Method for trace detection of leaks in gas path systems by photoacoustic spectroscopy and related devices

Abstract

The present application relates to the technical field of gas leakage detection, in particular to a photoacoustic spectroscopy trace detection method for gas path system leakage and related device. The photoacoustic spectroscopy trace detection method for gas path system leakage comprises the following steps: emitting specific wavelength laser with double-frequency intensity modulation to the suspected leakage area of the gas path system, the specific wavelength matches two adjacent absorption spectral lines of the target leakage gas, and the frequency difference of the double-frequency intensity modulation corresponds to the frequency interval of the two adjacent absorption spectral lines, so that the leakage gas molecules absorb laser energy of two frequencies at the same time to enhance the amplitude of the periodic photoacoustic signal. Through the cooperation of multiple technical links, the present application effectively solves the problems of weak trace signal, difficult noise suppression and low positioning accuracy in the existing method, and realizes the unification of high-sensitivity trace detection and centimeter-level accurate positioning for gas path system leakage.

Description

Photoacoustic spectroscopy trace detection method and related devices for gas system leaks Technical Field

[0001] This invention relates to the field of gas leak detection technology, specifically to a photoacoustic spectroscopy trace detection method and related apparatus for gas system leaks. Background Technology

[0002] In industrial production, leaks in gas systems not only waste rare gas resources but can also trigger major safety accidents such as explosions and poisoning. Therefore, high-sensitivity, high-precision, and high-reliability detection of gas system leaks is crucial for ensuring industrial safety. Photoacoustic spectroscopy (PAS), based on the principle of selective absorption of specific wavelengths of laser light by gas molecules, detects leaks by detecting the periodic photoacoustic signals generated after the gas absorbs laser energy. It boasts advantages such as high sensitivity, good selectivity, and non-contact measurement, and has become one of the mainstream technologies for trace gas leak detection. However, existing photoacoustic spectroscopy methods for gas system leak detection still suffer from insufficient synergistic optimization, making it difficult to meet the stringent requirements of industrial scenarios for trace leaks.

[0003] On the one hand, in order to match the single absorption spectrum of the target gas, existing methods mostly use single-frequency intensity modulated lasers, which causes the leaked gas molecules to absorb only the laser energy of a single frequency, resulting in a weak photoacoustic signal amplitude. Especially in trace leakage scenarios (such as below ppm level), the signal is easily submerged by environmental noise, making it difficult to achieve effective detection.

[0004] On the other hand, when the gas system is running, environmental vibration noise such as the vibration of the fixed support and the disturbance of the pipeline fluid will be transmitted to the photoacoustic signal acquisition device through the gas system structure. Existing adaptive filtering algorithms mostly use the vibration sensor signal with a fixed step size or a single direction as the reference input, which cannot dynamically adjust the filter coefficient to adapt to the time-varying characteristics of the noise, resulting in incomplete separation of leakage characteristic signal and vibration noise, and reduced signal-to-noise ratio.

[0005] Furthermore, spatial positioning relies heavily on traditional TDOA (Time Difference of Arrival) algorithms, which only utilize the time difference of arrival of photoacoustic signals without incorporating the gas system topology model (such as branch connection nodes and pipe diameter direction) and flow distribution data. This makes it difficult to achieve centimeter-level three-dimensional positioning of leak points in complex gas path branches (such as multi-branch, curved pipes), resulting in insufficient positioning accuracy. These problems are intertwined, making it difficult for existing methods to simultaneously achieve high-sensitivity signal detection and high-precision spatial positioning in trace leak detection, thus failing to meet the trace leak detection requirements of industrial gas system systems. Summary of the Invention

[0006] This disclosure presents a photoacoustic spectroscopy trace detection method and related device for gas system leaks, aiming to overcome at least one defect in the prior art.

[0007] To achieve the above objectives, the technical solution disclosed in this invention is as follows:

[0008] According to one aspect of this disclosure, a photoacoustic spectroscopy method for detecting trace amounts of leakage in a gas path system is provided, comprising the steps of:

[0009] A specific wavelength laser with dual-frequency intensity modulation is emitted towards the suspected leak area of ​​the gas path system. The specific wavelength matches two adjacent absorption spectral lines of the target leaking gas. The frequency difference of the dual-frequency intensity modulation corresponds to the frequency interval of the two adjacent absorption spectral lines, so that the leaking gas molecules absorb the laser energy of two frequencies at the same time, thereby enhancing the amplitude of the periodic photoacoustic signal.

[0010] The photoacoustic signal is acquired using a gas path branch array fiber optic probe. The gas path branch array fiber optic probe includes multiple fiber optic sensors arranged in multiple dimensions along the axial, radial and circumferential directions of the gas path branch. Each fiber optic sensor adopts a reflective optical path, and the optical path is twice the length of the fiber optic cable, which is used to acquire the spatial distribution information of the photoacoustic signal.

[0011] Based on the spatial distribution information, combined with the gas path system topology model and flow distribution data, the three-dimensional coordinates of the leak point are calculated by a spatial positioning algorithm that integrates the TDOA algorithm and the machine learning model, for centimeter-level positioning.

[0012] The photoacoustic signal is subjected to adaptive filtering. The fused signal of the multi-directional vibration sensor on the fixed support of the gas path system is used as the reference input. The filter coefficient is dynamically adjusted by a variable step size recursive least squares algorithm to separate the environmental vibration noise and leakage characteristic signal.

[0013] Based on the separated leakage characteristic signals, multi-dimensional features such as amplitude, phase difference, and frequency shift are extracted. Combined with a pre-calibrated photoacoustic response model and a machine learning model trained on historical leakage data, the existence and rate of leakage are determined.

[0014] Furthermore, the dual-frequency intensity modulation is implemented as follows: a tunable laser outputs the laser of a specific wavelength, and a dual-frequency signal generator generates two modulation frequencies f1 and f2, wherein the difference between f1 and f2 is equal to the frequency interval between two adjacent absorption spectral lines of the target leaking gas; the dual-frequency signal is input to an intensity modulator to modulate the intensity of the laser so that the intensity of the laser changes periodically with f1 and f2.

[0015] Furthermore, calculating the three-dimensional coordinates of the leak point includes the following steps:

[0016] The photoacoustic signals acquired by each fiber optic sensor are preprocessed: a bandpass filter is used to filter high-frequency noise, and then the signal envelope is extracted by Hilbert transform. The peak position of the envelope is used as the arrival time of the photoacoustic signal of the sensor.

[0017] Calculate the time difference of arrival (TDOA) between any two fiber optic sensors: For N fiber optic sensors, calculate C(N,2) time differences to form a time difference matrix;

[0018] Import the gas path system topology model: Convert the 3D CAD model of the gas path into a mathematical model, including the coordinates of the connection nodes of each branch, the pipe diameter and the direction, as geometric constraints for positioning;

[0019] Incorporate flow distribution data: Obtain real-time flow from the flow sensors of the gas path system for each branch, and correct the possible location range of the leak point based on the relationship between flow and leakage rate;

[0020] Training the machine learning model: Train a support vector machine model using experimental data from known leak points. The inputs are the time difference matrix, topology model parameters, and traffic data. The output is the coordinates of the leak points. Optimize the kernel function and penalty parameters of the model.

[0021] Calculate the coordinates of the leak point: Input the time difference matrix, the parameters of the topology model, and the flow distribution data into the trained support vector machine model, and output the three-dimensional coordinates of the leak point.

[0022] Furthermore, the specific steps of the variable step-size recursive least squares algorithm are as follows:

[0023] Initialization: Set the filter order M, initialize the filter coefficient vector w(0)=0, and initialize the covariance matrix. Where δ is a small positive number, It is the identity matrix;

[0024] Acquiring reference signal and original photoacoustic signal: The reference signal x(n) is a weighted fusion of vibration sensor signals from three directions, and the original photoacoustic signal d(n) is the signal acquired by the fiber optic probe, where n represents the number of samplings;

[0025] Calculate the filter output: , where y(n) is the filter output and w(n) is the filter coefficient vector. Let T be the reference signal vector, and let T denote the transpose of the vector.

[0026] Calculate the error signal: , where e(n) is the estimated value of the leakage characteristic signal after separation;

[0027] Adjusting the step size: Using an adaptive step size formula , where μ0 is the initial step size and α is the adjustment factor;

[0028] Update filter coefficients: ;

[0029] Update the covariance matrix: ;

[0030] The process of acquiring the reference signal and the original photoacoustic signal is repeated until the variance of the error signal e(n) is less than a preset threshold, thus completing the separation of environmental vibration and noise.

[0031] Furthermore, the calibration process of the photoacoustic response model includes:

[0032] Selecting standard leakage sources: Select standard leakage sources with known leakage rates, and select at least 3 samples for each leakage rate;

[0033] Set up a calibration test platform: Connect a standard leak source to the test section of the gas path system to simulate a real leak scenario, arrange gas path branch array fiber optic probes and vibration sensors, and set laser emission parameters;

[0034] Acquire calibration data: For each standard leakage source, acquire at least 10 sets of photoacoustic signals, with each set acquired for 60 seconds, and extract the amplitude, phase difference, and frequency shift of each set of signals;

[0035] Fitting the mapping relationship: For each leakage rate, calculate the average value of multi-dimensional features of multiple samples and multiple sets of photoacoustic signals to form a feature vector; use a multiple linear regression model to fit the mapping relationship between the leakage rate and the feature vector to obtain the photoacoustic response model. Where R is the leakage rate, A is the amplitude, Δφ is the phase difference, Δf is the frequency shift, and a, b, c, and d are regression coefficients;

[0036] Verify model accuracy: Test the model using a standard leak source that has not been calibrated, and calculate the relative error between the model's predicted values ​​and the actual values.

[0037] Furthermore, the deployment method of the fiber optic sensor includes: using phase-sensitive optical time-domain reflectometry to achieve high-resolution spatial sensing of photoacoustic signals, selecting polarization-maintaining fiber as the transmission medium to suppress polarization fading, and forming a reflective optical path by setting a gold-plated reflector at the end of the fiber to enhance the interaction between light and leaked gas.

[0038] Furthermore, the adaptive parameter adjustment method for the specific wavelength laser includes: automatically adjusting the output wavelength of the tunable laser based on two adjacent absorption spectral lines of the target leaking gas, with a range of 1.5μm-1.6μm and a wavelength tuning accuracy ≤0.01nm; adjusting the output power of the laser based on the pipe length of the gas path system to compensate for light attenuation during transmission; and adjusting the linewidth of the laser based on the temperature change of the gas path system to maintain the monochromaticity of the laser and improve gas absorption efficiency.

[0039] Furthermore, the dynamic construction method of the gas path system topology model includes: constructing an initial three-dimensional topology model of the gas path system using three-dimensional drawing software, including branch connection nodes, pipe diameter and direction, exporting it to STL format and converting it into a mathematical model, including the axis equation and node coordinates of each branch; collecting the flow data of each branch in real time through the flow sensor of the gas path system, and correcting the branch direction and node position in the topology model in combination with the flow distribution law, so as to dynamically update the gas path system topology model.

[0040] Furthermore, the multi-dimensional suppression method for environmental vibration noise includes: setting piezoelectric vibration sensors in multiple directions on the fixed support of the air circuit system, using a magnetic installation method, and optimizing the sensor installation points based on the vibration mode test results of the support; weighting and fusing the vibration sensor signals from multiple directions to form a reference input signal, so as to improve the suppression effect of adaptive filtering on environmental vibration noise.

[0041] Furthermore, the method for utilizing the historical leakage data includes: storing the historical leakage data in a structured relational database; extracting the correlation between leakage rate and environmental parameters, as well as the mapping pattern between photoacoustic signal characteristics and leakage point location, using data mining algorithms; and retraining the machine learning model with the mined patterns to optimize the model's leakage judgment accuracy.

[0042] Furthermore, it also includes visualizing the leak detection results, with the following steps:

[0043] The three-dimensional coordinates of the leak point are superimposed onto the gas path system topology model, and the location of the leak point is marked with different colors.

[0044] The real-time value of the leakage rate and the historical change curve are displayed on the same interface. The horizontal axis of the curve is time and the vertical axis is leakage rate.

[0045] The visual interface is developed using WebGL technology and supports scaling, rotation, and translation operations to adapt to display devices of different sizes.

[0046] Furthermore, this also includes periodically calibrating the gas path branch array fiber optic probe, following these steps:

[0047] Regular calibration is performed by connecting a standard leak source to the calibration section of the gas path system to simulate leak scenarios.

[0048] Acquire photoacoustic signals from the calibration section and extract amplitude, phase difference, and frequency shift;

[0049] The extracted features are compared with a pre-calibrated photoacoustic response model, and the deviation of the feature values ​​is calculated.

[0050] If the deviation exceeds the preset threshold, the sensitivity coefficient of the fiber optic probe is adjusted, and the calibration result is stored in the relational database.

[0051] According to another aspect of this disclosure, a photoacoustic spectral trace detection system for gas system leaks is provided, for implementing the photoacoustic spectral trace detection method for gas system leaks as described above, comprising:

[0052] A laser emitting module is used to emit a specific wavelength laser with dual-frequency intensity modulation towards a suspected leak area. The laser emitting module includes a tunable laser, a dual-frequency signal generator, and an intensity modulator.

[0053] The gas path branch array fiber optic probe module includes multiple fiber optic sensors arranged in multiple dimensions along the gas path branches, used to collect photoacoustic signals and acquire spatial distribution information.

[0054] The spatial positioning module is used to receive spatial distribution information of photoacoustic signals, gas path topology model and flow distribution data, and calculate the three-dimensional coordinates of the leak point through a spatial positioning algorithm that integrates the TDOA algorithm and machine learning model.

[0055] An adaptive filtering module is used to receive the fused signals from photoacoustic signals and multi-directional vibration sensors, and uses a variable step size recursive least squares algorithm to separate environmental vibration noise and leakage characteristic signals.

[0056] The leakage detection module is used to receive the separated feature signals, extract multi-dimensional features, and combine the photoacoustic response model and machine learning model to determine whether a leak exists and the leakage rate.

[0057] The visualization module is used to visualize the location and rate of leaks superimposed on the gas path topology model.

[0058] According to another aspect of this disclosure, a photoacoustic spectral trace detection device for gas system leaks is provided, comprising the photoacoustic spectral trace detection system for gas system leaks as described above, and further comprising:

[0059] A housing for accommodating the detection system, the housing being made of anti-static PC material;

[0060] A gas connection interface is used to connect the device to a suspected leak area of ​​the gas system, and the gas connection interface adopts a quick-connect structure;

[0061] The display interface is used to visually display the leak detection results;

[0062] The power supply module is used to provide a stable power supply for the detection system. The power supply module adopts a redundant design to ensure continuous operation.

[0063] The communication module is used for data transmission with the gas system monitoring platform.

[0064] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the photoacoustic spectral trace detection method for gas system leaks as described above.

[0065] According to another aspect of this disclosure, an electronic product is provided that integrates or assembles the computer-readable storage medium as described above, and further includes:

[0066] Processor for executing computer programs in the storage medium;

[0067] The communication module is used for data transmission with the gas system monitoring platform;

[0068] Input interface, used to receive parameter settings from the gas circuit system;

[0069] The output interface is used to connect to an external display device to output the leak detection results.

[0070] The beneficial effects of this invention are:

[0071] This invention effectively solves the problems of weak trace signals, difficulty in noise suppression, and low positioning accuracy in existing methods through the synergy of multiple technical aspects, and achieves the unification of highly sensitive trace detection and centimeter-level precise positioning of gas system leaks.

[0072] Specifically, by using dual-frequency intensity-modulated laser technology, two adjacent absorption spectral lines of the target leaking gas are matched, allowing the leaking gas molecules to absorb laser energy of two frequencies simultaneously. Compared with single-frequency modulation, this significantly enhances the amplitude of the periodic photoacoustic signal, effectively improving the detection sensitivity of trace leaks (such as ppb level) and solving the problem of insufficient amplitude of existing single-frequency modulated photoacoustic signals.

[0073] Furthermore, a gas path branch array fiber optic probe is adopted, with multiple fiber optic sensors arranged in multiple dimensions along the axial, radial, and circumferential directions of the gas path branch. Combined with a reflective optical path design, the optical path is twice the length of the fiber. This not only obtains the spatial distribution information of the photoacoustic signal but also enhances the interaction between the light and the leaked gas, further improving the signal strength. At the same time, a spatial positioning algorithm that integrates the TDOA algorithm and machine learning model is used, combined with the gas path system topology model (such as branch node coordinates and pipe diameter direction) and flow distribution data.

[0074] Furthermore, by mining the correlation between time difference, topology parameters and flow data through machine learning models, centimeter-level three-dimensional localization of leak points was achieved, solving the problem of insufficient localization accuracy of existing single TDOA algorithms in complex gas path branches. In addition, using the fused signal from multi-directional vibration sensors as a reference input, a variable step-size recursive least squares algorithm was used to dynamically adjust the filter coefficients. The step size was adaptively adjusted according to the time-varying characteristics of the error signal, effectively separating environmental vibration noise from leak characteristic signals, solving the problem of poor noise suppression effect of existing adaptive filtering algorithms.

[0075] This invention achieves end-to-end optimization of trace signal enhancement, adaptive noise separation, and precise spatial positioning. It not only improves the detection sensitivity of trace leaks but also enables centimeter-level positioning of leak points, meeting the high sensitivity and high precision detection requirements of industrial gas circuit systems for trace leaks and providing reliable technical support for the safe operation of industrial production. Attached Figure Description

[0076] Figure 1 is a flowchart of a photoacoustic spectral trace detection method for gas system leakage according to an embodiment of the present invention;

[0077] Figure 2 is a schematic diagram of the dual-frequency modulated laser and gas absorption enhancement principle of the present invention;

[0078] Figure 3 is a schematic diagram of the adaptive filtering noise separation effect of the present invention;

[0079] Figure 4 is a schematic diagram of the TDOA and machine learning fusion positioning algorithm of the present invention, wherein (a) is a fiber optic sensor signal arrival time extraction diagram, (b) is a time difference matrix (TDOA) heat map, (c) is a schematic diagram of the gas path topology model, and (d) is a schematic diagram of the machine learning model output.

[0080] Figure 5 is a schematic diagram of the calibration curve of the photoacoustic response model of the present invention;

[0081] Figure 6 is a visual schematic diagram of the leakage location effect of the present invention. Detailed Implementation

[0082] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0083] The present invention provides the following preferred embodiments:

[0084] Example 1: To address the issues of insufficient amplitude of single-frequency modulated photoacoustic signals, limited spatial positioning accuracy in complex gas path branches, and poor environmental noise suppression in existing photoacoustic spectral detection of gas path system leaks, this example refines the signal enhancement, spatial sensing, and positioning stages of the trace detection method, achieving a balance between high-sensitivity trace detection and centimeter-level precise positioning. As shown in Figure 1, the flow of the photoacoustic spectral trace detection method for gas path system leaks is as follows:

[0085] S100: Emits a specific wavelength laser with dual-frequency intensity modulation to the suspected leak area of ​​the gas path system. The specific wavelength matches two adjacent absorption spectral lines of the target leaking gas. The frequency difference of the dual-frequency intensity modulation corresponds to the frequency interval of the two adjacent absorption spectral lines, so that the leaking gas molecules absorb the laser energy of two frequencies at the same time, thereby enhancing the amplitude of the periodic photoacoustic signal.

[0086] S200: The air path branch array fiber optic probe is used to collect photoacoustic signals. The air path branch array fiber optic probe includes multiple fiber optic sensors arranged in multiple dimensions along the axial, radial and circumferential directions of the air path branch. Each fiber optic sensor adopts a reflective optical path, and the optical path is twice the length of the fiber optic cable, which is used to obtain the spatial distribution information of the photoacoustic signal.

[0087] S300: Based on spatial distribution information, combined with the gas path system topology model and flow distribution data, a spatial positioning algorithm that integrates the TDOA algorithm and machine learning model is used to calculate the three-dimensional coordinates of the leak point for centimeter-level positioning.

[0088] S400: Adaptive filtering of photoacoustic signals, using the fused signal from multi-directional vibration sensors on the fixed support of the gas path system as a reference input, and dynamically adjusting the filter coefficients using a variable step size recursive least squares algorithm to separate environmental vibration noise from leakage characteristic signals.

[0089] S500: Based on the separated leakage characteristic signal, it extracts multi-dimensional features such as amplitude, phase difference and frequency offset, and combines them with a pre-calibrated photoacoustic response model and a machine learning model trained on historical leakage data to determine whether a leakage exists and the leakage rate.

[0090] Specifically, a specific wavelength laser, modulated by dual frequencies, is emitted towards the suspected leak area of ​​the gas path system. This specific wavelength precisely matches two adjacent absorption lines of the target leaking gas. For example, when the target gas is methane, its adjacent absorption lines are located around 1.65 μm with a frequency interval of 1.2 GHz. The difference between the two frequencies f1 and f2 of the dual-frequency intensity modulation is then set to 1.2 GHz. As shown in Figure 2, the frequency difference of the dual-frequency intensity-modulated laser strictly corresponds to the frequency interval of the adjacent absorption lines of the target gas, causing the leaking gas molecules to resonate and absorb simultaneously at both frequencies. Compared to single-frequency modulation, the amplitude of the photoacoustic signal is significantly enhanced, effectively improving the detection sensitivity of trace leaks.

[0091] Furthermore, a gas path branch array fiber optic probe is used to collect photoacoustic signals. The probe is arranged in a multi-dimensional manner along the axial, radial, and circumferential directions of the gas path branch: one fiber optic sensor group is set every 5 cm in the axial direction, and three fiber optic sensors are evenly arranged in the radial direction along the circumference of the pipe in each group. One fiber optic sensor is arranged on each side of the axial direction of each radial sensor, forming a dense three-dimensional sensing network to ensure the acquisition of comprehensive spatial distribution information of photoacoustic signals. Each fiber optic sensor adopts a reflective optical path design. One end of the fiber is connected to the laser emission module and the signal acquisition module, and the other end is fixed to the outer wall of the gas path pipe. A high-reflectivity metal mirror is set at the end. The laser is transmitted through the fiber to the mirror and then returns along the original fiber. The optical path is twice the length of the fiber, which enhances the interaction between the light and the leaked gas and further improves the signal strength. As shown in Figure 4(a), the photoacoustic signals collected by the multi-dimensionally arranged fiber optic sensors provide basic data for subsequent time of arrival extraction and spatial positioning.

[0092] Furthermore, based on the spatial distribution information of the photoacoustic signal obtained by the fiber optic probe, combined with the gas path system topology model and flow distribution data, the three-dimensional coordinates of the leak point are calculated by a spatial positioning algorithm that integrates the TDOA algorithm and the machine learning model. First, the photoacoustic signals collected by each fiber optic sensor are preprocessed. A bandpass filter is used to filter high-frequency noise, and then the signal envelope is extracted through Hilbert transform. The peak position of the envelope is used as the arrival time of the photoacoustic signal of the sensor. Then, the arrival time difference between any two sensors is calculated to form a time difference matrix, as shown in Figure 4(b). The gas path system topology model is imported, and the three-dimensional CAD model of the gas path is converted into a mathematical model containing the coordinates of branch connection nodes, pipe diameter and direction, as a geometric constraint for positioning, as shown in Figure 4(c). Flow distribution data is incorporated. The real-time flow of each branch is obtained from the flow sensor of the gas path system, and the possible location range of the leak point is corrected according to the relationship between flow and leakage rate. When training the machine learning model, the support vector machine model is trained using experimental data of known leak points. The input is the time difference matrix, topology model parameters and flow data, and the output is the coordinates of the leak point. The kernel function and penalty parameters of the model are optimized. Finally, the time difference matrix, topology model parameters and flow data are input into the trained model, and the three-dimensional coordinates of the leak point are output, as shown in Figure 4(d) and Figure 6, to achieve centimeter-level positioning.

[0093] Furthermore, the photoacoustic signal undergoes adaptive filtering. Using the fused signal from the multi-directional vibration sensors on the fixed support of the gas path system as the reference input, a variable-step-size recursive least squares algorithm is employed to dynamically adjust the filter coefficients. The multi-directional vibration sensors are magnetically mounted at the vibration mode test optimization point on the support. The collected signals are weighted and fused to form the reference input. The variable-step-size recursive least squares algorithm dynamically updates the filter coefficients by adjusting the step-size factor based on the time-varying characteristics of the error signal, effectively separating the leakage characteristic signal from the environmental vibration noise. As shown in Figure 3, the signal-to-noise ratio of the separated leakage characteristic signal is significantly improved, laying the foundation for subsequent feature extraction.

[0094] Furthermore, based on the separated leakage characteristic signals, multi-dimensional features such as amplitude, phase difference, and frequency shift are extracted. These features are then combined with a pre-calibrated photoacoustic response model and a machine learning model trained on historical leakage data to determine the presence and rate of leakage. The pre-calibrated photoacoustic response model was established through experiments with standard leakage sources, as shown in Figure 5. The model fits the mapping relationship between leakage rate and multi-dimensional features of the photoacoustic signal. Historical leakage data is structured and stored in a relational database. Data mining is used to extract the correlation between leakage rate and environmental parameters, which is then used to optimize the machine learning model. The machine learning model employs a random forest algorithm. The inputs are photoacoustic signal features and environmental parameters, and the outputs are leakage state and leakage rate. After training and optimization, the model's generalization ability and judgment accuracy are improved.

[0095] This embodiment enhances the amplitude of the photoacoustic signal through dual-frequency intensity modulation, acquires spatial distribution information by arranging fiber optic probes in multiple dimensions, achieves centimeter-level positioning by integrating TDOA and machine learning, suppresses environmental noise by combining variable step size adaptive filtering, and finally realizes leakage judgment through multi-dimensional features and machine learning models. The entire link optimizes the trace leakage detection performance and meets the needs of industrial gas circuit systems for high sensitivity and high precision detection.

[0096] Example 2: To address the issue that the amplitude of a single-frequency intensity-modulated photoacoustic signal is insufficient to meet the requirements for trace leakage detection, this example further refines the specific implementation method of dual-frequency intensity modulation. Through the coordinated operation of a tunable laser, a dual-frequency signal generator, and an intensity modulator, the laser intensity presents a superposition of two frequency components, ensuring a significant improvement in the resonant absorption efficiency of the leaked gas molecules to the laser.

[0097] First, a tunable laser is used to output laser light at a specific wavelength. This wavelength must be precisely matched to two adjacent absorption lines of the target leaking gas. Taking methane as an example, its adjacent absorption lines are located around 1.65 μm. The tunable laser uses temperature or current tuning to stably lock the output wavelength within this spectral range, ensuring that the laser light can be effectively absorbed by the leaking methane molecules. The output of the tunable laser is transmitted to the optical input of the intensity modulator via optical fiber, providing a stable light source for subsequent modulation.

[0098] Furthermore, a dual-frequency signal generator is used to generate two modulation frequencies, f1 and f2, where the difference between f1 and f2 is equal to the frequency interval between two adjacent absorption lines of the target leaking gas, such as 1.2 GHz for methane. The dual-frequency signal generator needs to have high stability, typically employing phase-locked loop (PLL) technology to ensure the difference between the two frequencies remains constant, preventing gas molecules from being unable to simultaneously absorb both frequencies of laser light due to frequency drift. The output of the dual-frequency signal generator is connected to the control terminal of the intensity modulator via a coaxial cable, loading the electrical signal onto the modulator's electro-optic or acousto-optic crystal.

[0099] Furthermore, the intensity modulator converts the dual-frequency electrical signal into an intensity modulation signal, causing the input laser intensity to vary periodically with f1 and f2. Specifically, the intensity modulator changes the refractive index or diffraction efficiency of the crystal to make the laser intensity amplitude vary according to the waveform of the dual-frequency signal. The final output laser intensity is a superposition of two sinusoidal components, and its intensity expression can be described as a linear combination of the two frequency components, where the frequency of each component corresponds to the output frequency of the dual-frequency signal generator.

[0100] As shown in Figure 2, the frequency difference of the dual-frequency modulated laser strictly corresponds to the frequency interval of adjacent absorption lines in the target gas, ensuring that the leaking gas molecules simultaneously resonate and absorb the laser light at both frequencies. When the laser irradiates the leaking gas region, the gas molecules satisfy the Boltzmann transition conditions at both frequencies f1 and f2, and the total absorbed light energy is more than twice that of single-frequency modulation, thus significantly enhancing the amplitude of the photoacoustic signal. It is important to understand that the frequency difference of the dual-frequency signal must be perfectly consistent with the frequency interval of adjacent absorption lines; otherwise, the gas molecules cannot absorb the laser light at both frequencies simultaneously, and the signal enhancement effect cannot be achieved.

[0101] In this embodiment, the wavelength stability of the tunable laser, the frequency accuracy of the dual-frequency signal generator, and the modulation depth of the intensity modulator all need to meet stringent requirements: the wavelength drift of the tunable laser must be controlled within 0.01 nm to ensure that the laser always covers the target absorption spectrum; the frequency error of the dual-frequency signal generator must be less than 10 MHz to avoid a decrease in resonant absorption efficiency due to frequency difference deviation; and the modulation depth of the intensity modulator must be greater than 90% to ensure that the light intensity change is sufficiently significant. Through the coordinated operation of the above components, the output characteristics of the dual-frequency intensity-modulated laser can be precisely controlled, providing a sufficient photoacoustic signal intensity basis for trace leakage detection.

[0102] Example 3: To address the limited spatial positioning accuracy of leak points in complex gas path branches, this example further refines the three-dimensional coordinate calculation method of leak points based on the fusion of TDOA and machine learning. Through signal preprocessing, time difference matrix construction, gas path topology constraints, flow data fusion, and machine learning model training, centimeter-level positioning in complex gas path environments is achieved.

[0103] Specifically, the photoacoustic signals acquired by the fiber optic sensors are first preprocessed. Due to the presence of high-frequency electromagnetic noise and mechanical vibration noise in the airflow environment, a bandpass filter is used to filter out noise components higher than the photoacoustic signal frequency. The center frequency of the bandpass filter is set to the characteristic frequency of the photoacoustic signal, such as 1kHz, and the bandwidth is ±10% of the characteristic frequency, to retain the effective components of the photoacoustic signal. Subsequently, the analytical envelope of the signal is extracted using Hilbert transform. The peak position of the envelope curve corresponds to the arrival time of the photoacoustic signal, as shown in Figure 4(a). Due to differences in propagation paths, the peak times of the envelopes of photoacoustic signals acquired by different fiber optic sensors vary significantly. By extracting the peak position, the arrival time of the signal from each sensor can be accurately obtained. The core objective of the preprocessing step is to eliminate noise interference, ensure the accuracy of arrival time extraction, and provide a reliable data foundation for subsequent positioning calculations.

[0104] Furthermore, the time difference of arrival between any two fiber optic sensors is calculated to form a time difference matrix (TDOA matrix). For N fiber optic sensors, C(N,2) time differences need to be calculated, each time difference corresponding to the time interval between the two sensors receiving the same photoacoustic signal. As shown in Figure 4(b), the TDOA matrix is ​​presented in the form of a heatmap, with the color intensity representing the magnitude of the time difference. This matrix can intuitively reflect the propagation law of the photoacoustic signal in the air path. The time difference matrix is ​​the foundation of positioning calculation, and its accuracy directly affects the accuracy of subsequent positioning results.

[0105] Furthermore, a gas path system topology model is imported as a geometric constraint. The 3D CAD model of the gas path is converted into a mathematical model containing the coordinates of branch connection nodes, pipe diameter, and direction. For example, for a three-way branch gas path, the mathematical model needs to include parameters such as the coordinates of the connection points between the main branch and the branches, the length and angle of each branch, and the pipe diameter, as shown in Figure 4(c). The role of the topology model is to limit the possible location of the leak point to the geometric range of the gas path pipeline, avoiding deviations from the gas path in the location results. At the same time, the flow distribution data of the gas path system is incorporated. Real-time flow rates of each branch are obtained from flow sensors, and the possible location range of the leak point is corrected according to the relationship between flow rate and leakage rate. For example, in branches with higher flow rates, the leak point is more likely to be located in the downstream area because the downstream gas velocity is higher, and the leakage signal propagates faster.

[0106] Furthermore, a support vector machine model was trained to achieve a nonlinear mapping of leak point coordinates. Experimental data from known leak points were used for training: standard leak sources were set up at different locations in the gas path system, and the signal arrival time of each fiber optic sensor, the corresponding topology model parameters, and flow data were recorded, with the true coordinates of the leak points labeled. The time difference matrix was converted into vector form, and the topology model parameters, such as branch length, angle, and flow data, were converted into numerical features as input to the model; the three-dimensional coordinates of the leak points were used as the model output. During training, the kernel function and penalty parameters of the model were optimized through cross-validation. A radial basis function was chosen as the kernel function to handle the nonlinearity of the gas path topology; the penalty parameters were determined through grid search to ensure a performance balance between the training and validation sets.

[0107] Furthermore, the preprocessed time difference matrix, topology model parameters, and flow distribution data are input into the trained support vector machine model. The model outputs the three-dimensional coordinates of the leak point, as shown in Figure 4(d). The output results are presented in the form of a bar chart, corresponding to the coordinate values ​​in the X, Y, and Z directions, with the true values ​​marked by dashed lines to intuitively reflect the positioning accuracy. It is important to understand that the role of the machine learning model is to handle the nonlinear problems of multivariate interactions in complex gas paths, combining topological constraints and flow data to make the positioning results more consistent with the actual structure and operating state of the gas path.

[0108] This embodiment ensures the accuracy of arrival time through signal preprocessing, constructs the positioning basis through the TDOA matrix, provides constraints through topology model and flow data, and realizes nonlinear mapping through support vector machine model. All steps work together to finally achieve centimeter-level leak point location in complex gas path branches.

[0109] Example 4: To address the interference of multi-directional vibration noise in the gas path environment on photoacoustic signals, this example further refines the specific implementation of the variable step size recursive least squares (RLS) algorithm. By adaptively adjusting the step size and recursively updating the filter parameters, accurate separation of environmental vibration noise is achieved, ensuring the effective extraction of leakage characteristic signals.

[0110] First, the algorithm is initialized. Based on the frequency characteristics of vibration and noise in the gas path environment (such as narrowband interference and broadband noise including 200Hz and 500Hz), the filter order M is selected as 32 to cover the time delay range of the noise signal; the filter coefficient vector w(0) is initialized as a zero vector to ensure that the initial state is unbiased; the covariance matrix P(0) is set to δ -1 δ is taken as 0.01 (a small positive number) to avoid singularity of the initial matrix and to ensure the numerical stability of the recursive process.

[0111] Further, a reference signal and the original photoacoustic signal are acquired. The reference signal x(n) is generated by weighted fusion of the output signals from the three vibration sensors in each direction. The weights are determined based on the intensity of vibration noise in each direction; for example, a higher weight is assigned when axial vibration noise is stronger. The purpose is to fully capture the multi-directional characteristics of environmental vibration noise, making the reference signal highly correlated with the vibration noise components in the photoacoustic signal. The original photoacoustic signal d(n) is directly acquired by the fiber optic probe and contains the superposition of leakage characteristic signals and environmental vibration noise. Here, n represents the number of samples, and the sampling rate needs to match the characteristic frequency of the photoacoustic signal, such as 10kHz, to ensure coverage of the 1kHz photoacoustic signal frequency.

[0112] Furthermore, the filter output and error signal are calculated. The filter output y(n) is the inner product of the filter coefficient vector w(n) and the reference signal vector x(n), that is:

[0113] The reference signal vector x(n) consists of the current and the reference signals sampled in the previous M-1 times, i.e.:

[0114] This delay structure enables the filter to effectively capture the time correlation of noise. The error signal e(n) is the original photoacoustic signal d(n) minus the filter output y(n), that is:

[0115] Its physical meaning is the estimated value of the leakage characteristic signal after separation. When the filter coefficients accurately match the noise characteristics, y(n) will approximate the vibration noise component in the photoacoustic signal, and e(n) will approximate the pure leakage characteristic signal.

[0116] To balance the algorithm's convergence speed and steady-state error, an adaptive step size formula is used to adjust the step size μ(n). The step size formula is:

[0117] The initial step size μ0 is set to 0.1 based on the noise intensity, and the adjustment factor α is set to 0.1 to control the step size change rate. The logic of this formula is as follows: when the amplitude of the error signal e(n) is large, it indicates that the current filter coefficients have a low matching degree with the noise characteristics, so the step size μ(n) is increased to speed up the update speed of the filter coefficients and shorten the convergence time; when the amplitude of the error signal e(n) is small, it indicates that the filter coefficients are close to the optimal value, so the step size μ(n) is decreased to avoid steady-state fluctuations caused by excessive coefficient updates and to ensure the stability of the separated signal.

[0118] Based on the adjusted step size, update the filter coefficients and covariance matrix. The filter coefficient update formula is:

[0119] Where P(n) is the covariance matrix, this formula recursively transforms the correlation between the error signal and the reference signal into coefficient adjustment amounts, allowing the filter coefficients to gradually approach the optimal solution. The covariance matrix update formula is:

[0120] Its function is to avoid directly calculating the matrix inverse and maintain the positive definite property of P(n) through recursion, thus ensuring the real-time performance and stability of the algorithm.

[0121] Furthermore, the process of acquiring signals, calculating output and error, adjusting step size, updating coefficients and covariance matrix needs to be repeated until the variance of the error signal e(n) is less than a preset threshold, such as 0.001. At this point, the filter coefficients have fully matched the characteristics of environmental vibration and noise, and the vibration and noise components in the error signal e(n) are effectively separated. The remaining part is the clear leakage characteristic signal, as shown in Figure 3. After the observed signal, which includes leakage signal and vibration noise, is filtered by variable step size RLS, the waveform of the output signal is highly consistent with the original leakage signal. Noise components such as 200Hz and 500Hz narrowband interference and broadband noise are significantly suppressed, verifying the noise separation effect of the algorithm.

[0122] In this embodiment, the key to the variable step size RLS algorithm lies in achieving a balance between convergence speed and steady-state accuracy through adaptive step size adjustment, while the recursive update of the covariance matrix ensures the real-time performance of the algorithm, enabling it to meet the real-time requirements of gas path system leak detection. The multi-directional fusion design of the reference signal further improves the accuracy of noise separation, ensuring that leak characteristic signals can be accurately extracted even in complex vibration environments.

[0123] Example 5: To solve the problem of accurate mapping between the photoacoustic response model and the actual leakage rate, this example further refines the calibration process of the photoacoustic response model. By collecting multi-sample data from standard leakage sources and fitting multi-dimensional features, it is ensured that the model can accurately reflect the relationship between the leakage rate and the photoacoustic signal characteristics.

[0124] When selecting standard leak sources, choose standard sources with known leak rates for the target gas (e.g., methane), and select at least three samples for each leak rate to reduce the impact of individual sample differences on the calibration results. When setting up the calibration experimental platform, connect the standard leak sources to the test section of the gas path system to simulate a real leak scenario. The test section uses the same materials and branch structure as the actual gas path, and arranges gas path branch array fiber optic probes and vibration sensors to reproduce the superposition of photoacoustic signals and vibration noise in the actual environment. The laser emission parameters are set to a dual-frequency modulation mode consistent with the actual detection, as shown in Figure 2, with dual absorption lines near 1.65 μm, ensuring consistency between the calibration data and the actual detection data.

[0125] Furthermore, during calibration data acquisition, at least 10 sets of photoacoustic signals were collected for each standard leak source, with each set acquired for 60 seconds to obtain stable signal characteristics. The amplitude, phase difference, and frequency shift of each signal set were extracted: the amplitude reflects the intensity of the photoacoustic signal and is directly related to the gas concentration change caused by the leak rate; the phase difference is caused by the difference in light propagation time in the leak area and is related to the location and rate of the leak point; the frequency shift is caused by the slight change in laser frequency due to the change in the refractive index of the leaking gas, indirectly reflecting the leak rate. For each leak rate, the average value of the multi-dimensional features of multiple samples and multiple sets of photoacoustic signals was calculated to form a feature vector, thereby reducing the influence of random noise on the features.

[0126] Furthermore, when fitting the mapping relationship, a multiple linear regression model is used to fit the mapping relationship between the leakage rate and the feature vector. The multiple linear model is chosen because amplitude, phase difference, and frequency shift are all linearly correlated with the leakage rate, as shown in Figure 5. The amplitude increases linearly with the leakage rate, while the phase difference decreases linearly. The combination of multi-dimensional features can more comprehensively reflect the information of the leakage rate. During the fitting process, the regression coefficients are solved using the least squares method to ensure that the model optimally fits the training data.

[0127] Furthermore, to verify the model's accuracy, the model was tested using a standard leak source that was not calibrated, and the relative error between the model's predicted values ​​and the actual values ​​was calculated. Uncalibrated samples effectively test the model's generalization ability, ensuring accurate prediction of unknown leak rates in practical applications. As shown in Figure 5, the linear fitting curve of the amplitude feature and the leak rate shows that the deviation between the predicted and actual values ​​is small, verifying the model's fitting effect; the fitting curve of the phase difference feature also exhibits a stable linear relationship, further ensuring the model's multi-dimensional feature fusion effect.

[0128] The advantages of this embodiment are that, through multi-sample calibration and multi-dimensional feature fitting of standard leakage sources, the photoacoustic response model can accurately reflect the relationship between leakage rate and photoacoustic signal characteristics, providing a reliable model basis for subsequent quantitative detection of leakage rate; the calibration experimental platform simulating real leakage scenarios ensures the adaptability of the model in actual gas circuit systems; and the absence of participation in the verification process of calibration samples further guarantees the generalization ability of the model.

[0129] Example 6: To address the spatial resolution and signal stability issues of fiber optic sensors in pneumatic systems, this example further optimizes the deployment method of fiber optic sensors. By combining phase-sensitive optical time-domain reflectometry with polarization-maintaining fibers and high-reflectivity mirrors, high-resolution spatial sensing and stable signal transmission are achieved.

[0130] Specifically, phase-sensitive optical time-domain reflectometry (OTDR) is employed to achieve high-resolution spatial sensing of photoacoustic signals. This technique injects narrow-pulse laser light into an optical fiber and detects the phase change of the backscattered light to achieve spatial localization of the photoacoustic signal. Compared with traditional OTDR, phase-sensitive OTDR offers higher spatial resolution and can accurately identify small-interval leaks in the branch structure of the gas path system, as shown in the triflingual model in Figure (c). This provides precise signal arrival times for subsequent TDOA localization algorithms, as shown in the sensor signal peak time extraction in Figure 4(a).

[0131] Furthermore, polarization-maintaining fiber is selected as the transmission medium to suppress polarization fading. Fiber optic probes in gas path systems often experience polarization state changes due to pipe vibration or bending, leading to photoacoustic signal attenuation. Polarization-maintaining fiber maintains the polarization state of incident light through a special fiber structure, avoiding the impact of polarization state changes on the photoacoustic signal and ensuring signal transmission stability. In array-type fiber optic probe arrangements, polarization-maintaining fiber is uniformly laid along the gas path branches, with a fiber Bragg grating sensor installed at each branch node, further enhancing the spatial resolution of the signal.

[0132] Furthermore, a reflective optical path is formed by placing a gold-plated reflector at the end of the optical fiber. The gold-plated reflector has a reflectivity of ≥99.5%, completely reflecting the incident light back into the fiber, thus extending the optical path to twice the fiber length. A longer optical path means increased interaction time between the light and the leaking gas. Especially in trace leak scenarios, greater light absorption increases the intensity of the photoacoustic signal and enhances the signal-to-noise ratio. The reflective optical path design also reduces laser power loss, ensuring sufficient light intensity reaching the leak area and meeting the requirements of long-distance transmission in gas path systems.

[0133] The advantages of this embodiment are that the high-resolution spatial perception capability of phase-sensitive optical time-domain reflectometry can accurately identify leaks in small intervals of gas path branches, providing precise data for the positioning algorithm; the use of polarization-maintaining fiber effectively suppresses polarization fading and ensures the stability of signal transmission; the reflective optical path design of the gold-plated mirror enhances the interaction between light and leaking gas, improving the detection capability of trace leak signals.

[0134] Example 7: To address the issue of reduced gas absorption efficiency due to environmental factors affecting laser parameters, this example further refines the adaptive adjustment method for laser parameters of a specific wavelength. By dynamically adjusting the wavelength, power, and linewidth, it ensures that the laser always matches the absorption characteristics of the target gas.

[0135] Based on two adjacent absorption lines of the target leaking gas, the output wavelength of the tunable laser is automatically adjusted using a wavelength-locking circuit. The absorption lines of the target gas are fundamental to laser absorption. As shown in Figure 2, two spectral lines near 1.65 μm for methane indicate that wavelength shift will cause the laser to deviate from the absorption peak, reducing absorption efficiency. The wavelength-locking circuit compares the laser's output wavelength with the standard absorption lines and uses feedback control to adjust the laser's drive current, stabilizing the wavelength within the 1.5 μm-1.6 μm range with a tuning accuracy ≤0.01 nm, ensuring accurate laser alignment with the absorption peak.

[0136] Furthermore, the output power of the laser is adjusted by the laser's power control module based on the length of the gas path system's pipe. When light propagates through optical fibers, its intensity attenuates due to fiber losses such as absorption and scattering; the longer the pipe, the more severe the attenuation. The power control module detects the light intensity at the end of the optical fiber and adjusts the laser's output power to compensate for attenuation during transmission, ensuring sufficient light intensity reaches the leak point and maintaining the photoacoustic signal strength. For example, when the pipe length increases by 100 meters, the power control module automatically increases the output power by 10% to keep the light intensity at the leak point constant.

[0137] Furthermore, based on the temperature changes in the gas path system, the laser's linewidth control circuit adjusts the linewidth to maintain monochromaticity. Temperature changes intensify the thermal motion of gas molecules, causing the absorption spectral lines to broaden, i.e., the linewidth increases. If the laser linewidth is too wide, it will cover non-absorption regions, reducing absorption efficiency; if the linewidth is too narrow, the laser may not be able to cover the entire absorption peak due to spectral broadening. The linewidth control circuit adjusts the laser's resonant cavity length to change the laser linewidth, matching it to the spectral linewidth after the temperature change, maintaining the laser's monochromaticity and ensuring high absorption efficiency. For example, when the temperature increases by 10°C, the linewidth control circuit increases the laser linewidth from 0.05 nm to 0.08 nm to cover the broadened absorption spectral lines.

[0138] The advantages of this embodiment are that the wavelength-locking circuit ensures that the laser is always aligned with the absorption peak of the target gas, maintaining high absorption efficiency; the power control module compensates for the light transmission attenuation in long pipes, ensuring sufficient light intensity at the leak point; and the linewidth control circuit adapts to spectral line broadening caused by temperature changes, maintaining the monochromaticity of the laser. These adaptive adjustment measures enable the laser parameters to dynamically match environmental changes, ensuring that the laser maintains high absorption efficiency for the target gas under different gas path lengths and temperature conditions, thus improving the adaptability and reliability of photoacoustic spectroscopy detection.

[0139] Example 8: To address the issue of decreased leak point location accuracy caused by static gas system topology models, this example further optimizes the dynamic construction method of the gas system topology model. First, an initial 3D topology model of the gas system is constructed using 3D drawing software (such as SolidWorks). The model fully includes the branch connection nodes, pipe diameters, and routing information of the gas path. For example, the initial model of the three-branch gas system shown in Figure 4(c) is constructed based on design drawings, clearly defining the coordinates of the connection nodes of the three branches, the diameter of each pipe segment, and its routing trend. The initial 3D topology model is exported to STL format and converted into a mathematical model using a 3D model conversion tool (such as MeshLab). The axis of each branch is represented by parametric equations. For example, the axis equation of a straight line branch is x=at+b, y=ct+d, z=et+f, t∈[0,1], a, b, c, d, e, and f are constants, and the node coordinates are stored in Cartesian coordinates, forming the mathematical topology model of the gas system.

[0140] Furthermore, flow sensors installed on each branch of the gas path system collect flow data in real time. For example, electromagnetic flow sensors are installed at the end of each branch of a three-way branch to obtain the medium flow rate value in real time. Combining the flow distribution laws of the gas path system, such as the law of conservation of mass (the flow rate into a node equals the flow rate out of a node), the branch orientation and node positions in the topology model are corrected. For example, if the measured flow rate of a certain branch is lower than the expected flow rate in the initial model, it may be due to a deviation between the branch's orientation design and the actual fluid flow, such as an excessively large bend angle leading to increased flow resistance. In this case, based on the flow data and fluid dynamics calculations, such as using CFD software to simulate the flow state of the branch, the branch's orientation parameters are adjusted. For example, the original 90-degree bend is adjusted to a 45-degree bend, and the node position is adjusted, such as moving the connection node between the branch and the main road 5 cm upstream of the main road, so that the corrected topology model better matches the actual fluid flow state of the gas path system.

[0141] Through the above dynamic construction method, the gas path system topology model can reflect the actual structure and flow state of the gas path in real time. For example, the three-branch topology model shown in Figure 4(c), after being corrected by flow data, has branches that are more in line with the actual medium flow path and the node positions are more accurate. This provides a more reliable spatial model basis for locating leak points and improves the accuracy of leak point location.

[0142] Example 9: To address the interference of environmental vibration and noise on photoacoustic signal detection, this example further optimizes the multi-dimensional suppression method for environmental vibration and noise. First, piezoelectric vibration sensors are installed on the fixed support of the pneumatic system along multiple directions, such as the x-axis, y-axis, and z-axis (three orthogonal directions). The sensors are magnetically attached to the metal surface of the support using strong magnets, allowing for easy adjustment of the installation position according to testing requirements. For example, an x-axis sensor is installed on the cantilever section of the support to collect horizontal vibration, a y-axis sensor is installed on the vertical column of the support to collect vertical vibration, and a z-axis sensor is installed on the bottom connecting beam of the support to collect axial vibration, thus comprehensively collecting vibration signals from the support in all three directions.

[0143] Furthermore, the sensor mounting points are optimized through vibration modal testing. Experimental modal testing methods, such as the impact test, are used to perform vibration modal analysis on the fixed support: different locations on the support are struck with a hammer, and the vibration response at each point is measured using piezoelectric sensors to obtain the support's natural frequencies and mode shapes. The locations with the largest vibration amplitudes on the support, such as the cantilever ends or the connection points between two supports, are identified. Piezoelectric sensors are then installed at these locations to maximize the acquisition of environmental vibration and noise signals. For example, a curve showing the vibration amplitude versus location is plotted using the impact test, and the location with the largest vibration amplitude is determined as the sensor mounting point, thus improving the efficiency of vibration signal acquisition.

[0144] Furthermore, the vibration sensor signals from multiple directions are weighted and fused to form a reference input signal. The weights are determined based on the degree of influence of vibration in each direction on the photoacoustic signal: through correlation analysis, the correlation coefficient between the vibration signal in each direction and the output signal of the photoacoustic sensor is calculated. The larger the correlation coefficient, the greater the interference of vibration in that direction on the photoacoustic signal. Weights are assigned according to the correlation coefficients. For example, if the correlation coefficients for the x-axis direction are 0.8, y-axis is 0.5, and z-axis is 0.3, then the weights for the x-axis signal are set to 0.5, y-axis to 0.3, and z-axis to 0.2. The fused reference signal better reflects the vibration components that have the greatest impact on the photoacoustic signal. It is input into an adaptive filter, such as the variable step size RLS filter shown in Figure 3, and processed as a noise reference signal and the observed signal of the photoacoustic sensor. For example, in the schematic diagram of the adaptive filtering noise separation effect shown in Figure 3, the reference signal is the multi-directional fused vibration signal. After filtering, the separated leakage signal is closer to the real leakage signal, effectively suppressing the interference of environmental vibration noise.

[0145] Example 10: To address the issue of insufficient accuracy in machine learning models due to the underutilization of historical leakage data, this example further optimizes the method for utilizing historical leakage data. First, the historical leakage data is structured and stored in a relational database (such as MySQL). The data includes the three-dimensional coordinates of the leakage point, leakage rate, environmental parameters (temperature, humidity), and photoacoustic signal characteristics (amplitude, phase difference, frequency offset). For example, a "Leakage Event" table is created, with fields including "Leakage Point X Coordinate," "Leakage Point Y Coordinate," "Leakage Point Z Coordinate," "Leakage Rate (sccm)," "Ambient Temperature (°C)," "Ambient Humidity (%RH)," "Photoacoustic Signal Amplitude (au)," "Photoacoustic Signal Phase Difference (rad)," and "Photoacoustic Signal Frequency Offset (Hz)." Relevant data for each leakage event is entered into the table, forming a structured historical database.

[0146] Furthermore, data mining algorithms were used to extract the correlation between leakage rate and environmental parameters, as well as the mapping pattern between photoacoustic signal characteristics and leakage point location. Association rule mining algorithms, such as the Apriori algorithm, were employed, with minimum support (e.g., 0.2) and minimum confidence (e.g., 0.7) set to mine the correlation between environmental parameters and leakage rate. For example, when the ambient temperature is above 30℃ and the humidity is below 50%, the amplitude characteristic of the leakage rate increases by 15% compared to normal temperature and humidity conditions. This indicates that temperature and humidity have a significant impact on the amplitude characteristic of the leakage rate. Classification algorithms (e.g., decision trees) were used to extract the mapping pattern between photoacoustic signal characteristics and leakage point location. For example, when the phase difference of the photoacoustic signal is between -0.5 rad and -0.3 rad and the frequency shift is between 0.2 Hz and 0.4 Hz, the leakage point is located at the end of a branch in the gas path system. This pattern reflects the correspondence between photoacoustic signal characteristics and leakage point location.

[0147] The machine learning model is retrained using the patterns and rules discovered through data mining. For example, ambient temperature and humidity are added as additional input features to the machine learning model, as shown in the support vector machine model in Figure 4(d). The weight of the temperature feature in the model is adjusted based on the "temperature-amplitude" correlation obtained through association rule mining; the classification boundaries of these features in the model are adjusted based on the "phase difference-frequency offset-leakage location" mapping pattern obtained through decision tree mining. The retrained model can more accurately utilize the patterns in historical data. For example, in the schematic diagram of the machine learning model output shown in Figure 4(d), the X, Y, and Z coordinates predicted by the retrained model are closer to the true values, thus optimizing the accuracy of leak detection.

[0148] Example 11: To address the problem of insufficient visualization of leak detection results, which makes it difficult for operators to quickly understand the information, this example further optimizes the visualization method for leak detection results. First, the three-dimensional coordinates of the leak point are overlaid onto the gas path system topology model, and the leak point location is marked with different colors. For example, using the leak location visualization diagram shown in Figure 6, the three-dimensional topology model of the gas path system, such as a three-way branch model, is displayed in the center of the interface. The three-dimensional coordinates of the leak point (e.g., X=1.2m, Y=0.8m, Z=2.5m) are marked with red dots at the corresponding positions on the model. The size of the dots is adjusted according to the leak rate; the higher the leak rate, the larger the dot. The colors are divided according to the leak level: red indicates a high leak rate, yellow indicates a medium leak rate, and green indicates a low leak rate, enabling operators to quickly locate the leak point and its severity.

[0149] Furthermore, the real-time leakage rate value and historical variation curve are displayed on the same interface. The real-time value is displayed numerically on the interface in sccm, such as "Current leakage rate: 0.15sccm"; the historical variation curve plots the leakage rate change over the past 24 hours with time on the horizontal axis in minutes and leakage rate on the vertical axis in sccm, allowing operators to intuitively see the trend of leakage rate changes, such as whether it is increasing or remaining stable.

[0150] The visualization interface is developed using WebGL technology, supporting zoom, rotation, and pan operations to adapt to display devices of different sizes. For example, on a computer, operators can zoom the interface using the mouse wheel, rotate the interface by dragging the mouse, and pan the interface with the left mouse button; on a tablet, operators can zoom, rotate, and pan by swiping their fingers, making it convenient to view detection results on mobile devices. For example, as shown in Figure 6, the visualization interface allows operators to rotate the model to view the back of the leak point, zoom to view the specific location of the leak point, and pan to browse the entire gas path system, improving the flexibility and ease of use of visualization.

[0151] Example 12: To address the sensitivity drift issue caused by long-term use of fiber optic probes, this example further optimizes the periodic calibration method for gas path branch array fiber optic probes. First, periodic calibration is performed: a standard leak source is connected to the calibration section of the gas path system, such as a 0.05 sccm or 0.1 sccm hydrogen leak source. The calibration section is located at the connection point between the main path and the branch of the gas path system, and has standard pipe dimensions, such as a diameter of 25 mm and a length of 1 m, to ensure the consistency of the simulated leak and to simulate real leak scenarios.

[0152] Furthermore, photoacoustic signals from the calibration section are acquired, and their characteristic parameters are extracted. An array of fiber optic probes is used to acquire the photoacoustic signals from the calibration section. Signal processing algorithms such as Fast Fourier Transform are used to extract the amplitude, cross-correlation is used to calculate the phase difference, and spectral analysis is used to obtain the frequency offset. The three characteristic parameters—amplitude, phase difference, and frequency offset—are extracted. For example, for a standard leakage source with a 0.05 sccm amplitude, the acquired photoacoustic signal has an amplitude of 0.3 au, a phase difference of -0.1 rad, and a frequency offset of 0.15 Hz; for a standard leakage source with a 0.1 sccm amplitude, the acquired signal has an amplitude of 0.6 au, a phase difference of -0.2 rad, and a frequency offset of 0.3 Hz.

[0153] Furthermore, the extracted feature parameters are compared with the pre-calibrated photoacoustic response model to calculate the deviation of the feature values. The pre-calibrated photoacoustic response model is shown in Figure 5: the relationship between amplitude and leakage rate is a linear fitting curve, the relationship between phase difference and leakage rate is a linear fitting curve, and the relationship between frequency shift and leakage rate is a quadratic fitting curve. For example, for a standard leakage source of 0.05 sccm, the model predicts an amplitude of 0.32 au, the measured amplitude is 0.3 au, and the deviation is -0.02 au; the model predicts a phase difference of -0.11 rad, the measured phase difference is -0.1 rad, and the deviation is 0.01 rad; the model predicts a frequency shift of 0.16 Hz, the measured frequency shift is 0.15 Hz, and the deviation is -0.01 Hz.

[0154] Furthermore, if the deviation exceeds a preset threshold, such as amplitude deviation exceeding ±5%, phase difference deviation exceeding ±0.05 rad, or frequency offset deviation exceeding ±0.02 Hz, the sensitivity coefficient of the fiber optic probe is adjusted. For example, if the amplitude deviation is -0.02 au, the sensitivity coefficient of the fiber optic probe is increased: by adjusting the alignment position of the fiber optic cable and the photoacoustic cell, such as moving the fiber optic cable 1 mm towards the center of the photoacoustic cell, the coupling efficiency of the fiber optic cable is improved, increasing the amplitude of the acquired photoacoustic signal. For example, if the measured amplitude after adjustment is 0.31 au, the deviation is reduced to -0.01 au, meeting the threshold requirement. After adjustment, the photoacoustic signal is acquired again to verify whether the deviation is within the threshold range. The calibration results are stored in a relational database, such as MySQL, recording information such as the calibration date, standard leakage source, measured characteristic value, deviation, and adjusted sensitivity coefficient, as reference data for subsequent detection. For example, as shown in Figure 5, the measured data points after calibration are closer to the fitted curve in the photoacoustic response model calibration curve, improving the accuracy of the photoacoustic signal characteristics and thus improving the accuracy of leakage detection.

[0155] Example 13: To address the issues of module coordination and spatial positioning accuracy in the photoacoustic spectroscopy trace detection system, this example further refines the specific implementation of the photoacoustic spectroscopy trace detection system for gas path system leaks. The laser emission module serves as the system's signal source. Its tunable laser selects a specific wavelength based on the characteristic absorption peak of the leaking gas, such as the 1064nm absorption peak of hydrogen. A dual-frequency signal generator generates a modulation signal composed of two superimposed sinusoidal signals of different frequencies. The intensity modulator receives this signal and performs dual-frequency intensity modulation on the laser, enabling the laser to carry dual-frequency information and be emitted to the suspected leak area. The gas path branch array fiber optic probe module is arranged in multiple dimensions along the connection nodes, bends, and ends of the main gas path and its branches. Multiple fiber optic sensors are connected to the laser emission module via fiber bundles to collect the photoacoustic signals generated after the leaking gas absorbs laser energy. The spatial distribution information of the photoacoustic signals is formed by the installation coordinates of each sensor (pre-recorded x, y, and z axis positions of the system).

[0156] Furthermore, the spatial positioning module receives spatial distribution information from the fiber optic probe module and simultaneously retrieves gas path topology model data from the system database, including parameters such as pipe direction, node coordinates, and pipe diameter, as well as flow distribution data, including real-time flow of each branch. This data is then input into the spatial positioning algorithm, which fuses the TDOA algorithm and the machine learning model. The TDOA algorithm first calculates the time difference between the photoacoustic signals received by different fiber optic sensors and, combined with the spatial location of the sensors, preliminarily determines the range of the leak point. The machine learning model utilizes the pipe connection relationships and flow distribution data in the gas path topology model; for example, if the flow rate of a branch is abnormally reduced, it indicates that the leak may be located in that branch, correcting the TDOA results and finally outputting the three-dimensional coordinates of the leak point. The adaptive filtering module receives the photoacoustic signal from the fiber optic probe module and the fused signal from the multi-directional vibration sensors (x, y, z axes) mounted on the gas path fixed support. The fusion weights are determined based on the correlation analysis between vibration and photoacoustic signals in each direction. Using the photoacoustic signal as the observation signal and the fused vibration signal as the reference signal, a variable step-size recursive least squares algorithm is used to dynamically adjust the filtering parameters, separating environmental vibration noise from the leak characteristic signal, and outputting a clean characteristic signal to the leak judgment module.

[0157] Furthermore, the leak detection module extracts multi-dimensional features from the characteristic signals, including the amplitude and phase difference of the signal reflecting the concentration of the leaking gas (i.e., the phase difference of the signals received by different sensors, used to assist in localization), and frequency offset (i.e., the shift in laser frequency due to gas absorption, correlated with the leak rate). These features are combined with a pre-calibrated photoacoustic response model and simultaneously input into a machine learning model. The model determines the existence of a leak through feature matching and calculates the leak rate based on the photoacoustic response model. The visualization module overlays the three-dimensional coordinates of the leak point output by the spatial positioning module onto the gas path topology model, marking the leak location with color-gradient dots. Simultaneously, it displays the real-time value of the leak rate and the change curve over the past 24 hours on the right side of the interface. Operators can zoom, rotate, and pan the model using a mouse or touch screen to intuitively view the specific location of the leak point in the gas path system.

[0158] The advantages of this embodiment are that the dual-frequency intensity modulation of the laser emission module ensures the recognizability of the photoacoustic signal, the multi-dimensional arrangement of the gas path branch array fiber optic probe achieves spatial coverage of the photoacoustic signal, the fusion algorithm of the spatial positioning module improves the accuracy of the leak point coordinates, the variable step size algorithm of the adaptive filtering module effectively suppresses environmental vibration noise, the multi-feature fusion of the leak judgment module improves the reliability of leak identification, and the three-dimensional overlay display of the visualization module enhances the readability of the detection results. The collaborative work of these modules forms a complete photoacoustic spectral trace detection system.

[0159] Example 14: To address the issues of ease of field deployment and integration in photoacoustic spectroscopy trace detection systems, this example further refines the specific implementation of the photoacoustic spectroscopy trace detection device for gas path system leakage. The device housing is an anti-static PC housing, with reserved installation positions for the display interface and gas path connection interfaces. Heat dissipation holes are provided on both sides, a non-slip handle is provided on the top, and rubber feet are installed on the bottom to reduce vibration transmission. The gas path connection interfaces are located on the lower front of the housing, employing a quick-connect structure. The interface specifications match the gas path system piping, and each interface has an embedded nitrile rubber sealing gasket. Connections can be quickly inserted and sealed without tools. The number of interfaces corresponds to the number of gas path branches; for example, a three-way branch gas path has four interfaces, connecting to the main path and three branches respectively.

[0160] Furthermore, the display interface is located on the outer casing and supports multi-touch. The interface layout is divided into a gas path topology model display area on the left (overlaying leak points), a leak rate curve area on the right (real-time values ​​and historical changes), and an operation button area at the bottom (start detection, stop detection, calibration). The display screen is covered with scratch-resistant tempered glass to adapt to frequent operation in industrial environments. The power module is built into the casing and adopts a redundant design. The main power supply is AC220V input. When the main power supply is interrupted, the backup power supply automatically switches to ensure continuous operation of the device. The power module also integrates overload protection and voltage stabilization circuits to prevent voltage fluctuations from damaging the internal electronic modules.

[0161] Furthermore, the communication module is located inside the casing and uses wireless communication. It connects to an external antenna via an antenna interface on the back of the casing to enhance the signal. The communication module transmits data such as the detected leak location, leak rate, and the operating status of each module (e.g., laser emission power, fiber optic probe sensitivity) to the gas path system monitoring platform in real time. Simultaneously, it receives instructions from the monitoring platform, such as adjusting the laser wavelength and modifying the detection threshold, enabling remote management and control. The device integrates a photoacoustic spectroscopy trace detection system. The laser emission module, gas path branch array fiber optic probe module, spatial positioning module, adaptive filtering module, leak detection module, and visualization module are all mounted on a metal bracket inside the casing. The bracket features a perforated design to facilitate heat dissipation. All modules are connected via high-speed data cables to ensure timely data transmission.

[0162] The advantages of this embodiment are that the anti-static PC shell effectively prevents electrostatic interference, the quick-connect pneumatic connection interface simplifies the on-site installation process, the touch display interface improves the ease of operation, the redundant power supply module ensures continuous working capability, the wireless communication module enables remote monitoring, and the overall device has high integration and good portability, adapting to the complex environmental requirements of industrial sites.

[0163] Example 15: To address the issues of computer implementation and data storage stability in the photoacoustic spectroscopy trace detection method, this example further refines the specific implementation of the computer-readable storage medium. The computer-readable storage medium uses a solid-state drive (SSD), which features high-speed read / write capabilities and is suitable for storing large amounts of detection data and computer programs. The computer program within the storage medium adopts a modular design, including functional modules such as laser emission control, fiber optic probe data acquisition, spatial positioning calculation, adaptive filtering, leakage detection, and visualization display. These modules coordinate through a data bus.

[0164] The laser emission control module sends wavelength adjustment commands to the tunable laser and simultaneously controls the dual-frequency signal generator to generate superimposed signals with frequencies of 1kHz and 10kHz. This signal is then input to the intensity modulator to achieve dual-frequency intensity modulation of the laser. The fiber optic probe data acquisition module connects to the gas path branch array fiber optic probe via a USB interface, acquiring the photoacoustic signals of each fiber optic sensor in real time and recording the sensor's installation coordinates (x, y, z axis positions) to form the spatial distribution information of the photoacoustic signals, which is stored in a temporary data area on the solid-state drive. The spatial positioning calculation module reads the spatial distribution information from the temporary data area and retrieves the stored gas path topology model and flow distribution data. It then executes a spatial positioning algorithm that integrates the TDOA algorithm and a machine learning model: the TDOA algorithm calculates the approximate range of the leak point based on the signal arrival time difference, while the machine learning model uses the pipe connection relationships and flow data in the gas path topology to correct this range, outputting the three-dimensional coordinates of the leak point to the result data area.

[0165] The advantages of this embodiment are that the high-speed read and write of the solid-state drive ensures the real-time performance of data processing, the modular computer program enables the step-by-step execution of the detection method, the data partitioning storage improves the orderliness of data management, and the overall storage medium design meets the computer implementation requirements of the photoacoustic spectroscopy trace detection method, providing data support for the stable operation of the detection system.

[0166] Example 16: To address the issues of portability and interaction with external devices in photoacoustic spectroscopy trace detection methods, this example further refines the specific implementation of an electronic product integrating a computer-readable storage medium. The electronic product uses an industrial panel PC with an aluminum alloy casing, an IP65 protection rating, and is dustproof and waterproof, suitable for handheld or fixed installation in the field. The panel PC integrates a solid-state drive to store computer programs and detection data, used to execute the computer programs stored in the storage medium, achieving full-process control of photoacoustic spectroscopy trace detection.

[0167] The communication module employs dual-mode Wi-Fi 6 and 5G, transmitting data via an antenna on the top of the tablet. It sends the detected leak location, leak rate, and equipment status to the gas path system monitoring platform in real time, while simultaneously receiving remote commands from the monitoring platform, such as adjusting the laser wavelength or initiating calibration procedures. Input interfaces include a USB Type-C port for connecting gas path system parameter setting devices, such as a keyboard or touchscreen, and a Micro SD card slot for expanded storage. Operators can input parameters such as the gas path system's pipe diameter and medium type via the USB interface. These parameters are received by the computer program and used to adjust the parameters of the photoacoustic response model. For example, the pipe diameter affects the propagation path of the photoacoustic signal, and the program corrects the path parameters in the model based on the input value. The output interface is HDMI, used to connect external display devices, such as large-screen monitors, to overlay the leak point's gas path topology model and leak rate curve onto the visual interface for easy viewing by on-site operators.

[0168] Furthermore, the tablet's operating system supports industrial-grade stability, and the computer program, written in C++, boasts excellent real-time performance. During program execution, the laser emission control module first sends commands to the tunable laser, selecting a specific wavelength, such as 1064nm for hydrogen, and controls the dual-frequency signal generator and intensity modulator to achieve dual-frequency intensity modulation. Next, the fiber optic probe data acquisition module connects to the gas path branch array fiber optic probe via a USB interface to acquire photoacoustic signals and spatial distribution information. The spatial positioning calculation module uses the acquired information and stored gas path topology and flow data to execute a fusion algorithm to calculate the three-dimensional coordinates of the leak point. The adaptive filtering module separates environmental vibration noise from the photoacoustic signal. The leak detection module extracts features and determines the leak and its rate. Finally, the visualization display module generates an interface, outputting it to an external monitor via an HDMI interface or displaying it directly on the tablet screen.

[0169] The tablet features a high-capacity lithium battery that supports fast charging and is suitable for extended field use. The touchscreen is compatible with gloves. Ventilation vents on the sides of the casing ensure the processor operates normally in high-temperature environments, and a magnetic bracket at the bottom allows for secure mounting to the pneumatic system support, enabling hands-free operation.

[0170] The advantages of this embodiment are that the portability and protection level of the industrial tablet PC meet the needs of industrial field use, the communication module enables remote monitoring and interaction, the input / output interface enhances the expandability of the device, the integrated solid-state drive and high-performance processor ensure the efficient operation of the computer program, and the overall electronic product design realizes the portability and intelligence of the photoacoustic spectroscopy trace detection method, providing a flexible solution for on-site leak detection.

[0171] Although the present invention has been specifically described above with reference to preferred embodiments, it should be understood that the present invention is not limited to the embodiments described above. Various modifications and variations can be made by those skilled in the art without departing from the spirit of the present invention, and such modifications and variations should fall within the scope defined by the appended claims and their equivalents.

Claims

1. A photoacoustic spectroscopy method for detecting trace leaks in a gas system, characterized in that the steps include... include: A specific wavelength laser, modulated by dual-frequency intensity, is emitted towards the suspected leak area of ​​the gas path system. The specific wavelength matches two adjacent absorption lines of the target leaking gas, and the frequency difference of the dual-frequency intensity modulation corresponds to the frequency interval of the two adjacent absorption lines. This allows the leaking gas molecules to simultaneously absorb laser energy at two frequencies, thereby enhancing the amplitude of the periodic photoacoustic signal. The photoacoustic signal is acquired using a gas path branch array fiber optic probe. This probe includes multiple fiber optic sensors arranged axially, radially, and circumferentially along the gas path branches. Each fiber optic sensor employs a reflective optical path, with a path length twice the fiber length, to acquire the spatial distribution information of the photoacoustic signal. Based on the... Spatial distribution information, combined with the gas path system topology model and flow distribution data, is used to calculate the three-dimensional coordinates of the leak point through a spatial positioning algorithm that integrates the TDOA algorithm and a machine learning model, enabling centimeter-level positioning. The photoacoustic signal is adaptively filtered, using the fused signal from a multi-directional vibration sensor on the gas path system's fixed support as a reference input. A variable step-size recursive least squares algorithm is used to dynamically adjust the filter coefficients, separating environmental vibration noise from the leak characteristic signal. Based on the separated leak characteristic signal, multi-dimensional features such as amplitude, phase difference, and frequency shift are extracted. Combined with a pre-calibrated photoacoustic response model and a machine learning model trained on historical leak data, the existence and rate of the leak are determined.

2. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The dual-frequency intensity modulation is implemented as follows: a tunable laser outputs a laser of a specific wavelength, and a dual-frequency signal generator generates two modulation frequencies f1 and f2, where the difference between f1 and f2 is equal to the frequency interval between two adjacent absorption spectral lines of the target leaking gas; the dual-frequency signal is input to an intensity modulator to modulate the laser intensity so that the laser intensity changes periodically with f1 and f2.

3. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The calculation of the three-dimensional coordinates of the leak point includes the following steps: Preprocessing the photoacoustic signal collected by each fiber optic sensor: using a bandpass filter to filter high-frequency noise, and then extracting the signal envelope through Hilbert transform, finding the peak position of the envelope as the arrival time of the photoacoustic signal of that sensor; Calculating the time difference of arrival (TDOA) between any two fiber optic sensors: for N fiber optic sensors, calculating C(N,2) time differences to form a time difference matrix; Importing the gas path system topology model: converting the three-dimensional CAD model of the gas path into a mathematical model, including the coordinates of the connection nodes, pipe diameter, and direction of each branch, as geometric constraints for positioning; Incorporating flow distribution data: obtaining the real-time flow of each branch from the flow sensors of the gas path system, and correcting the possible location range of the leak point based on the relationship between flow rate and leakage rate; Training the machine learning model: training a support vector machine model using experimental data of known leak points, with the input being the time difference matrix, topology model parameters, and flow data, and the output being the leak point coordinates, optimizing the kernel function and penalty parameters of the model; Calculating the leak point coordinates: inputting the time difference matrix, the parameters of the topology model, and the flow distribution data into the trained support vector machine model, and outputting the three-dimensional coordinates of the leak point.

4. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The specific steps of the variable step-size recursive least squares algorithm are as follows: Initialization: Set the filter order M, initialize the filter coefficient vector w(0)=0, and initialize the covariance matrix. Where δ is a small positive number, It is the identity matrix; Acquiring reference signal and original photoacoustic signal: The reference signal x(n) is a weighted fusion of vibration sensor signals from three directions, and the original photoacoustic signal d(n) is the signal acquired by the fiber optic probe, where n represents the number of samplings; Calculate the filter output: , where y(n) is the filter output and w(n) is the filter coefficient vector. Let T be the reference signal vector, and T denote the transpose of the vector; calculate the error signal: , where e(n) is the estimated value of the leakage characteristic signal after separation; Adjusting the step size: Using an adaptive step size formula Where μ0 is the initial step size and α is the adjustment factor; update the filter coefficients: Update the covariance matrix: ； The process of acquiring the reference signal and the original photoacoustic signal is repeated until the variance of the error signal e(n) is less than a preset threshold, thus completing the separation of environmental vibration and noise.

5. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The calibration process of the photoacoustic response model includes: selecting standard leakage sources: selecting standard leakage sources with known leakage rates, and selecting at least 3 samples for each leakage rate; building a calibration experimental platform: connecting the standard leakage sources to the test section of the gas path system to simulate a real leakage scenario, arranging gas path branch array fiber optic probes and vibration sensors, and setting laser emission parameters; collecting calibration data: for each standard leakage source, collecting at least 10 sets of photoacoustic signals, with each set collected for 60 seconds, and extracting the amplitude, phase difference, and frequency shift of each set of signals; fitting the mapping relationship: for each leakage rate, calculating the average value of the multi-dimensional features of multiple samples and multiple sets of photoacoustic signals to form a feature vector; using a multiple linear regression model to fit the mapping relationship between the leakage rate and the feature vector to obtain the photoacoustic response model. Where R is the leakage rate, A is the amplitude, Δφ is the phase difference, Δf is the frequency offset, and a, b, c, and d are regression coefficients; Verify model accuracy: Test the model using a standard leakage source that has not been calibrated, and calculate the relative error between the model's predicted value and the actual value.

6. The photoacoustic spectral trace detection method for gas system leaks as described in claim 1, characterized in that, The deployment method of the fiber optic sensor includes: using phase-sensitive optical time-domain reflectometry to achieve high-resolution spatial sensing of photoacoustic signals, selecting polarization-maintaining fiber as the transmission medium to suppress polarization fading, and forming a reflective optical path by setting a gold-plated reflector at the end of the fiber to enhance the interaction between light and leaked gas.

7. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The adaptive parameter adjustment method for the specific wavelength laser includes: automatically adjusting the output wavelength of the tunable laser based on two adjacent absorption spectral lines of the target leaking gas, with a range of 1.5μm-1.6μm and a wavelength tuning accuracy ≤0.01nm; adjusting the output power of the laser based on the pipe length of the gas path system to compensate for light attenuation during transmission; and adjusting the linewidth of the laser based on the temperature change of the gas path system to maintain the monochromaticity of the laser and improve gas absorption efficiency.

8. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The dynamic construction method of the gas path system topology model includes: using 3D drawing software to construct an initial 3D topology model of the gas path system, including branch connection nodes, pipe diameter and direction, exporting it to STL format and converting it into a mathematical model, including the axis equation and node coordinates of each branch; collecting the flow data of each branch in real time through the flow sensor of the gas path system, and correcting the branch direction and node position in the topology model in combination with the flow distribution law, so as to dynamically update the gas path system topology model.

9. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The multi-dimensional suppression method for environmental vibration and noise includes: setting piezoelectric vibration sensors in multiple directions on the fixed support of the air circuit system, using a magnetic installation method, and optimizing the sensor installation points based on the vibration mode test results of the support; weighting and fusing the vibration sensor signals from multiple directions to form a reference input signal to improve the suppression effect of adaptive filtering on environmental vibration and noise.

10. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, The method for utilizing the historical leakage data includes: storing the historical leakage data in a structured relational database; extracting the correlation between leakage rate and environmental parameters, as well as the mapping pattern between photoacoustic signal characteristics and leakage point location, using data mining algorithms; and retraining the machine learning model with the mined patterns to optimize the model's leakage judgment accuracy.

11. The photoacoustic spectral trace detection method for gas system leakage as described in claim 1, characterized in that, It also includes visualizing the leak detection results, with the following steps: superimposing the three-dimensional coordinates of the leak point onto the gas path system topology model and marking the location of the leak point with different colors; displaying the real-time value of the leak rate and the historical change curve on the same interface, with the horizontal axis of the curve representing time and the vertical axis representing the leak rate; The visual interface is developed using WebGL technology and supports scaling, rotation, and translation operations to adapt to display devices of different sizes.

12. The photoacoustic spectral trace detection method for gas system leakage as described in claim 10, characterized in that, It also includes periodically calibrating the gas path branch array fiber optic probe, with the following steps: periodically calibrate by connecting a standard leak source to the calibration section of the gas path system to simulate a leak scenario; collect the photoacoustic signal of the calibration section and extract the amplitude, phase difference, and frequency shift; compare the extracted features with the pre-calibrated photoacoustic response model and calculate the deviation of the feature values; If the deviation exceeds the preset threshold, the sensitivity coefficient of the fiber optic probe is adjusted, and the calibration result is stored in the relational database.

13. A photoacoustic spectral trace detection system for gas system leaks, used to implement the photoacoustic spectral trace detection method for gas system leaks as described in any one of claims 1-12, characterized in that, The system includes: a laser emitting module for emitting a specific wavelength laser modulated by dual-frequency intensity towards a suspected leak area; the laser emitting module comprising a tunable laser, a dual-frequency signal generator, and an intensity modulator; a gas path branch array fiber optic probe module comprising multiple fiber optic sensors arranged in multiple dimensions along the gas path branches for acquiring photoacoustic signals and obtaining spatial distribution information; a spatial positioning module for receiving spatial distribution information of photoacoustic signals, a gas path topology model, and flow distribution data, and calculating the three-dimensional coordinates of the leak point using a spatial positioning algorithm that fuses the TDOA algorithm and a machine learning model; an adaptive filtering module for receiving fused signals from photoacoustic signals and multi-directional vibration sensors, and separating environmental vibration noise and leak characteristic signals using a variable step-size recursive least squares algorithm; a leak detection module for receiving the separated characteristic signals, extracting multi-dimensional features, and determining the existence and rate of a leak by combining a photoacoustic response model and a machine learning model; and a visualization module for visually displaying the location and rate of the leak point superimposed on the gas path topology model.

14. A photoacoustic spectral trace detection device for gas system leaks, characterized in that, The photoacoustic spectroscopy trace detection system for gas system leaks as described in claim 13 further includes: a device housing for accommodating the detection system, the device housing being made of anti-static PC material; a gas path connection interface for connecting the device to a suspected leak area of ​​the gas path system, the gas path connection interface employing a quick-connect structure; a display interface for visually displaying the leak detection results; a power module for providing stable power to the detection system, the power module employing a redundant design to ensure continuous operation; and a communication module for data transmission with the gas path system monitoring platform.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the photoacoustic spectral trace detection method for gas system leakage as described in any one of claims 1-12.

16. An electronic product, characterized in that, An integrated or assembled computer-readable storage medium as described in claim 15 is characterized in that it further comprises: a processor for executing a computer program in the storage medium; a communication module for transmitting data with a gas path system monitoring platform; an input interface for receiving parameter settings of the gas path system; and an output interface for connecting to an external display device to output leak detection results.

Images