Intelligent fire hydrant for water supply pipeline leakage detection and algorithm

Abstract

The invention relates to the technical field of pipeline water leakage detection and fire fighting equipment, and discloses a fire hydrant device and method for intelligent detection of pipeline water leakage. In order to solve the problems of limited operation time, high experience dependency, environmental noise interference and the like in the traditional manual detection technology, the invention provides an integrated solution that an embedded control unit, a broadband underwater sound sensing module and a solar energy supply system are arranged on a fire hydrant base body, and pipeline sound wave signals are collected in real time. After the preprocessing of pre-emphasis, 43-dimensional feature vectors are extracted from three aspects of time domain, frequency domain and auditory perception, and leakage probability and confidence coefficient prediction is realized in combination with a pre-trained SVM classification model, so that the detection efficiency and the confidence coefficient are improved; and the cloud platform receives the data based on the MQTT protocol and judges whether an early warning mechanism is triggered or not. The device supports long-term unattended operation, significantly reduces the manual inspection cost through a confidence threshold and a trigger alarm mechanism, is suitable for intelligent monitoring and early leakage early warning of a municipal water supply network, and is suitable for popularization and application.

Description

[Technical Field]

[0001] This invention patent relates to the field of pipeline leak detection and fire protection equipment technology. [Background Technology]

[0002] Fire hydrants, as key nodes in urban water supply networks, play a vital role in urban fire prevention. The annual water loss due to pipe leaks poses a significant threat to public safety. Prolonged pipe leaks can damage underground pipe network structures, leading to widespread flooding or soil erosion, and posing safety risks to the stability of roads and buildings. Water waste not only affects urban water supply, especially for fire prevention, but also has negative environmental impacts. Large leaks can seep into the soil, causing groundwater pollution and potentially triggering public health emergencies. As a crucial infrastructure component of the urban water supply network, the operational status of fire hydrants directly impacts the city's fire safety capabilities. Current traditional detection methods have significant technical limitations. The existing manual auscultation method is limited by human hearing thresholds and experience; traditional testing requires nighttime operation, increasing labor costs and posing safety hazards. To address the problem of pipeline leak detection, Chinese patent application (publication number CN118856248A) discloses a pipeline leak detection and location method based on multi-scale Mann-Whitney test. This method involves performing a multi-scale Mann-Whitney test on acoustic signals and calculating the Pipeline Leakage Status Index (PLSI). The PLSI is used to calculate the effect size (ES) and detect pipeline leaks online. If a leak is detected, the acoustic emission signal at the leak point is identified, high-energy acoustic emission events are filtered out and stored in an event database, and the pipeline length is divided into segments, with the KS density calculated for each segment. Areas with high KS density are identified as leak locations.

[0003] This solution still has some shortcomings in practical applications. Due to the complex distribution of urban water supply networks and the generally long distances between pipelines, collecting acoustic emission signals from the pipelines to be assessed is cumbersome, involves a large amount of manual work, and results in low overall detection efficiency. Acoustic emission technology, as an active non-destructive testing method, exhibits high-frequency characteristics and short-duration bursts. Affected by material attenuation coefficients and propagation paths, signal attenuation is significant, limiting the effective monitoring range to a few meters. This technology is mainly used for the early identification of microscopic defects within materials, and its detection window precedes the formation of macroscopic leaks. When monitoring characteristic acoustic emission signals, the structure may still be in a subcritical damage state, not yet reaching the threshold for fluid leakage. This temporal sensitivity advantage combined with spatial monitoring range limitations determines its specific applicability in industrial scenarios. To address these limitations, this study proposes an intelligent pipeline monitoring solution: by deploying intelligent sensing terminals at the topological nodes of the water supply network and constructing a cloud-based data machine learning model training framework, online identification and early warning of pipeline leak characteristics can be achieved. Empirical data verification shows that the system can achieve a detection efficiency of ≥90% for leak event identification in typical municipal pipeline network environments. Compared with the traditional single acoustic emission detection mode, it can effectively improve the detection accuracy of pipeline leaks, thereby greatly reducing the workload and effectively realizing the quantitative assessment of pipeline health status and optimization of resource allocation. [Summary of the Invention]

[0004] This invention provides an intelligent fire hydrant device for detecting pipeline leaks. The technical solution includes: a fire hydrant base, an embedded control unit, a hydroacoustic sensing module, and an autonomous power supply system. Furthermore, the embedded control unit uses a rectangular sealed chamber fixed to the front face of the fire hydrant base. This control panel is connected to a hydrophone located inside the pipeline via a cable. Internally, it integrates a high-density lithium-ion battery pack (operating voltage 12VDC, capacity 12000mAh) and a double-layer PCB control motherboard. The control motherboard integrates a digital signal processor, a control unit (STM32F105), a wireless Cat.1-WAN communication module (ML302), and a data storage unit (real-time transmission SRAM area). The acquisition chip is AD7768-1, with a 24-bit acquisition bit depth, a maximum sampling rate of 256kHz, and a range of ±4.192V. The positioning system uses the ATGM336H, a high-performance GPS module that provides high-precision, high-reliability location information services with low power consumption. It adopts SiRFstarIII technology and supports multiple satellite positioning systems.

[0005] Furthermore, the self-powered system is composed of monocrystalline silicon photovoltaic panels, which are connected to an intelligent charging controller on top of the fire hydrant base via cables to optimize photoelectric conversion efficiency.

[0006] Furthermore, the underwater acoustic sensing module employs a wideband piezoelectric hydrophone, which converts acoustic signals to electrical signals via a waterproof coaxial cable and a built-in programmable preamplifier (fixed gain 20dB). The front panel of the device is equipped with switches and status indicator lights, forming a complete electromechanical integrated monitoring terminal. This structural design meets fire hydrant standards and, through modular assembly, achieves spatial decoupling of the sensor, power supply, and communication subsystems, effectively improving the device's reliability and maintenance convenience in complex pipe network environments.

[0007] Furthermore, the control circuit board in the system integrates a wireless communication module for data interaction with an external cloud database device. The cloud database receives acoustic signals collected by the hydrophone and uploaded via the control circuit board. Features are automatically extracted from the signal and input into a classification model for prediction. The cloud system is designed based on the ONENET IoT open platform, with an overall architecture including a device access layer, a data processing layer, a service support layer, and an application and visualization layer. The system uses a Time Series Database (TSDB) to store time-series data collected by various devices, providing efficient data storage and retrieval capabilities. For communication protocols, the MQTT protocol is used to meet the needs of efficient and stable communication between various devices. Furthermore, the cloud database performs 43-dimensional feature extraction in the time domain, frequency domain, and auditory perception after pre-emphasis on the existing acoustic data. Furthermore, the time-domain features refer to the mean, variance, root mean square (RMS), kurtosis, skewness, maxima, and interquartile range (IQR) of the signal window obtained in the time domain, totaling 7 dimensions. Furthermore, the frequency domain features refer to the 13 dimensions of the frequency domain centroid, spectral entropy, frequency domain variance, mean of power spectral density, skewness of the spectrum, kurtosis of the spectrum, flatness of the spectrum, and energy ratio of the frequency band below 1 kHz, as well as the mean and standard deviation of the Mel frequency cepstral coefficients (MFCC), for a total of 33 dimensions, obtained through Fourier transform.

[0008] Furthermore, the auditory perception features refer to loudness, sharpness, and pitch in the perception parameters, totaling three dimensions.

[0009] The aforementioned feature parameters constitute a 43-dimensional feature vector for each frame of signal, comprehensively covering time-domain statistical characteristics, frequency-domain spectral features, and auditory perception indicators. Based on a standardized experimental design process, ductile iron pipes and galvanized steel pipes were used as research objects, focusing on two typical pipe diameters, DN100 and DN200, to construct experimental systems. Acoustic wave data was repeatedly collected and manually labeled according to whether leakage occurred, constructing a labeled acoustic wave dataset. Supervised learning was then conducted using cross-validation (k=5) and Bayesian optimization strategies, employing an ensemble model of Support Vector Machine (SVM: linear kernel, penalty coefficient C=1.0). The model was optimized using key performance indicators such as F1-score and AUC (Area Under the ROC Curve) and deployed to a cloud system to achieve efficient and reliable automated acoustic wave leakage detection.

[0010] During the testing phase, the communication module automatically collects 30-second acoustic signals from a hydrophone late at night and uploads them to the cloud. The cloud platform receives the acoustic data, extracts feature vectors, and calls a pre-trained model to output a prediction of whether the pipe is leaking, the probability of leakage, and the confidence level. When the model determines that a leak exists and the prediction confidence level is high, the system will automatically trigger an early warning mechanism and push an early warning notification. For samples with repeated alarms, the system sets cases where alarms are triggered for three consecutive days with a gradually increasing confidence level as key investigation targets, which are then prioritized for manual on-site verification. The verification results are transmitted back to the cloud platform via the wireless communication module and automatically incorporated into the acoustic feature database. These samples are used as labeled samples to further improve the database and enhance the predictive capabilities of the trained model, achieving a closed-loop iterative upgrade of intelligent diagnostic capabilities. This invention patent achieves online monitoring and data feedback through the wireless communication module of the control component, effectively improving the accuracy and efficiency of leak detection.

[0011] Table 1. Description of 43-dimensional features

[0012]

[0013] [Attached Image Description]

[0014] Figure 1 Front view structural diagram

[0015] Figure 2 Side view structural diagram

[0016] Figure 3 Top view of the structure

[0017] Figure 4 :flow chart

Detailed Implementation Methods

[0018] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0019] like Figures 1 to 3 As shown, an intelligent fire hydrant for pipeline leak detection includes a hydrant body 1, a control component 2, a hydrophone 4, and a solar panel 6. The control component 2, in a rectangular structure, is fixed to the front of the hydrant body 1. A wiring conduit 3 is located in the middle of the front of the hydrant body 1, housing a power cord electrically connected to a lithium battery 21. The solar panel 6 is located directly above the hydrant body 1 and is connected to and powered by the lithium battery 21 in the control component 2 via the wiring conduit 3. A wireless communication module 23 is located on one side of the control circuit board 22, enabling communication with external devices. The control component 2, with its lithium battery 21, control circuit board 22, and wireless communication module 23, is positioned at the front of the hydrant body 1. The lithium battery 21 provides power storage, and the wireless communication module 23 enables wireless communication with external devices. Figures 1 to 3 As shown, hydrophones 4 are installed on the inner bottom of the fire hydrant body 1. The hydrophones 4 are connected to the control circuit board 22. A main switch 5 is provided on the front side of the control circuit board 22 and is connected to the circuit board 22. A main water pipe 11 is provided in the middle of the inner side of the fire hydrant body 1. Water inlet pipes 12 are provided on both sides of the fire hydrant body 1. The water inlet pipes 12 are connected to the main water pipe 11 and have a cover 13 at their ends.

[0020] After preprocessing the acoustic signal of the pipe under test by the hydrophone with pre-emphasis, 43-dimensional feature extraction is performed in the time domain, frequency domain and auditory perception.

[0021] In the time domain, the mean, variance, root mean square (RMS), kurtosis, skewness, maximum value, and interquartile range (IQR) of the window are obtained; the specific formulas are as follows:

[0022] • Mean: x i Let N be the i-th sample, and N be the total number of samples.

[0023] ·variance:

[0024] • Root mean square:

[0025] ·Cumber:

[0026] • Skewness:

[0027] • Maximum value: Max = max(x1, x2, ... x N )

[0028] • Interquartile Range (IQR): IQR = Q3 - Q1, where Q1 is the 25th percentile and Q3 is the 75th percentile.

[0029] Fourier transform yields the spectral centroid, spectral entropy, frequency domain variance, mean of power spectral density, skewness, kurtosis, flatness, and energy ratio of the frequency band below 1 kHz, as well as the mean and standard deviation of the 13 dimensions of the Mel-frequency cepstral coefficients (MFCC). The specific formulas are as follows:

[0030] X(f k ) is the amplitude (or power spectral density) of the k-th frequency component.

[0031] f k It corresponds to the frequency.

[0032] N is the total number of frequency points.

[0033]

[0034] • Spectral centroid:

[0035] • Spectral entropy: ε is a constant to prevent the minimum value of log(0).

[0036] • Frequency domain variance: μ f Spectral centroid

[0037] • Mean power spectral density:

[0038] • Spectral kurtosis:

[0039] • Spectral skewness:

[0040] • Frequency domain flatness:

[0041] • Energy ratio of frequencies below 1kHz:

[0042] MFCC: Let MFCC i It is the vector of the i-th MFCC coefficients.

[0043] • The mean of the i-th MFCC:

[0044] • The standard deviation of the i-th MFCC: Loudness, sharpness, and pitch are parameters of auditory perception. The specific formulas are as follows:

[0045] Loudness: N is the total loudness, in sone; N'(z) is the sympathetic loudness distribution along the cochlear position z; loudness contribution / sharpness of each Bark band: Unit: acum, g(z) is the weighting function, the higher the frequency band, the greater the weight, where z: Critical band frequency (0-24 bar)

[0046] • Pitch: S i : The i-th significant tonal peak; N i Noise near the peak frequency band

[0047] The aforementioned feature parameters constitute a 43-dimensional feature vector for each frame of signal, comprehensively covering time-domain statistical characteristics, frequency-domain spectral characteristics, and auditory perception indicators. Based on whether leakage occurs, manual annotation is performed to construct a labeled sound wave dataset.

[0048] Based on a standardized experimental design process, this study focuses on ductile iron and galvanized steel pipes, specifically DN100 and DN200 pipes. An experimental system with controllable leakage points was constructed, and leakage acoustic data was repeatedly collected and manually labeled according to whether leakage occurred. A labeled acoustic dataset was then built, and supervised learning was implemented. Cross-validation (k=5) and Bayesian optimization strategies were employed, along with an ensemble model using Support Vector Machines (SVM: linear kernel, penalty coefficient C=1.0). The model was optimized using key performance indicators such as F1-score and AUC (Area Under the ROC Curve) and deployed to a cloud system to achieve efficient and reliable automated acoustic leakage detection. The basic principle of SVM is as follows:

[0049] Kernel function support vector machine

[0050] Support Vector Machine (SVM) is a discriminative model used for binary classification. It achieves linear separability in a high-dimensional space through a kernel function.

[0051] Original objective function: s.ty i (w T x i +b)≥1

[0052] Dual problem after adding kernel function

[0053]

[0054] Where α i : Lagrange multipliers, C: penalty parameter, K(x) i ,x jKernel function: Linear kernel function: K(x,x')=x T x'

[0055] Final classification function:

[0056] During the testing phase, the communication module automatically collects short-duration acoustic signal segments from the hydrophone at night and uploads them to the cloud. The cloud platform receives the acoustic data, extracts feature vectors, and calls a pre-trained model to output a judgment on whether the pipe is leaking, the probability of leakage, and the confidence level. When the model determines that a leak exists and the prediction confidence level is high, the system will automatically trigger an early warning mechanism and push an early warning notification. For cases with repeated alarms over several consecutive days and a gradually increasing confidence level, the system will mark them as key investigation targets, which will be prioritized for manual on-site verification. The verification results will be transmitted back to the cloud platform via the wireless communication module and automatically incorporated into the acoustic feature database as labeled samples to further improve the database and enhance predictive capabilities, achieving a closed-loop iterative intelligent diagnostic capability.

[0057] As can be seen from the above description of this invention patent, compared with the prior art, the advantages of this invention patent are as follows:

[0058] It adopts an integrated intelligent monitoring hydrophone device, which is embedded inside the pipeline. It can operate stably for a long time without human supervision, significantly reducing the frequency of manual inspections and maintenance costs. It is suitable for large-scale deployment and remote scenarios.

[0059] The system is based on a high-sensitivity hydrophone and acoustic sensor, which can directly collect sound wave signals in the pipeline and directly listen to the leakage sound without the need for external sensor arrays or auxiliary devices, effectively avoiding environmental noise interference and improving detection accuracy.

[0060] The system automatically triggers an alarm mechanism based on the confidence level of the classification model's output results. Samples with consistently high confidence levels over multiple consecutive days, and whose confidence levels increase daily, are designated as key targets for review. Furthermore, the cloud-based monitoring platform...

[0061] System parameters support dynamic adjustment.

[0062] The above are merely specific embodiments of the present invention, but the concept involved in the present invention is not limited thereto.

[0063] This is only limited to the above; any non-substantial improvements made to this invention patent using this concept are considered...

[0064] This should be considered an act that infringes on the scope of protection of this invention patent.

Claims

1. A fire hydrant device for intelligent detection of leaks in water supply pipelines, characterized in that, include: Fire hydrant body; An embedded control unit fixed to the front end of a fire hydrant, the control unit includes a lithium-ion battery, a circuit control board and a wireless communication module; A hydroacoustic sensing module connected to the control unit, the hydroacoustic sensing module including a broadband hydrophone and a preamplifier; and an autonomous power supply system connected to the control unit, the power supply system including a monocrystalline silicon photovoltaic panel.

2. The fire hydrant device according to claim 1, characterized in that, The bottom of the control unit is connected to a hydrophone placed at the bottom of the fire hydrant pipe via a cable.

3. The fire hydrant device according to claim 1, characterized in that, The underwater acoustic sensing module is connected to the preamplifier via a waterproof coaxial cable.

4. The fire hydrant device according to claim 1, characterized in that, The control unit interacts with a cloud database via a wireless communication module. The cloud database is built on the ONENET IoT platform, uses the MQTT protocol for communication, and uses a time-series database (TSDB) to store device data.

5. The fire hydrant device according to any one of claims 1 to 3, characterized in that, The device achieves pipeline leakage detection through the following algorithm steps: pre-emphasis and preprocessing of the acoustic signal; extraction of a 43-dimensional feature vector, including 7-dimensional time-domain features, 33-dimensional frequency-domain features, and 3-dimensional auditory perception features; inputting the features into a support vector machine (SVM) model for leakage status discrimination.

6. The time-domain feature according to claim 4, characterized in that: The time-domain features include mean, variance, root mean square (RMS), kurtosis, skewness, maxima, and interquartile range (IQR).

7. The frequency domain feature according to claim 4, characterized in that: The frequency domain features include spectral centroid, spectral entropy, frequency domain variance, mean power spectral density, spectral skewness, spectral kurtosis, frequency domain flatness, band power ratio, and the mean and standard deviation of the first 13 dimensions of MFCC.

8. The auditory perception feature according to claim 4, characterized in that: The auditory perception features mentioned include loudness, sharpness, and pitch.

9. The method for detecting pipeline leakage in a fire hydrant device according to any one of claims 1-4, characterized in that: In the communication module, short-duration acoustic signal fragments automatically collected by the hydrophone at night are uploaded to the cloud. The cloud platform receives the acoustic data, extracts feature vectors, and calls a pre-trained model to output a judgment on whether the pipe is leaking, the probability of leakage, and the confidence level. When the model determines that a leak exists and the prediction confidence level is high, the system will automatically trigger an early warning mechanism and push an early warning notification. For situations where alarms are repeatedly triggered over several consecutive days and the confidence level shows a gradual upward trend, the system will mark it as a key investigation target, which will be prioritized for manual on-site verification.

Images