis, resulting in insufficient accuracy in water leakage event identification and location; Insufficient system linkage and response capabilities: Most solutions lack intelligent alarm linkage mechanisms, which cannot achieve automatic identification and rapid response to water leakage events, affecting the safety assurance of practical applications. Summary of the Invention

[0010] The purpose of this invention is to provide a pipeline leakage detection method, system, and storage medium based on multimodal composite sensing, in order to solve the above-mentioned problems.

[0011] This invention is mainly achieved through the following technical solutions: A pipeline leakage detection method based on multimodal composite sensing includes the following steps: Step S1: Based on the sensor array, the incremental changes of temperature, pressure and conductivity signals around the pipe are collected to obtain the temperature increment ΔT, pressure increment ΔP and conductivity increment Δσ arranged in an array; then, data preprocessing is performed to form a standardized multidimensional feature dataset; Step S2: Based on a standardized multidimensional feature dataset, a fusion analysis model is used to identify leakage feature patterns; Step S21: Extract time series signal features; Based on the time series data of temperature increment ΔT, pressure increment ΔP and conductivity increment Δσ, the Long Short-Term Memory (LSTM) network is used to extract the time correlation features of the multidimensional feature data, and capture the dynamic laws and abrupt changes of temperature, pressure and conductivity. Step S22: Extract spatial distribution features; Based on the spatial distribution data of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ, a convolutional neural network (CNN) is used to extract two-dimensional or three-dimensional features; Step S23: Multimodal feature fusion; The temporal signal features from step S21 and the spatial distribution features from step S22 are fused in a multimodal manner through a feature fusion layer; Step S24: Finally, based on a multi-level classifier or a multi-layer fully connected neural network, output the leakage status, leakage location, and leakage severity.

[0012] To better realize the present invention, further, in step S21, firstly, the time-series data of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ are normalized and time-synchronized; then, the processed time-series data are input into a Long Short-Term Memory (LSTM) network. The LSTM network captures the abrupt changes caused by the leakage event through memory units, while retaining trend information over a long time scale, thus taking into account both short-term drastic changes and long-term variations; finally, a time feature vector containing the dynamic patterns and time dependencies of each channel at different times is output.

[0013] To better implement this invention, further, in step S22, firstly, the collected one-dimensional spatial distribution data of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ are aligned and superimposed, so that a spatial location x simultaneously contains the detected values ​​of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ, forming a data matrix; then, the spatial distribution characteristics of leakage are obtained based on convolutional neural network (CNN) processing. Specifically, the spatial location x corresponds to the sequence number position of the point sensor in each sensor array.

[0014] To better realize the present invention, the feature fusion layer is further defined as an attention mechanism fusion module, and the weights of each feature channel are dynamically adjusted based on the attention mechanism.

[0015] To better implement this invention, further, in step S24, the spatiotemporal features fused by the feature fusion layer are input into the MLP (Multilayer Perceptron) to obtain the leakage status, leakage location, and leakage severity; the MLP includes three parallel classification branches: The leakage status branch outputs the probability distributions for no leakage and leakage. The leak location branch outputs the coordinates of the leak location of the sample. The leakage severity branch outputs the probability distribution of minor, moderate, and severe leaks.

[0016] To better realize the present invention, it further includes step S3: outputting the two-dimensional / three-dimensional coordinates of the leak point based on the leak location and sensor location information.

[0017] This invention is mainly achieved through the following technical solutions: A pipeline leakage detection system based on multimodal composite sensing, implemented according to the aforementioned pipeline leakage detection method based on multimodal composite sensing, includes a material layer module, a sensor array module, and a control and processing module. The control and processing module includes a data acquisition and preprocessing module and an intelligent analysis and recognition module. The sensor array module is disposed inside the material layer module and is connected to the data acquisition and preprocessing module. The data acquisition and preprocessing module is used to acquire temperature, pressure, and conductivity signals from the sensor array module in real time and perform preprocessing to form a standardized multidimensional feature dataset. The intelligent analysis and recognition module is used to identify leakage feature patterns based on the multidimensional feature dataset using a fusion analysis model. The material layer module includes, from bottom to top, a bottom layer, a moisture-absorbing and heat-releasing expansion layer, and an outer protective layer; the sensor array module includes, from bottom to top, a temperature sensor array, a pressure sensor array, and a conductivity sensor array; the moisture-absorbing and heat-releasing expansion layer includes adsorbed salt and an expansion material, wherein the adsorbed salt reacts with water to produce an exothermic reaction and transfers heat to the temperature sensor array; the expansion material absorbs water and expands to generate mechanical stress, which directly acts on the pressure sensor array; the conductivity sensor array changes its resistance and generates a conductivity characteristic signal after encountering water.

[0018] To better realize the present invention, the adsorbent salt is calcium chloride, the expansion material is cross-linked sodium polyacrylate, and the mass ratio of calcium chloride to cross-linked sodium polyacrylate is 4:1.

[0019] To better realize the present invention, it further includes an alarm and linkage module connected to the control processing module, the alarm and linkage module being used to trigger multi-channel alarms.

[0020] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for detecting pipe leaks based on multimodal composite sensing.

[0021] The beneficial effects of the present invention are as follows: (1) This invention forms a standardized multidimensional feature dataset based on the incremental changes in temperature, pressure, and conductivity signals. It employs a Long Short-Term Memory (LSTM) network to extract the temporal dependence of the multidimensional feature data, capturing the dynamic patterns and abrupt changes in temperature, pressure, and conductivity. A Convolutional Neural Network (CNN) is used to convolve the spatial distribution data of the sensor array, extracting spatial patterns of leakage heat, pressure distribution, and conductivity changes to assist in locating leakage points and areas. Then, a feature fusion layer fuses the temporal and spatial distribution features in a multimodal manner, inputting the fusion into a classifier to complete the state recognition, location estimation, and severity classification of the leakage event. The high-dimensional features fused in this invention integrate multi-physics information about the leakage, significantly improving the ability to identify minute leaks and complex environmental changes. Simultaneously, this invention exhibits high robustness, effectively suppressing the interference of abnormal signals and environmental noise on the judgment results.

[0022] (2) This invention integrates the exothermic reaction of adsorbed salt, pressure sensing of expanding materials, and conductivity changes of conductive fine iron wires through an innovatively designed multi-functional composite material layer, achieving multi-physics sensitive capture of water leakage events and significantly improving the sensitivity and accuracy of water leakage detection. Compared with traditional single-sensor methods, this invention effectively overcomes the technical difficulties of detecting small water leaks, accurately locating water leaks, and high false alarm rates caused by environmental interference. Secondly, based on a fusion analysis model, this invention uses machine learning and deep learning algorithms to jointly analyze multimodal time-series data, realizing intelligent discrimination of water leakage status, location, and severity, effectively distinguishing water leakage signals from environmental interference, and improving the robustness and adaptability of the system.

[0023] (3) This invention further constructs a real-time alarm mechanism, which can ensure a rapid response and accurate alarm when a water leakage event occurs. This invention supports linkage with pipeline control devices, such as automatically closing valves and notifying maintenance personnel, improving the efficiency of water leakage handling. This invention has the advantages of flexible deployment, fast response speed, high detection accuracy, and low maintenance cost, meeting the needs of large-scale applications in multiple scenarios such as homes, businesses, and industries. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating the principle of the pipeline leakage detection method based on multimodal composite sensing of the present invention. Figure 2 The simulated temperature change curve ΔT-t, pressure change curve ΔP-t, and conductivity change curve Δσ-t are shown in Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the principle of the pipeline leakage detection system based on multimodal composite sensing of the present invention. Figure 4 This is a schematic diagram illustrating the principle of the linkage alarm of the present invention; Figure 5 This is a schematic diagram of the moisture-absorbing and heat-releasing expansion layer and the sensor array module.

[0025] Wherein: 1-bottom layer, 2-moisture-absorbing and heat-releasing expansion layer, 3-sensor array module, 4-outer protective layer. Detailed Implementation

[0026] Example 1: A pipeline leakage detection method based on multimodal composite sensing identifies leakage characteristic patterns based on a fusion analysis model, such as... Figure 1 As shown, the fusion analysis model in this invention includes a Long Short-Term Memory (LSTM) network and a Convolutional Neural Network (CNN) arranged in parallel, along with a multi-level classifier. This invention performs intelligent analysis based on multi-dimensional sensor data, employing advanced machine learning and deep learning technologies to achieve accurate identification, location, and severity assessment of water leakage events. Specifically, it includes the following steps: (1) Multi-dimensional data collection and preprocessing; This invention utilizes a sensor array composed of multiple types of sensing units integrated within a composite material layer to collect multi-dimensional physical signals such as temperature, pressure, and conductivity in real time. By fusing these multi-physical field signals (temperature, pressure, conductivity, etc.) and performing preprocessing such as denoising, normalization, and synchronization, a unified format multi-dimensional spatiotemporal dataset is formed, enabling sensitive detection and accurate identification of water leakage events. Specific acquisition features include: 1) Temperature change characteristics; Based on the exothermic reaction between adsorbed salt and leaking water, the temperature sensor captures the instantaneous temperature rise and calculates the incremental temperature change (ΔT) to reflect the occurrence and intensity of the leak.

[0027] 2) Pressure change characteristics; The local pressure changes caused by the water-absorbing and expanding material on the pipeline are measured by a pressure sensor, forming a pressure increment (ΔP), which provides a basis for judging the severity of the leak.

[0028] 3) Characteristics of conductivity variation; The conductive fine iron wires embedded in the material layer exhibit high resistance when dry, but the resistance drops sharply when exposed to water. The system can quickly detect water leakage by monitoring the change in conductivity (Δσ) in real time.

[0029] The characteristics of the above three dimensions are derived by calculating the increment, that is, the total change relative to the baseline value before the leak. Multi-channel sensor data collected by the system at a certain time before the leak (e.g., the past minute) can be used to calculate the "normal baseline" under the current environment through moving average or median filtering. For example, if the temperature sensor's average value over the past 60 seconds is 25.0℃, then 25℃ is used as the baseline. If the temperature increases to 30.0℃ due to water absorption and heat release, then ΔT is 5. Therefore, after real-time data acquisition, the algorithm module performs a simple calculation.

[0030] The calculation of pressure increment ΔP and conductivity increment Δσ follows the same principle. This yields changes relative to the current environmental conditions, rather than rigid absolute values. Furthermore, it adapts to environmental changes such as seasons, day / night cycles, and air conditioning operation. Under normal conditions, the data changes during a leak are as follows: The temperature increment ΔT can generally reach 8~20℃ or even higher, and the change is particularly obvious; The pressure increment ΔP causes a significant pressure change in the pipeline due to water absorption and expansion, typically ranging from 5 to 50 kPa. The conductivity increment Δσ may jump from an extremely low state to a high conductivity state, exhibiting a step-like abrupt change.

[0031] All raw signals undergo filtering, noise reduction, drift correction, normalization, and time synchronization by the acquisition and preprocessing module, forming a standardized multidimensional feature dataset.

[0032] (2) Subsequently, the data is input into the fusion analysis model; For time-series signals, such as temperature change curve ΔT-t, pressure change curve ΔP-t, and conductivity change curve Δσ-t, a Long Short-Term Memory (LSTM) network is used to extract time-related features. Specifically, this invention utilizes a Temporal Convolutional Network (TCN) and a Long Short-Term Memory (LSTM) network to extract the time dependence of sensor data, capturing the dynamic patterns and abrupt changes in temperature, pressure, and conductivity.

[0033] For spatial distribution data, the spatial variation curves of temperature (Tx), pressure (Px), and conductivity (σ-x) are extracted using a convolutional neural network (CNN) for two-dimensional or three-dimensional feature extraction. Specifically, this invention uses a CNN to perform convolution operations on the spatial distribution data of the sensor array to extract the spatial patterns of leakage heat, pressure distribution, and conductivity changes, thereby assisting in locating the leakage point and area.

[0034] Finally, a feature fusion layer integrates temporal and spatial distribution features in a multimodal manner, and inputs this into a classifier to complete the state identification (whether there is a leak), location estimation (coordinates or area of ​​the leak point), and severity classification (minor leak / moderate / severe). Specifically, the feature fusion layer jointly learns temporal and spatial distribution features, dynamically adjusting the weights of each feature channel through an attention mechanism to improve the model's sensitivity to key signals and recognition accuracy. The fused high-dimensional features integrate multi-physics information about the leak, significantly improving the ability to identify minor leaks and complex environmental changes; at the same time, this layer has a certain degree of robustness, effectively suppressing the interference of abnormal signals and environmental noise on the judgment results.

[0035] Based on deep feature representation, the system uses a multi-layer fully connected neural network or a lightweight classifier (e.g., XGBoost, Random Forest) to perform binary classification (leaking / normal) of leakage status and multi-level severity classification (minor leak, moderate leak, severe leak). The classifier is trained on a large number of labeled leakage datasets, supporting continuous iterative optimization of the model to improve generalization ability. It can also be combined with a threshold strategy to achieve a balance between sensitivity and false alarm rate, ensuring stable system operation.

[0036] Specifically, such as Figure 1 As shown, the fusion analysis model includes the following steps: Step A1: Time series signal feature extraction; For three types of time-series signals—temperature change curve ΔT-t, pressure change curve ΔP-t, and conductivity change curve Δσ-t—normalization and time synchronization processing are first performed (i.e., the time is adjusted to be consistent so that the three detection values ​​are at the same time). Then, the signals are input into a Long Short-Term Memory (LSTM) network to extract time-related features.

[0037] LSTM captures the abrupt changes caused by leakage events through its memory cells (e.g., instantaneous peak of ΔT, step increase of ΔP, sudden jump of Δσ); at the same time, it retains trend information over long time scales (e.g., slow signal decline, sustained high voltage range), thus taking into account both short-term drastic changes and long-term variations; the final output time feature vector contains the dynamic patterns and time dependencies of each channel at different times.

[0038] Step A2: Spatial distribution feature extraction; First, there are three curves representing Tx, Px, and σ-x that vary with location, representing one-dimensional spatial distribution data. Assume there are n spatial locations (x1, x2, ..., x...). n Let Tx be the temperature increment ΔT, Px be the pressure increment ΔP, and σ be the conductivity increment Δσ. Then Tx, Px, and σ-x are all vectors of length n. Superimposing these three vectors results in: Temperature increment per layer: ΔT1 ΔT2 ΔT3 ... ΔTn Second-layer pressure increment: ΔP1 ΔP2 ΔP3 ... ΔPn The conductivity increments of the three layers are: Δσ1 Δσ2 Δσ3 ... Δσn. The above superposition forms a 3-row, n-column "matrix," where 3 represents the number of channels and n represents the number of spatial locations. When the CNN convolutional kernel scans this matrix, it's equivalent to simultaneously examining three physical quantities at the same location and observing how they change in adjacent locations, thus obtaining the spatial distribution characteristics of the leak. For example, leak location features or leak range features.

[0039] Step A3: Multimodal feature fusion; The temporal feature vector output by LSTM and the spatial distribution feature vector output by CNN are concatenated and weighted in a feature fusion layer (e.g., a fully connected layer or an attention mechanism fusion module). The attention mechanism fusion module dynamically allocates the weights of different feature channels based on the attention mechanism. For example, in the initial leakage stage, the temporal feature has a higher weight, while in the diffusion stage, the spatial distribution feature has a higher weight.

[0040] For example, suppose the temporal feature vector ft=[0.2,0.5,0.1] output by LSTM and the spatial distribution feature vector fs=[0.3,0.7,0.4] output by CNN are fused by a feature fusion layer to obtain the spatiotemporal feature ffusion=[0.2,0.5,0.1,0.3,0.7,0.4]. The fused vector contains both temporal evolution information and spatial distribution information, which can be directly input into a classifier for leak identification, location estimation, and severity judgment.

[0041] A4: Classification tasks; The fused spatiotemporal features are input into an MLP (Multilayer Perceptron), which outputs three tensors representing three classification tasks. Specifically, the MLP includes: The leak status branch (status_pred) outputs two values, representing the probability distribution of "no leak" and "leaking", with the classification result being the event with the highest probability. The leak location branch (position) is converted into a probability without going through an activation function (such as softmax), and outputs two sample coordinates. Therefore, it also outputs two values, representing the leak location coordinates [x1, x2] of each sample. The severity branch (severity_pred) outputs three values, representing the probability distribution of "minor leak / moderate / severe", with the classification result being the severity level with the highest probability.

[0042] like Figure 1 As shown, the loss functions for the leakage status branch and the leakage severity branch are cross-entropy loss functions, and the loss function for the leakage location branch is the mean squared error loss function. The loss function of the MLP multilayer perceptron is the sum of the loss functions of the three branches.

[0043] The MLP (Multilayer Perceptron) was trained using simulated data. The model was optimized through backpropagation by jointly calculating the losses of each task and controlling the activation and deactivation of the softmax activation function. This demonstrated that the MLP model can simultaneously predict three tasks. The MLP model was tested using simulated data. As shown in Table 1, in the leak status column, 1 indicates a leak, and 0 indicates no leak. Analysis, as shown in Table 2, shows that the performance report of the MLP model indicates a leak status detection accuracy of 92% and a leak severity detection accuracy of 83%, indicating usable prediction information. This simulation proves that the MLP model can indeed classify "leak status, leak location, and leak severity" using a large amount of diverse feature data. Further iterative learning with large amounts of data can refine the model's accuracy.

[0044] Table 1 Comparison of Model Predictions After Initial Simulation Training (Table 1)

[0045] Table 2 Performance Report of the MLP Multilayer Perceptron Model

[0046] (3) Estimation of the location of the leak; Combining the leak location coordinates and sensor location information output above, a regression model or localization network can be used to output the two-dimensional / three-dimensional coordinates of the leak point. Through model inversion technology, the spatial distribution of the sensor signals is mapped back to the actual pipeline location, achieving precise localization. The regression model, localization network, and model inversion technology mentioned here are all existing technologies and will not be elaborated further.

[0047] This invention can automatically update model parameters based on continuously collected operational data, adapting to different pipeline environments and sensor characteristic changes. Through transfer learning and semi-supervised learning techniques, it can effectively expand the model's application scope using a small amount of labeled data; achieving intelligent maintenance and performance self-optimization, extending system lifespan and reducing maintenance costs.

[0048] As shown in Table 3, to verify the effect of multi-dimensional feature fusion, we simulated data under multiple scenarios, including: T1 micro-leakage (1-3 drops per minute, located close to the sensor); T2 has moderate leakage (thin water flow, flow rate 0.1-0.3 L / min, location close to the sensor); T3 is severely leaking (continuous water flow, >1 L / min); T4 remote micro-leakage (leak point is far from sensor, signal is weak); T5 is in normal condition (no change in conductivity or pressure, and will be ruled out); T6 Temperature Fluctuation False Signal (No water leakage, but significant change in external temperature).

[0049] like Figure 2 As shown in Table 4, the present invention simulated the temperature change curve ΔT-t, the pressure change curve ΔP-t, and the conductivity change curve Δσ-t, respectively. The analysis results based on the simulation data of the present invention are accurate and have high confidence levels.

[0050] Table 3 Simulated Scene Information Scene ΔT (°C) Δσ(mS / cm) ΔP(kPa) illustrate T1 6~8 0→50 mutation 5~10 There was significant heat release, a jump in electrical conductivity, and a slight increase in pressure. T2 10~12 0→150 mutation 15~30 Significant heat release, substantial increase in electrical conductivity, and marked increase in pressure. T3 12~18 0→300 near short circuit 30~50 Extremely strong heat release, electrical conductivity approaching short circuit, and pressure significantly increased. T4 4~6 0→20 slight change 2~5 The signal weakens, but it can still be captured by the algorithm. T5 2~3 0 0 Without electrical conductivity and pressure change, it will be excluded. T6 12+6 300+100 40+20 AI can distinguish superimposed signals. Table 4. Analysis Information Based on Simulation Data

[0051] Example 2 A pipeline leakage detection system based on multimodal composite sensing, such as Figure 3As shown, the system includes a material layer module, a sensor array module 3, and a control processing module. The control processing module includes a data acquisition and preprocessing module and an intelligent analysis and recognition module. The sensor array module 3 is located inside the material layer module and is connected to the data acquisition and preprocessing module. The data acquisition and preprocessing module is used to acquire temperature, pressure, and conductivity signals from the sensor array module 3 in real time and perform preprocessing to form a standardized multidimensional feature dataset. The intelligent analysis and recognition module is used to identify leakage characteristic patterns based on the multidimensional feature dataset using a fusion analysis model.

[0052] like Figure 5 As shown, the material layer module includes a bottom layer 1, a moisture-absorbing and heat-releasing expansion layer 2, and an outer protective layer 4 arranged sequentially from bottom to top; the material layer module is directly attached to the outer wall of the pipe to be monitored and is responsible for sensing the physical changes (including temperature changes, pressure changes, and conductivity changes) caused by water leakage.

[0053] The sensor array module 3 includes a temperature sensor array, a pressure sensor array, and a conductivity sensor array arranged sequentially from bottom to top inside the material layer module; the sensor array module 3 can achieve high-density acquisition of multi-point, multi-dimensional data, improving the spatial resolution and sensitivity of leakage detection.

[0054] The moisture-absorbing and heat-releasing expansion layer 2 includes adsorbed salt and an expansion material. The adsorbed salt reacts exothermically with water and transfers heat to the temperature sensor array. The expansion material absorbs water and expands, generating mechanical stress that directly acts on the pressure sensor array. The conductivity sensor array includes conductive fine iron wires (e.g., nickel-plated copper-clad iron wires). The resistance of the conductive fine iron wires changes upon contact with water, generating a conductivity characteristic signal. Preferably, the conductive fine iron wires can be embedded inside the moisture-absorbing and heat-releasing expansion layer 2.

[0055] Specifically, such as Figure 5 As shown, this invention deploys a multifunctional composite structure at key monitoring locations on the pipeline, consisting of a bottom layer 1, a hygroscopic and exothermic expansion layer 2 (a mixture of adsorbed salt and expansion material), an embedded temperature sensor array, an embedded pressure sensor array, a conductivity sensor array, and an outer protective layer 4. When a leak occurs, the adsorbed salt reacts exothermically with water, and the temperature change is transmitted to the temperature sensor via the composite material, enabling rapid capture of the temperature signal. The expansion material absorbs water and expands, generating mechanical stress that directly acts on the pressure sensor, acquiring pressure change characteristics. The conductive wire undergoes a significant change in resistance (short circuit or continuity) upon contact with water, generating a conductivity characteristic signal. This integrated material layer simultaneously provides three independent sensing channels: temperature, pressure, and conductivity, forming a multi-physical feature fusion detection mechanism. This allows the system to achieve a sufficiently high signal-to-noise ratio even under conditions of minor leaks, thereby improving the accuracy and robustness of leak detection.

[0056] Preferably, the bottom layer 1 is a non-woven fabric, and the outer protective layer 4 is a polyurethane (PU) film. The adsorbent salt and the expanding material are calcium chloride (CaCl2) and cross-linked sodium polyacrylate (SAP), respectively, and are mixed together in a mass ratio of approximately 4:1.

[0057] The role of CaCl2 is to release heat immediately upon absorbing water, allowing the temperature sensor to quickly detect changes; the role of SAP is to rapidly expand in volume after absorbing water, directly providing a measurable "push" to the pressure sensor. The 4:1 mass ratio is based on a comprehensive consideration of the reaction rates and signal contribution proportions of the two materials: calcium chloride has a strong exothermic reaction, but its duration is shorter than that of the expansion material, and the heat release is short-lived, making it suitable for providing instantaneous temperature peaks; the water absorption process of the expansion material is relatively slow, but it can maintain its expansion for a longer period, and once expanded, it may remain expanded indefinitely. This ratio ensures both rapid generation of a temperature signal in the early stages of leakage and maintenance of a continuous pressure signal in the later stages of the reaction, thus achieving complementary and enhanced multi-channel signals throughout the entire leakage process.

[0058] Preferably, the data acquisition and preprocessing module is responsible for receiving, digitizing, and initially organizing the temperature, pressure, and conductivity signals collected by the sensor array in real time; at the same time, it performs data filtering, noise reduction, calibration, and feature extraction to ensure that the data input to the intelligent analysis and recognition module has high accuracy and high reliability.

[0059] The intelligent analysis and identification module performs unified modeling and correlation analysis of multi-dimensional signals such as temperature, pressure, and conductivity, and uses a fusion analysis model to identify leakage characteristic patterns. It can simultaneously analyze time-series features (temperature / pressure change rate, duration, etc.) and spatial distribution features (sensor array response location, etc.), thereby achieving accurate determination of leakage events. The machine learning model is trained through historical data accumulation and simulated data generation, supporting periodic updates or online learning to adapt to environmental changes. It supports combining user feedback or manual correction mechanisms to continuously optimize the accuracy of leak identification and location.

[0060] Specifically, the control processing module (including the main control chip and integrated circuit board) is designed with a compact box structure, facilitating installation in access panels, equipment cabinets, or shallow wall grooves, and possesses excellent environmental adaptability and heat dissipation performance. For already renovated environments, it is recommended to reserve small maintenance channels in concealed locations such as wall corners, baseboards, ceilings, and pipe shafts for wiring, replacement of adsorbent materials, and sensor maintenance, ensuring long-term reliable operation and convenient maintenance of the system.

[0061] Preferably, such as Figure 3As shown, it also includes an alarm and linkage module connected to the control processing module, which is used to trigger multi-channel alarms. The alarm and linkage module includes existing smart home central control, local sound and light alarms, and a maintenance module. Figure 4 As shown, if a water leak is detected, the control module can issue a water shut-off control command to close the solenoid valve, activate the audible and visual alarm, remotely push alarm information to the user terminal, and send maintenance information to maintenance personnel, ensuring that maintenance personnel can be informed of the water leak event in the shortest possible time. Specifically, the control module supports local and remote communication functions and can automatically link with the pipeline control system, emergency valves, fire protection system, and building management system to realize emergency valve closure, start drainage equipment, or trigger other emergency measures, thereby effectively reducing property damage and safety risks caused by water leaks.

[0062] Preferably, concealed access ports can be reserved in inconspicuous locations such as wall panels, cabinets, or baseboards. These access ports are mainly used to facilitate quick repair of leaks by on-site personnel when a pipe leaks.

[0063] The core principle of this invention lies in achieving sensitive detection of water leakage events through a specially designed material layer structure, and combining it with machine learning / deep learning analysis of multi-dimensional sensing features to significantly improve the accuracy and robustness of water leakage identification; at the same time, it is combined with intelligent alarm and linkage mechanisms to achieve fast and accurate response control.

[0064] This system adopts a modular and flexible design, facilitating rapid deployment in various practical scenarios. Its applicability covers the synchronous pre-installation of new pipelines and the later renovation and upgrade of existing buildings. The core monitoring component of this system is a multifunctional composite flexible strip with an integrated moisture-absorbing and heat-releasing expansion layer 2, a temperature sensor array, a pressure sensor array, and a conductivity sensor array, with a thickness of approximately 0.5 cm. This flexible strip is lightweight, flexible, and highly adhesive, allowing it to adhere tightly to the outer wall of hot and cold water pipes, enabling multi-dimensional synchronous signal sensing. It offers various installation methods, including adhesive bonding, binding, clipping, or nesting fixation, ensuring no damage to the original pipe structure. It adapts to various pipe diameters and bending angles, guaranteeing good contact with the pipe surface, thereby improving sensing sensitivity and response speed.

[0065] This invention prioritizes the deployment of this flexible strip at key nodes (such as pipe corners, interfaces, and ends) in high-leakage areas (e.g., bathrooms, kitchens), forming a three-dimensional, distributed, multi-point sensing network. Each sensor segment is connected in series via a bus to a unified main control module, enabling centralized data acquisition and unified intelligent management, thus improving the overall stability and scalability of the system. Based on a non-destructive, flexible composite structure and combined with a multi-dimensional sensor array, this system achieves highly sensitive and accurate leak monitoring and intelligent early warning. Its distributed, modular design balances ease of installation, environmental adaptability, and aesthetics, providing an ideal solution for intelligent early warning systems in new buildings and a practical technical path for the intelligent transformation of existing buildings.

[0066] The fields in which this invention can be applied include: (1) Centralized residential buildings such as hotels and apartments: The system of this invention is installed around the centralized water supply pipelines in the building and the sanitary facilities of each household. It can collect multi-dimensional sensor data such as temperature, pressure, and conductivity in real time and upload the data to the property management system. This invention achieves accurate location and classification of water leakage events by integrating intelligent recognition algorithms, and promptly triggers multi-level linkage alarms. The system can be linked with the fire protection and emergency response system to quickly execute automatic valve shut-off and personnel notification, effectively preventing structural seepage and wall mold caused by chronic water leakage, reducing maintenance costs, and realizing large-area distributed water leakage monitoring and centralized management.

[0067] (2) Industrial plants and production workshops: Applicable to leakage detection of key nodes in the internal water supply and cooling pipe network of factories, densely equipped areas, and underground concealed pipes. This invention can simultaneously sense temperature rise, pressure change, and electrical short circuit, identify leakage events in real time and trigger alarms, and support automatic linkage to close relevant valves. The system can be integrated into an industrial automation platform to provide remote monitoring and maintenance, improve equipment safety and production continuity, and reduce equipment damage and safety hazards caused by leakage.

[0068] (3) Public buildings such as hospitals and schools: This system is deployed in areas with high water usage and high personnel flow, such as dormitories and restrooms, to support accurate identification and rapid alarm of local water leaks. This invention can reduce the false alarm rate and ensure stable system operation through multi-sensor fusion algorithms. This invention connects to the building management system to realize regional water leak risk classification management, ensuring timely detection and handling of water leak hazards in the event of long-term unattended operation, and protecting public safety and building structural safety.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A pipeline leakage detection method based on multimodal composite sensing, characterized in that, Includes the following steps: Step S1: Based on the sensor array, collect the incremental changes of temperature, pressure and conductivity signals around the pipe to obtain the temperature increment ΔT, pressure increment ΔP and conductivity increment Δσ arranged in an array; Then, data preprocessing is performed to form a standardized multidimensional feature dataset; Step S2: Based on a standardized multidimensional feature dataset, a fusion analysis model is used to identify leakage feature patterns; Step S21: Extract time series signal features; Based on the time series data of temperature increment ΔT, pressure increment ΔP and conductivity increment Δσ, the Long Short-Term Memory (LSTM) network is used to extract the time correlation features of the multidimensional feature data, and capture the dynamic laws and abrupt changes of temperature, pressure and conductivity. Step S22: Extract spatial distribution features; Based on the spatial distribution data of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ, a convolutional neural network (CNN) is used to extract two-dimensional or three-dimensional features; Step S23: Multimodal feature fusion; The temporal signal features from step S21 and the spatial distribution features from step S22 are fused in a multimodal manner through a feature fusion layer; Step S24: Finally, based on a multi-level classifier or a multi-layer fully connected neural network, output the leakage status, leakage location, and leakage severity.

2. The pipeline leakage detection method based on multimodal composite sensing according to claim 1, characterized in that, In step S21, firstly, the time-series data of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ are normalized and time-synchronized; then, the processed time-series data are input into a Long Short-Term Memory (LSTM) network, which captures the abrupt changes caused by the leakage event through memory units while retaining trend information over a long time scale; finally, a time feature vector containing the dynamic patterns and time dependencies of each channel at different times is output.

3. The pipeline leakage detection method based on multimodal composite sensing according to claim 1, characterized in that, In step S22, firstly, the one-dimensional spatial distribution data of the collected temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ are aligned and superimposed so that a spatial location x simultaneously contains the detected values ​​of temperature increment ΔT, pressure increment ΔP, and conductivity increment Δσ, forming a data matrix; then, the spatial distribution characteristics of the leakage are obtained based on the processing of a convolutional neural network CNN.

4. The pipeline leakage detection method based on multimodal composite sensing according to claim 1, characterized in that, In step S23, the feature fusion layer is an attention mechanism fusion module, and the weights of each feature channel are dynamically adjusted based on the attention mechanism.

5. The pipeline leakage detection method based on multimodal composite sensing according to claim 4, characterized in that, In step S24, the spatiotemporal features fused by the feature fusion layer are input into the MLP (Multilayer Perceptron) to obtain the leakage status, leakage location, and leakage severity. The MLP includes three parallel classification branches: The leakage status branch outputs the probability distributions for no leakage and leakage. The leak location branch outputs the coordinates of the leak location of the sample. The leakage severity branch outputs the probability distribution of minor, moderate, and severe leaks.

6. The pipeline leakage detection method based on multimodal composite sensing according to claim 1, characterized in that, It also includes step S3: based on the location of the leak and the sensor location information, output the two-dimensional / three-dimensional coordinates of the leak point.

7. A pipeline leakage detection system based on multimodal composite sensing, implemented based on the pipeline leakage detection method based on multimodal composite sensing as described in any one of claims 1-6, characterized in that, It includes a material layer module, a sensor array module, and a control processing module. The control processing module includes a data acquisition and preprocessing module and an intelligent analysis and recognition module. The sensor array module is located inside the material layer module and is connected to the data acquisition and preprocessing module. The data acquisition and preprocessing module is used to acquire temperature, pressure, and conductivity signals from the sensor array module in real time and perform preprocessing to form a standardized multidimensional feature dataset. The intelligent analysis and recognition module is used to identify leakage feature patterns based on a multi-dimensional feature dataset and a fusion analysis model. The material layer module includes, from bottom to top, a bottom layer, a moisture-absorbing and heat-releasing expansion layer, and an outer protective layer; the sensor array module includes, from bottom to top, a temperature sensor array, a pressure sensor array, and a conductivity sensor array; the moisture-absorbing and heat-releasing expansion layer includes adsorbed salt and an expansion material, wherein the adsorbed salt reacts with water to produce an exothermic reaction and transfers heat to the temperature sensor array; the expansion material absorbs water and expands to generate mechanical stress, which directly acts on the pressure sensor array; the conductivity sensor array changes its resistance and generates a conductivity characteristic signal after encountering water.

8. A pipeline leakage detection system based on multimodal composite sensing according to claim 7, characterized in that, The adsorbed salt is calcium chloride, the expanding material is cross-linked sodium polyacrylate, and the mass ratio of calcium chloride to cross-linked sodium polyacrylate is 4:

1.

9. A pipeline leakage detection system based on multimodal composite sensing according to claim 7, characterized in that, It also includes an alarm and linkage module connected to the control processing module, which is used to trigger multi-channel alarms.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements a pipeline leakage detection method based on multimodal composite sensing as described in any one of claims 1-6.

Images