Water pipeline leakage detection method based on fiber grating sensor

Abstract

The invention relates to a water pipe leakage detection method based on a distributed fiber bragg grating sensing network, and belongs to the technical field of cross application of optical fiber sensors and artificial intelligence. In the process of the method, firstly, a three-dimensional fluid dynamic simulation model of the water conveying pipeline system is established, and training data are generated; secondly, the fiber grating sensors are laid along the pipeline in a distributed mode to form a real-time monitoring network which is used for collecting negative pressure wave signals generated by leakage; then, aiming at a signal noise problem, innovatively adopting a variation mode decomposition method (VMD) combining a genetic algorithm (GA) and a correlation coefficient (CC) to carry out intelligent noise reduction and reconstruction on the signal; then, dimension reduction is carried out on the reconstructed signal by using a compressed sensing (CS) technology so as to reduce the calculation complexity; then, the signals after dimension reduction are converted into logarithmic spectrograms, the logarithmic spectrograms are input into a deep residual neural network (ResNet) for feature extraction, and then the signals are accurately classified into noise, normal, slight leakage, moderate leakage and serious leakage through a two-dimensional convolutional neural network (2D CNN); and finally, according to a classification result and the position of the sensor, calculating and positioning a leakage point by using a negative pressure wave method. According to the method, by fusing advanced signal processing and deep learning technologies, the accuracy, robustness and intelligent level of leakage detection are remarkably improved, and effective monitoring of the health state of the water conveying pipeline is achieved.

Description

Technical Field

[0001] This invention relates to a novel sensing method and device for monitoring the structural health of water pipelines, specifically to a method for detecting and locating leaks in water pipelines based on fiber optic grating sensors, belonging to the field of cross-application of fiber optic sensing and artificial intelligence technologies. Background Technology

[0002] Water pipelines are a crucial component of urban infrastructure and industrial production. Pipeline leaks not only waste precious water resources and increase operating costs, but can also trigger secondary disasters such as ground subsidence and environmental pollution, threatening public safety. Therefore, developing an efficient, accurate, and real-time pipeline leak detection and location technology has significant economic and social value. Traditional pipeline leak detection methods mainly include flow balance methods, pressure gradient methods, and acoustic detection methods. These methods generally suffer from low sensitivity, poor location accuracy, and susceptibility to environmental noise interference. In recent years, detection technology based on negative pressure wave methods has received widespread attention due to its advantages such as fast response speed and high location accuracy. However, negative pressure wave signals are susceptible to noise interference and attenuation during long-distance transmission, leading to blurred signal characteristics and affecting detection accuracy. To address these issues, this invention proposes a detection method that integrates advanced signal processing algorithms and deep learning models, aiming to improve the robustness and location accuracy of leak detection. Summary of the Invention

[0003] The water pipeline leakage detection system based on fiber optic grating sensors proposed in this invention is a novel online detection device with the following unique advantages: (1) Fiber optic grating sensors are small in size, lightweight, easy to install, and possess intrinsic safety, electromagnetic interference resistance, and corrosion resistance, making them very suitable for long-term deployment in complex industrial environments; (2) The variational mode decomposition method (GA-CC-VMD) combining genetic algorithms and correlation coefficients can adaptively remove environmental noise and signal aliasing, effectively reconstructing high-quality leakage characteristic signals; (3) The introduction of compressed sensing (CS) technology to reduce the dimensionality of the signal significantly reduces the computational complexity of subsequent deep learning models and improves processing efficiency; (4) The deep learning framework composed of deep residual networks (ResNet) and two-dimensional convolutional neural networks (2D CNN) can automatically learn and accurately classify pipeline signals in different states from logarithmic spectra, greatly improving the accuracy and automation level of leakage identification. This invention includes the following steps:

[0004] S1: Establish a three-dimensional fluid dynamics model of the water pipeline system using ANSYS Fluent or similar CFD software. Simulate different flow rate and pressure conditions, and pre-determine leak holes of different locations and sizes to generate a simulation dataset including normal operation and various leakage states for subsequent algorithm training and preliminary verification. The physical basis for simulating water flow and leakage processes within the pipeline using CFD software is the incompressible Navier-Stokes equations:

[0005]

[0006] Where: u is the velocity vector (m / s), p is the pressure (Pa), and ρ is the fluid density (kg / m³). 3 ), where μ is the dynamic viscosity (Pa·s) and f is the volume force (such as gravity). A simulated leak is generated by setting a local pressure drop or mass source term on the pipe wall to simulate the leak, thereby generating a simulated signal for the propagation of a negative pressure wave.

[0007] S2: Based on the pipeline's length, direction, and key nodes, determine the number and optimal spatial layout of fiber optic grating sensors. Install the sensors flush against or embedded in the pipeline's outer wall, connecting them to a data acquisition card via optical fiber. Data is then transmitted to a host computer for centralized processing. Simultaneously, deploy auxiliary sensors such as pressure gauges, accelerometers, and water meters to comprehensively monitor the pipeline's operating status.

[0008] S3: Start the water pump to operate the pipeline under set conditions. A negative pressure wave is triggered by artificially creating a leak (e.g., opening a valve). A distributed fiber Bragg grating sensor network captures the negative pressure wave signal propagating from the leak point to both ends in real time, and the data is acquired at high speed by a data acquisition card. B To sense the strain changes caused by negative pressure waves, the basic relationship is as follows:

[0009]

[0010] Where: λ B p is the initial Bragg wavelength. e α is the effective photoelastic coefficient (approximately 0.22 for silica fiber), ε is the axial strain (caused by the negative pressure wave), and α is the effective photoelastic coefficient. Λ α is the coefficient of thermal expansion. n Δλ is the thermo-optic coefficient, and ΔT is the temperature change. Under isothermal or temperature-compensated conditions, Δλ B It primarily reflects strain ε, and thus characterizes the intensity of the negative pressure wave.

[0011] S4: For the acquired noisy negative pressure wave signal, perform the following processing: 1) Initialize VMD parameters (number of modes K, penalty factor α). 2) Using the correlation coefficient (CC) between the reconstructed signal and the original signal as the fitness function, use a genetic algorithm (GA) to globally search for the optimal values ​​of K and α. 3) Use the optimized parameters to perform VMD decomposition on the signal to obtain K intrinsic mode functions (IMFs). 4) Calculate the CC value of each IMF component relative to the original signal. 5) Select IMF components with CC values ​​higher than a set threshold for signal reconstruction to obtain the denoised effective leakage signal. The VMD algorithm achieves signal decomposition by solving the following variational problem:

[0012]

[0013] Where: u k (t) represents the k-th intrinsic mode function (IMF), w k Let K be the corresponding center frequency, δ(t) be the Dirac delta function, and * denote convolution. This model achieves adaptive signal decomposition by constraining the minimization of bandwidth for each mode. A genetic algorithm is used to optimize K and the penalty parameter α (regularization term weights).

[0014] S5: For the reconstructed effective signal, dimensionality reduction is performed using compressed sensing theory: 1) Design a random Gaussian matrix as the measurement matrix Φ. 2) Obtain the low-dimensional measurement vector y through linear projection y = Φx, where x is the original signal vector. 3) Use the low-dimensional vector y as the input to the subsequent deep learning model to reduce the computational burden. Assume that the reconstructed signal x has sparsity under a certain orthogonal basis Ψ (such as a wavelet basis), i.e.:

[0015] x = Ψθ, ||θ|| 0 ≤ S << N

[0016] Where θ is the sparse coefficient vector and S is the number of non-zero elements. Combining the measurement y = Φx = ΦΨθ, x can be accurately reconstructed by minimizing l1 (when the RIP condition is satisfied).

[0017] S6: Convert the dimensionality-reduced signal data into a logarithmic spectrogram to more clearly show the energy distribution and subtle differences in the signal. Construct a deep residual network (ResNet) model, using the logarithmic spectrogram as input. Leveraging ResNet's powerful feature extraction capabilities, automatically learn and identify leakage features in the spectrogram, outputting preliminary leakage probabilities or feature vectors. ResNet employs residual learning units, and its output is:

[0018] y=F(x,{W i})+x

[0019] Where F(·) represents stacked convolutional layers (such as two 3×3 convolutions), and x represents the input feature map. This structure allows gradients to be directly backpropagated through the identity mapping, effectively alleviating the gradient vanishing problem in deep network training.

[0020] S7: Construct a 2D convolutional neural network (2D CNN) with an output layer containing 5 neurons. Input the features or logarithmic spectra extracted by ResNet directly into the 2D CNN. Train the CNN model to accurately classify the input signal into five categories: "noise signal," "normal signal," "slight leakage signal," "moderate leakage signal," and "severe leakage signal." The last layer of the 2D CNN outputs the probabilities of these five categories using the Softmax function.

[0021]

[0022] Where z = [z1, z2, z3, z4, z5] T The outputs of the fully connected layer correspond to five categories: "noise, normal, slight, moderate, and severe".

[0023] S8: When the system detects a leak, it selects the two adjacent sensors where the leak signal first arrives. The time difference Δt between the leak signal and these two sensors is precisely calculated. Based on the known propagation speed v of the negative pressure wave in the pipe fluid, the distance from the leak point to these two sensors is calculated using the formula L=(v*Δt) / 2, thus achieving precise location.

[0024] S9: The detection model proposed in this invention is comprehensively evaluated from four dimensions: accuracy, precision, recall, and F1 score. It is compared with traditional methods such as thresholding and support vector machines (SVM) to demonstrate the superiority of this method in terms of detection accuracy and efficiency. The formulas for calculating the four performance evaluation metrics—accuracy, precision, recall, and F1 score—are as follows:

[0025]

[0026] Where TP, TN, FP, and FN represent true positives, true negatives, false positives, and false negatives, respectively. Attached Figure Description

[0027] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0028] Figure 1 This is an overall flowchart of the water pipeline leakage detection method based on fiber optic grating sensor described in this invention;

[0029] Figure 2 This is a flowchart of the signal preprocessing process based on the GA-CC-VMD algorithm described in this invention.

[0030] Figure 3 This is a flowchart of the deep learning signal recognition and classification process based on ResNet and 2D CNN described in this invention. Detailed Implementation

[0031] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] This invention discloses a method for detecting leaks in water pipelines based on fiber Bragg grating sensors. First, the pipeline system is simulated using CFD software to generate training data. Next, multiple fiber Bragg grating sensors are distributed across the pipeline surface to construct a sensor network. When a leak occurs, the sensors capture a negative pressure wave signal. This signal is first intelligently denoised and reconstructed using the GA-CC-VMD algorithm to eliminate environmental noise and signal distortion. Subsequently, compressed sensing technology is used to reduce the dimensionality of the reconstructed signal, simplifying the data. The dimensionality-reduced data is converted into a logarithmic spectrum and input into a ResNet model for deep feature extraction and preliminary identification. Finally, a 2D CNN model performs a five-class classification task on the signal to accurately determine the leak status, and the leak point is located by combining sensor location information.

[0033] like Figure 1 As shown, this invention mainly comprises four core parts: system modeling and data acquisition, signal preprocessing, deep learning identification and classification, and leakage location and assessment.

[0034] like Figure 2 As shown, the core of the signal preprocessing section is the GA-CC-VMD algorithm. This algorithm adaptively optimizes VMD parameters through a genetic algorithm and uses correlation coefficients to screen effective mode components, thereby effectively filtering out noise while preserving leakage characteristics to the greatest extent, providing a high-quality input signal for subsequent processing.

[0035] like Figure 3 As shown, the deep learning part adopts a two-stage architecture of "ResNet + 2D CNN". ResNet is responsible for extracting deep, abstract leakage features from the logarithmic spectrogram, and its residual structure effectively solves the training problem of deep networks. 2DCNN acts as a classifier, using convolutional kernels to perform spatial analysis on the feature map, and finally outputs accurate classification results.

[0036] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for detecting leaks in water pipelines based on fiber optic grating sensors, characterized in that, The method includes the following steps: A three-dimensional simulation model of the water pipeline system was established to simulate the fluid dynamics behavior under different working conditions, providing data support for sensor layout and algorithm verification. Multiple fiber Bragg grating (FBG) sensors are distributed along the water pipeline and connected to a data acquisition card and a host computer to form a real-time monitoring network. The system also includes a water tank, a water pump, a pressure gauge, an accelerometer, and a water meter for controlling and monitoring the pipeline's operating status. By artificially creating or simulating a leak event, the negative pressure wave signal generated by the leak in the pipeline is collected by a fiber optic grating sensor. The acquired raw negative pressure wave signal was preprocessed, and the variational mode decomposition (VMD) method, which combines genetic algorithm (GA) and correlation coefficient (CC), was used for noise reduction and signal reconstruction in order to extract effective leakage characteristic signals. The reconstructed effective signal is subjected to dimensionality reduction using compressed sensing (CS) technology to reduce data dimensionality and simplify the computational complexity of subsequent models. The dimensionality-reduced signal is converted into a logarithmic spectrum and input into a deep residual neural network (ResNet) model to efficiently and accurately identify and capture leaked signals using deep learning strategies. A two-dimensional convolutional neural network (2D CNN) model is used to classify the signals output by ResNet into five categories: noise signals, normal signals, slightly leaky signals, moderately leaky signals, and severely leaky signals. Based on the classification results and combined with the sensor location information, the leak point is calculated and located using the propagation speed and arrival time difference of the negative pressure wave. The model performance is evaluated using four metrics: accuracy, precision, recall, and F1 score. The results are compared with traditional methods to verify the superiority of the proposed method.

2. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, When building a three-dimensional simulation model, computational fluid dynamics (CFD) software is used to simulate the water flow state inside the pipe, and leak holes of different locations and sizes are preset to generate a simulation dataset for training and validating the algorithm.

3. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, The variational mode decomposition method (GA-CC-VMD) combining genetic algorithms and correlation coefficients specifically includes: 1) Adaptively optimize the number of mode decompositions K and the penalty factor α in the VMD algorithm using a genetic algorithm; 2) Calculate the correlation coefficient (CC) between each modal component (IMF) and the original signal; 3) Select modal components with high correlation coefficients for reconstruction and discard components with low correlation coefficients to achieve effective noise reduction.

4. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, The compressed sensing (CS) dimensionality reduction process uses a random Gaussian matrix as the measurement matrix to linearly project the reconstructed time-domain signal to obtain low-dimensional measurement values.

5. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, The deep residual neural network (ResNet) is used to automatically extract deep leakage features from the logarithmic spectrum and overcome the gradient vanishing problem in deep network training, thereby improving the identification efficiency and accuracy of leakage signals.

6. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, The output layer of the two-dimensional convolutional neural network (2D CNN) model contains five neurons, which correspond to five classification labels: noise signal, normal signal, slight leakage signal, moderate leakage signal, and severe leakage signal.

7. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, The formula for locating the leak point is: Where L is the distance from the leak point to the two adjacent sensors, v is the propagation speed of the negative pressure wave in the fluid in the pipe, and Δt is the time difference between the arrival of the leak signal to the two adjacent sensors.

8. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, In the compressive sensing dimensionality reduction process, the original signal vector x ∈ R N is linearly projected through a measurement matrix Φ ∈ R M×N (where M << N) to obtain a low-dimensional measurement vector y ∈ R M , and its mathematical expression is: y = Φx Wherein, Φ is a random Gaussian matrix that satisfies the Finite Isometry Property (RIP), which is used to significantly reduce the data dimensionality while preserving the main information of the signal.

9. The method for detecting leaks in water pipelines based on fiber optic grating sensors as described in claim 1, characterized in that, The logarithmic spectrum is obtained by performing a short-time Fourier transform (STFT) on the dimension-reduced time-domain signal s(t) and taking the logarithmic amplitude. Its expression is as follows: Where w(·) is the window function, f is the frequency, t is the time, and ε is a small positive number (such as 10) to prevent logarithmic divergence. -6 The resulting logarithmic spectrogram is used as the input image for the deep learning model.

10. The GA-CC-VMD signal noise reduction method as described in claim 3, characterized in that, Genetic algorithms for reconstructing signals The correlation coefficient ρ with the original noisy signal x(t) is used as the fitness function, and it is defined as follows: Where cov(·,·) represents the covariance. and σ x ρ represents the standard deviation of the reconstructed signal and the original signal, respectively; the genetic algorithm optimizes the number of modes K and the penalty factor α of VMD by maximizing ρ.

11. The GA-CC-VMD signal noise reduction method as described in claim 3, characterized in that, The correlation coefficient ρ between the i-th intrinsic mode function (IMF(_i)) and the original noisy signal x(t) i Defined as: in, and Let be the IMF(_i) and the mean of the original signal, respectively, and N be the signal length; only |ρ is retained. i The modal components of |>τ are reconstructed, where τ is a preset threshold (e.g., 0.3).

12. The leak point location method as described in claim 7, characterized in that, The propagation speed v of a negative pressure wave in a pipe fluid is determined by the following formula: Where K is the bulk modulus of the fluid, ρ is the fluid density, D is the inner diameter of the pipe, h is the pipe wall thickness, and E is the Young's modulus of the pipe material. This formula takes into account the coupling effect between the fluid and the pipe wall and is suitable for estimating the propagation velocity of water in metal or plastic water pipes.

Images