A method for monitoring leakage of a sealed connection structure for hydrogen delivery

Abstract

The application discloses a kind of sealing connection structure's leakage monitoring method for hydrogen delivery, it is related to gas delivery and storage monitoring field, comprising: S1: layout annular negative pressure sampling cavity collection multimodal data and calculate angle segment suspicious degree;S2: apply heat pulse and sweep frequency sound wave excitation, collection response signal generates thermoacoustic leakage score;S3: environment compensation and multi-sensor fusion obtain angle segment hydrogen concentration time sequence;S4: based on comprehensive response and sparse-smooth inversion flux inversion realizes leakage positioning;S5: calculate risk score and alarm according to threshold classification.By circumferential layout of sealing connection structure multiple point detection unit and using signal fusion, realize sealing structure full coverage leakage monitoring, high-precision positioning, and combined with risk classification automatic trigger emergency measures, improve discovery timeliness and safety.

Description

Technical Field

[0001] This invention relates to the field of gas transportation and storage monitoring, specifically a method for monitoring leaks in a sealed connection structure used for hydrogen transportation. Background Technology

[0002] Hydrogen, as a highly efficient and clean energy carrier, is finding increasingly wider applications in chemical, energy, and transportation industries. During hydrogen transportation, a sealed connection structure is crucial for ensuring the safe operation of the system. However, hydrogen has a small molecular weight and high permeability, making it extremely prone to leakage through even the smallest gaps. Furthermore, it is colorless and odorless, making it undetectable by conventional senses. Once a leaked hydrogen gas mixes with air to form an explosive mixture, it poses a serious threat to personnel safety and equipment operation.

[0003] Existing hydrogen leak monitoring methods largely rely on single-point sensor detection or manual inspection, which suffers from problems such as response delay, low positioning accuracy, and susceptibility to environmental interference. Under complex operating conditions, especially in multi-connection points and enclosed or semi-enclosed spaces, the timeliness and reliability of leak detection still fall short of the requirements for safe operation. Furthermore, current technologies typically cannot achieve continuous coverage monitoring of the entire circumference of sealed connection structures, lack effective grading mechanisms for different leak intensities, and the detection systems are prone to sensitivity degradation and false alarms in high humidity, high temperature, or vibration environments, making it difficult to provide accurate and stable monitoring data for subsequent safety control or emergency response. Therefore, there is an urgent need for a hydrogen leak monitoring technology capable of rapid detection, precise positioning, and risk grading within the circumference of sealed connection structures to comprehensively improve the system's safety protection capabilities. Summary of the Invention

[0004] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide a leakage monitoring method for a sealed connection structure for hydrogen transportation, so as to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring leakage in a sealed connection structure for hydrogen transportation, comprising:

[0006] S1: Set up an annular negative pressure sampling chamber, extract gas from the chamber according to a preset pattern, use multimodal sensors to collect data from the chamber to calculate the observation scalar, collect machine status to calculate the comprehensive exposure intensity, and generate corner segment suspicion by maximizing operation.

[0007] S2: Apply thermal pulses and sweep frequency acoustic waves to the outer wall of the annular cavity, collect the excitation response signals, calculate the excitation gain of each corner segment, and generate a thermoacoustic leakage score by combining the corner segment suspicion level.

[0008] S3: Based on intracavitary data and acquired environmental data, after dynamic interference compensation, drift correction and cross-sensitivity compensation, multi-task signal fusion is performed by combining a lightweight long short-term memory network to obtain the time series of hydrogen concentration in the corner segment.

[0009] S4: Calculate the comprehensive response index based on the observation scalar, calculate the path fusion concentration by combining the comprehensive exposure intensity and intracavitary data, calculate the inversion flux based on the gradient descent method and introduce the sparse-smoothing inversion objective function, and obtain the leak location by maximizing the inversion flux.

[0010] S5: Calculate risk scores based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leakage score, classify them according to thresholds, and generate corresponding alarm information.

[0011] The present invention is further configured such that S1 includes:

[0012] An annular negative pressure sampling chamber is set up on the outside of each monitored sealed connection structure, and a multimodal sensor group is installed in several corner channels of the sampling chamber;

[0013] Gas is extracted from the cavity by a micro-pump-electronically controlled valve array according to a preset spatiotemporal coding pattern. The cavity data and machine status are collected simultaneously and preprocessed using a filter. The cavity data includes: original hydrogen concentration, cavity micro-pressure difference, and branch flow rate. The machine status includes: valve opening degree and pumping energy.

[0014] The present invention is further configured to synthesize a pattern-level observation scalar based on intracavity data of the same corner segment according to a power transformation, and to perform weighted aggregation of the observation scalars of each corner segment according to the acquired valve opening degree to generate the overall response quantity;

[0015] The overall exposure is calculated by statistically analyzing the power-weighted fusion values ​​of valve opening and pumping energy under all spatiotemporal coding patterns.

[0016] The present invention is further configured to perform sparse reconstruction calculation based on the overall response quantity, valve opening degree and pumping energy, and use the compressed sensing iterative subspace algorithm to recover the corner segment coefficient representation, and obtain the corner segment suspicion degree representing the sparse response intensity of each corner segment by maximizing the calculation.

[0017] The present invention is further configured such that S2 includes:

[0018] Based on the intracavity data collected by the multimodal sensor at the previous moment, the baseline energy value is calculated.

[0019] Micro heating elements and miniature ultrasonic transducers are installed on the outer wall of the annular negative pressure sampling cavity. Thermal pulse signals and swept frequency sound wave signals are applied sequentially according to a preset cycle.

[0020] A new excitation response signal is constructed by synchronously collecting intracavitary data using a multimodal sensor during the excitation process. The excitation response signal includes the following after excitation: excitation hydrogen concentration, intracavitary micro-pressure difference, and excitation branch flow rate.

[0021] The present invention is further configured to calculate the response energy by performing multi-resolution wavelet analysis and high-resolution spectral estimation on the excitation response signal, combined with the baseline energy.

[0022] The excitation gain is calculated for each corner segment based on the difference between the baseline energy value and the response energy before and after excitation.

[0023] The excitation gain at each corner position is weighted and fused with the corresponding corner position suspicion level to obtain a thermoacoustic leakage score that represents the possibility of leakage after thermoacoustic enhancement at each corner position.

[0024] The present invention is further configured such that S3 includes:

[0025] Collect environmental datasets output from a multimodal environmental sensor array. The environmental datasets include: temperature, humidity, ambient pressure, and wind speed.

[0026] The original hydrogen concentration in the cavity data is correlated with the machine's operating status, temperature, and humidity at the corresponding time. Using a compensation model based on historical interference patterns, the original hydrogen concentration under the evacuation state is corrected to the static hydrogen concentration.

[0027] The present invention is further configured to call historical data of the reference channel and compare the hydrogen concentration ratio at adjacent sampling times to obtain the internal drift factor characterizing the long-term drift of the sensor;

[0028] Based on the environmental dataset and the internal drift factor, the corner segment compensation signal after eliminating the cross-sensitivity effect is obtained by separating and calculating the component signals of different interference sources.

[0029] By combining static hydrogen concentration and corner compensation signal, corner hydrogen concentration data of corresponding time series are generated.

[0030] The present invention is further configured such that S4 includes:

[0031] The comprehensive response index is calculated based on the observed scalar values ​​of each corner segment and the valve opening degree in the machine status.

[0032] The comprehensive transfer coefficient of each corner segment is obtained by linear weighting based on the comprehensive response index, comprehensive exposure intensity and pumping energy in the intracavitary data, and the fusion concentration is calculated in combination with the corner segment path length marked in the structural design drawing;

[0033] Based on the fusion concentration, the inversion flux of each corner segment is obtained by using the gradient descent method and combining the inversion objective function with data consistency and circumferential smoothing constraints.

[0034] Based on the magnitude of the inverted flux, the leak location is determined using the maximization function.

[0035] The present invention is further configured such that S5 includes:

[0036] The risk score is calculated based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leakage score.

[0037] Risk levels are determined by threshold segmentation based on risk scores.

[0038] When the risk level is greater than the threshold, it is determined to be low risk, and the execution log is recorded.

[0039] When the risk score is less than or equal to threshold one and greater than threshold two, it is judged as medium risk, the execution log is recorded, an early warning message is generated and pushed to the cloud platform;

[0040] When the risk score is less than or equal to the threshold of 2, it is judged as high risk, triggering valve closure and shutdown operations, recording execution logs, generating alarm information, and pushing it to the cloud platform.

[0041] This invention provides a leakage monitoring method for a sealed connection structure used in hydrogen transportation. The method comprises: S1: Deploying an annular negative pressure sampling chamber, extracting gas from the chamber according to a preset pattern, collecting data from the chamber using a multimodal sensor to calculate the observed scalar, collecting machine status data to calculate the comprehensive exposure intensity, and generating corner segment suspicion by maximizing operation; S2: Applying thermal pulses and swept-frequency acoustic wave excitation to the outer wall of the annular chamber, collecting the excitation response signal, calculating the excitation gain for each corner segment, and combining the corner segment suspicion to generate a thermoacoustic leakage score; S3: Based on the data from the chamber and the collected environmental data, and after dynamic interference compensation and drift... After correction and cross-sensitivity compensation, multi-task signal fusion is performed using a lightweight long short-term memory network to obtain the time series of hydrogen concentration in the corner segment; S4: The comprehensive response index is calculated based on the observed scalar, and the path fusion concentration is calculated by combining the comprehensive exposure intensity and intracavity data. The inversion flux is calculated based on the gradient descent method and by introducing a sparse-smooth inversion objective function. The leak location is obtained by maximizing the inversion flux; S5: The risk score is calculated based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leak score. The risk score is then classified according to the threshold, and corresponding alarm information is generated. The beneficial effects include:

[0042] Achieving full-coverage leak monitoring of the circumferential range of the sealed connection structure: By evenly distributing multiple detection units around the circumference of the sealed connection structure and using a signal fusion algorithm to achieve full-coverage detection, compared to existing single-point sensors that can only monitor local areas, this effectively avoids the problem of missed detection due to blind spots. Even in complex operating conditions or with multiple potential leak locations, 360° real-time monitoring without blind spots can be achieved, significantly improving the timeliness and reliability of leak detection.

[0043] This solution offers high-precision leak location identification: By analyzing the response time difference and concentration gradient of multiple sensors, this technical solution can quickly determine the precise location of the leak in the circumferential direction of the sealed connection structure. Compared to existing methods that rely on experience to infer the leak location after manual inspection or single-point detection, this solution can output the leak coordinates within milliseconds, achieving accurate location, reducing maintenance and disassembly workload, and significantly shortening troubleshooting time.

[0044] A risk grading mechanism is introduced to assist in safety decision-making: it can not only detect leaks, but also conduct risk grading assessments by combining leak rate, concentration distribution, and environmental ventilation conditions. Compared with existing detection methods that "only report whether there is a leak" but cannot determine the degree of danger, this solution can automatically trigger different emergency response strategies based on the risk level, such as early warning, partial shutdown, or system-wide interlock shutdown, to avoid larger safety accidents caused by delayed response.

[0045] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0047] Figure 1 This is a flowchart illustrating a leakage monitoring method for a sealed connection structure for hydrogen transportation, as shown in an exemplary embodiment of the present invention. Detailed Implementation

[0048] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0049] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0050] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0051] Example:

[0052] A method for monitoring leaks in a sealed connection structure used for hydrogen transportation, such as Figure 1 As shown, it includes:

[0053] S1: Set up an annular negative pressure sampling chamber, extract gas from the chamber according to a preset pattern, use multimodal sensors to collect data from the chamber to calculate the observation scalar, collect machine status to calculate the comprehensive exposure intensity, and generate corner segment suspicion by maximizing operation.

[0054] S2: Apply thermal pulses and sweep frequency acoustic waves to the outer wall of the annular cavity, collect the excitation response signals, calculate the excitation gain of each corner segment, and generate a thermoacoustic leakage score by combining the corner segment suspicion level.

[0055] S3: Based on intracavitary data and acquired environmental data, after dynamic interference compensation, drift correction and cross-sensitivity compensation, multi-task signal fusion is performed by combining a lightweight long short-term memory network to obtain the time series of hydrogen concentration in the corner segment.

[0056] S4: Calculate the comprehensive response index based on the observation scalar, calculate the path fusion concentration by combining the comprehensive exposure intensity and intracavitary data, calculate the inversion flux based on the gradient descent method and introduce the sparse-smoothing inversion objective function, and obtain the leak location by maximizing the inversion flux.

[0057] S5: Calculate risk scores based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leakage score, classify them according to thresholds, and generate corresponding alarm information.

[0058] The present invention is further configured such that S1 includes:

[0059] An annular negative pressure sampling chamber is set up on the outside of each monitored sealed connection structure, and a multimodal sensor group is installed in several corner channels of the sampling chamber;

[0060] Gas is extracted from the sampling chamber using a micro-pump-electronically controlled valve array according to a preset spatiotemporal coding pattern. Simultaneously, data from within the chamber and machine status are collected and preprocessed using filters. The data includes: original hydrogen concentration, internal micro-pressure difference, and branch flow rate. Machine status includes: valve opening and pumping energy. Specifically, an annular negative pressure sampling chamber is fixed outside each monitored sealed connection, and multiple micro-solenoid valves are installed at equal intervals along the circumference. The micro-pump is connected to the main extraction pipe of the sampling chamber. The pump and valves are driven by a field controller, which collects position / flow feedback. A multimodal sensor group is installed in several corner channels of the sampling chamber; each corner channel includes at least a hydrogen sensing unit, a micro-pressure difference sensor, and a micro-flow sensor, and temperature, humidity, and reference sensors are arranged for synchronous environmental recording. Upon system startup, the annular negative pressure sampling chamber and the multimodal sensor group enter a standby detection state. After the sampling program begins, the micro-pump and electronically controlled valve array extract gas from the sampling chamber sequentially according to the preset time and channel order. During the extraction process of each channel, the sensor records the hydrogen concentration, the micro-pressure difference within the cavity relative to the outside, and the gas flow rate of that channel in real time, constructing intra-cavity data. This intra-cavity data includes: the initial hydrogen concentration, the intra-cavity micro-pressure difference, and the branch flow rate. Simultaneously, the control system records the valve opening ratio and pump output power at that time, constructing the machine status, which includes: valve opening degree and pumping energy. All collected data is stored chronologically and corresponds one-to-one with the channel location. After the data enters the filtering module, the system removes outliers caused by transient interference or electrical noise and smooths the data to obtain a continuous and stable measurement sequence.

[0061] The present invention is further configured to synthesize a pattern-level observation scalar based on intracavity data of the same corner segment according to a power transformation, and to perform weighted aggregation of the observation scalars of each corner segment according to the acquired valve opening degree to generate the overall response quantity;

[0062] The overall exposure is calculated by statistically analyzing the power-weighted fusion values ​​of valve opening and pumping energy under all spatiotemporally coded patterns. Specifically, the pattern-level observation scalar is a single scalar obtained by fusing the multimodal raw measurements at the lower corner of the pattern after power transformation and scale balancing. It is used to map the multidimensional raw quantities into the effective response intensity of the pattern to that corner, taking into account concentration sensitivity, dilution / enrichment effects, and negative pressure traction effects. The calculation logic of the pattern-level observation scalar is as follows: Where, Φ k,j For pattern-level observation scalars; k is the pattern index; j is the corner segment index; t k Let k be the sampling time for pattern k; This represents the original hydrogen concentration. This refers to the micro-pressure difference within the cavity; The branch flow rate is given by ε, where ε is the numerical stability constant, used to ensure numerical stability, and its value is 1 × 10⁻⁶. -6 1.2 and 1.1 are power exponents used to adjust the contribution of each channel to the observation. 1.2 indicates an emphasis on hydrogen concentration dominance and measures the amplification of the pattern-level observation scalar by the original hydrogen concentration. 1.1 reflects the penalty imposed by flow rate on the concentration dilution effect. The overall response is represented by the weighted sum of the pattern-level observation scalars of all corner segments under the pattern according to the valve opening, yielding the global intensity of the pattern, used to measure the total influence of the pattern on the circumference. The calculation logic for the overall response is as follows: Among them, G k The total response is represented by J; the total number of corner segments is represented by U. k,j The valve opening is represented by 1.8 and 1.6, which are the pattern aggregation weights. 1.8 is the power-law weight of the valve opening, and 1.6 is the power-law weight of the pattern observation variable. The overall exposure is essentially a comprehensive performance or risk quantification index. It unifies the measurement of multi-source data such as different angle segments, different valve openings, and different pumping energies to reflect the overall exposure level of the entire system under a certain time period or operating condition. The calculation logic for the overall exposure is as follows: Among them, Ξ j For overall exposure; K is the total number of patterns; Q is the overall exposure level. k To pump energy.

[0063] The invention is further configured to perform sparse reconstruction calculations based on the overall response, valve opening, and pumping energy, employing a compressed sensing iterative subspace algorithm to recover the corner segment coefficient representation, and maximizing the calculation to obtain the corner segment suspicion score representing the sparse response intensity of each corner segment. Specifically, the corner segment suspicion score is a single scalar representation of the "suspected leakage intensity" of a discrete corner segment in the circumferential direction of the annular sampling cavity. This scalar comprehensively considers: the overall response intensity under the excitation of the coded pattern, the valve opening of the corner segment under the corresponding pattern, and the pumping intensity at that time. A "sparse priority" rule is used to select the pattern that best explains the observation for that corner segment, thus forming a sparse proxy score for the corner segment. The corner segment suspicion score calculation logic is as follows: Among them, S j The degree of suspicion for the corner segment; G k U represents the total response. k,j Q represents the valve opening degree. kThe values ​​are: pumping energy; 1.0, 1.8, and 1.1 are power exponent weights, representing the amplification weight of the overall response, the amplification weight of the valve opening, and the penalty weight of the pumping energy, respectively. These are used to control the relative importance in the composite segment suspicion. The power exponent 1.0 can be omitted in mathematical calculations, but it is recommended to retain it in engineering deployment. Setting 1.0 as the default value means that if modifications are made later, only the parameters need to be adjusted without modifying the code.

[0064] The present invention is further configured such that S2 includes:

[0065] Based on the intracavity data collected by the multimodal sensor at the previous moment, the baseline energy value is calculated.

[0066] Micro heating elements and miniature ultrasonic transducers are installed on the outer wall of the annular negative pressure sampling cavity. Thermal pulse signals and swept frequency sound wave signals are applied sequentially according to a preset cycle.

[0067] Multimodal sensors are used to synchronously acquire intracavitary data during excitation, constructing a new excitation response signal. This excitation response signal includes the following after excitation: excitation hydrogen concentration, intracavitary micro-pressure difference, and excitation branch flow rate. Specifically, before execution, ensure that all multimodal sensors, valve opening readouts, pump energy meters, and excitation equipment (micro-heating pads and ultrasonic transducers) are installed, calibrated, and connected to a unified time synchronization bus. Configure excitation sequence parameters in the controller: thermal pulse amplitude and duration, ultrasonic sweep frequency range and rate, and excitation sequence timetable. Set two sampling paths in the data acquisition system: a low-speed path for hydrogen concentration / micro-pressure / flow rate, etc.; and a high-speed path for acoustic sampling. Set the sampling rate, trigger events, and file storage format. Then, before each excitation cycle begins, trigger the acquisition of a baseline window. The baseline window lasts for a preset time, recording steady-state data from all channels and saving it as a "baseline sample." Then, execute the excitation sequence in a predetermined order: first, apply a micro-heating pulse as planned, followed by or in parallel with an ultrasonic sweep frequency. Each excitation event is triggered by a synchronous trigger signal output by the controller to mark the data sampling time axis. During excitation, data from all channels is continuously acquired and divided into low-speed excitation response signals and high-speed acoustic data segments. The excitation response signals are saved for later use. Baseline energy characterizes the overall energy level of multimodal observations within the cavity before active excitation and within a preset baseline time window. Before each excitation cycle, the excitation equipment is turned off and a steady state is reached. The system then acquires and timestamps data from the concentration and differential pressure channels of each corner segment at a set sampling rate. Within the baseline window, the absolute value of the normalized concentration reading is taken and a preset concentration power exponent is applied to obtain a concentration energy proxy. The absolute value of the normalized differential pressure plus a slight positive offset is taken and a preset differential pressure power exponent is applied to obtain a differential pressure energy proxy. These two proxies are recorded sequentially and accumulated separately to form the window energy measures for the concentration and differential pressure channels. Finally, they are added together to obtain the baseline energy value for that corner segment and pattern. The baseline energy value calculation logic is as follows: Where, Θ k,j Baseline energy value; H k,j,t The original hydrogen concentration; P k,j,t H represents the intracavity micro-pressure difference; Δ represents the window length, which is physically equal to the baseline window duration multiplied by the sampling rate. The specific value is determined based on the actual site conditions during deployment. The chosen value should balance: sufficient to cover low-frequency fluctuations in the environment / sensor, yet not so long as to include irrelevant events; k,j,t The original hydrogen concentration; P k,j,t 1 represents the intracavitary micro-pressure difference; 1.2 and 1.5 are the baseline power exponents, which are the exponents used to power the concentration channel and the pressure difference channel, respectively, to control the degree of amplification / compression of the amplitude.

[0068] The present invention is further configured to calculate the response energy by performing multi-resolution wavelet analysis and high-resolution spectral estimation on the excitation response signal, combined with the baseline energy.

[0069] The excitation gain is calculated for each corner segment based on the difference between the baseline energy value and the response energy before and after excitation.

[0070] The excitation gain at each corner location is weighted and fused with the corresponding corner location suspicion level to obtain a thermoacoustic leakage score representing the possibility of leakage after thermoacoustic enhancement at each corner location. Specifically, the response energy represents the overall energy level of intracavity changes observed by the multimodal sensor within a certain corner location and a certain excitation cycle. It is used to quantify the influence of the excitation signal on the intracavity hydrogen concentration and micro-pressure difference, equivalent to a "measure of the amplitude of the excitation response"; the response energy calculation logic is as follows: Among them, E k,j In response to energy; A k This is the thermal pulse weighting coefficient, used to amplify or attenuate the weight of the concentration channel in the response energy calculation. The initial value is 1, but in actual deployment, if the concentration signal is weak or has low sensitivity to leakage, the value can be appropriately increased to 1.2 or 1.5. If the concentration signal is easily affected by noise, it can be appropriately reduced to 0.5 or 0.7; H1 k,j,t To stimulate hydrogen concentration; P1 k,j,t To excite the micro-pressure difference within the cavity; T k ε is the acoustic excitation weighting coefficient, used to amplify or attenuate the weight of the differential pressure channel in the response energy calculation. Its initial value is 1, and if the differential pressure signal significantly affects leakage characteristics, the value can be increased to 1.5. ε is the numerical stability constant, used to ensure numerical stability, and its value is 1 × 10⁻⁶. -6 ;Θ k,j The baseline energy value is used. Excitation gain represents the amplification factor or gain degree of the excitation-induced response energy relative to the baseline energy. It characterizes the degree of enhancement in the observability of the leakage by the intracavity observation signal after thermal pulse and frequency sweep acoustic excitation. Excitation gain calculation logic: Among them, Ψ j E is the excitation gain; k,j In response to energy; Θ k,j The baseline energy value is represented by k; the pattern index is represented by k; the total number of patterns is represented by K; and ε is the numerical stability constant, used to ensure numerical stability, with a value of 1 × 10⁻⁶. -6 Thermoacoustic leakage score is a quantitative indicator of the final leakage probability for each corner segment. It combines the suspicion level of sparse corner segments with the excitation gain to comprehensively consider the original leakage signal strength within the cavity and the enhancement effect of the excitation response, providing a more reliable corner segment leakage probability score. Thermoacoustic leakage score calculation logic: R j =(S j +ε) 1.5 (Ψ j )1.2 , where R j Score for thermal acoustic leakage; S j The degree of suspicion for the corner segment; Ψ j This is the excitation gain.

[0071] The present invention is further configured such that S3 includes:

[0072] Collect environmental datasets output from a multimodal environmental sensor array. The environmental datasets include: temperature, humidity, ambient pressure, and wind speed.

[0073] The raw hydrogen concentration in the cavity data is correlated with the machine's operating status, temperature, and humidity at the corresponding time. A compensation model based on historical interference patterns is used to correct the raw hydrogen concentration under evacuation conditions to a static hydrogen concentration. Specifically, environmental data is synchronously collected at a unified time point using multimodal environmental sensors and integrated into an environmental dataset, which includes temperature, humidity, ambient pressure, and wind speed. Static hydrogen concentration refers to the true and stable characterization of the hydrogen concentration within the sealed connection cavity, excluding the effects of dynamic operations such as pump evacuation, valve opening and closing causing airflow disturbances, and environmental factors. The static hydrogen concentration calculation logic is as follows: Among them, H′ j,t This represents the static hydrogen concentration; H k,j,t The original hydrogen concentration is the hydrogen concentration parameter in the intracavity data before excitation; Q k For pumping energy; U k,j τ represents the valve opening degree. t For temperature; v t Humidity; 1 is the coupling term between pumping energy and valve opening, used to describe the effect of pump and valve action on gas dilution; 2 is used for adjustment scale. This is a compensation item for pump and valve operation. This is for temperature and humidity compensation.

[0074] The present invention is further configured to call historical data of the reference channel and compare the hydrogen concentration ratio at adjacent sampling times to obtain the internal drift factor characterizing the long-term drift of the sensor;

[0075] Based on the environmental dataset and the internal drift factor, the corner segment compensation signal after eliminating the cross-sensitivity effect is obtained by separating and calculating the component signals of different interference sources.

[0076] By combining static hydrogen concentration and corner compensation signals, corner hydrogen concentration data for the corresponding time series is generated. Specifically, the internal drift factor is used to characterize the effect of slow drift on a single corner sensor during long-term operation. This drift may originate from sensor aging, temperature and humidity changes, or long-term environmental cumulative effects. It is used to correct the sensor's output at a certain moment back to the reference state, thereby improving the stability and reliability of the corner concentration data. The internal drift factor calculation logic is as follows: M j,t =M j,t-1 (1+0.02· Among them, M j,t M is the internal drift factor; j,t-1 The internal drift factor of the previous time step; Υ j,t The reference channel ratio is used to capture the relative change between the far-end blind zone sensor and the current corner segment, directly characterizing the slow drift trend; The reference hydrogen concentration is collected by a reference sensor deployed at a remote location or in a blind area. It is used as a standard value to reflect a baseline concentration unaffected by leakage, and is used for drift compensation and recursive correction. The reference hydrogen concentration is ε; ε is the numerical stability constant, used to ensure numerical stability, and its value is 1 × 10⁻⁶. -6 The corner compensation signal is a hydrogen concentration signal that, based on drift correction, further eliminates environmental interferences such as temperature, humidity, air pressure, and wind speed. It is an instantaneous correction value for the hydrogen concentration. The calculation logic for the corner compensation signal is as follows: Among them, Z j,t For corner segment compensation signal; H′ j,t τ represents the static hydrogen concentration. t For temperature; v t For humidity; π t For ambient pressure; ω t For wind speed; M j,t The internal drift factor is used. The corner hydrogen concentration data is the final concentration time series for each corner segment throughout the entire sampling period, reflecting the trend of concentration changes in the corner segment. This data undergoes drift compensation and environmental interference correction, and incorporates multiple types of signals generated by virtual sensors. Specifically, the corner compensation signal, along with environmental parameters such as temperature and ambient pressure, is input into a lightweight long short-term memory network (LSTM) model to simulate the outputs of electrochemical and thin-film sensors. Subsequently, the outputs of different virtual sensor types are summarized to obtain the corner hydrogen concentration value for a specific corner segment at a given time. Arranging the data from each time point in chronological order constitutes the complete time series of corner hydrogen concentration.

[0077] The present invention is further configured such that S4 includes:

[0078] The comprehensive response index is calculated based on the observed scalar values ​​of each corner segment and the valve opening degree in the machine status.

[0079] The comprehensive transfer coefficient of each corner segment is obtained by linear weighting based on the comprehensive response index, comprehensive exposure intensity and pumping energy in the intracavitary data, and the fusion concentration is calculated in combination with the corner segment path length marked in the structural design drawing;

[0080] Based on the fusion concentration, the inversion flux of each corner segment is obtained by using the gradient descent method and combining the inversion objective function with data consistency and circumferential smoothing constraints.

[0081] Based on the magnitude of the inversion flux, the leak location is determined according to the maximization function. Specifically, the comprehensive response index represents the overall response capability at the j-th corner segment, weighted by the observed scalar values ​​under all sampling patterns and the valve opening. It reflects the response strength of each corner segment to the overall system leak signal; a larger index indicates that the corner segment is more likely to be related to a leak event. The calculation logic of the comprehensive response index is as follows: Among them, Γ j For comprehensive response indicators; K is the total number of patterns; G k U represents the total response. k,j The valve opening is represented by the fused concentration, which is the estimated intracavity concentration after weighting by the integrated response and corner segment length. The integrated transfer coefficient is calculated by multiplying the integrated response index of the corner segment by the integrated exposure of the corner segment, and then adding the total pumping flow rate, which is obtained by summing the pumping energy of all patterns. The integrated transfer coefficient of the corner segment is then multiplied by the inversion flux of the corner segment to obtain the fused concentration of each corner segment. The inversion flux is obtained by solving an objective function, which includes the sum of squared errors between the fused concentration and the observed hydrogen concentration data of the corner segment, as well as the sum of squared differences in the inversion flux of adjacent corner segments. At the beginning of the program operation, the initial flux value is substituted into the formula to calculate the first fused concentration. Then, this fused concentration is used in the objective function, and the inversion flux is updated step by step through gradient descent to optimize iteratively until convergence, obtaining the final inversion flux of each corner segment. The initial flux value can be set to 1.0, or it can be obtained by experts in the actual deployment scenario by calculating the integrated response index × total pumping flow rate / total number of corner segments. Finally, find the corner segment with the maximum value in the inversion flux of all corner segments, and output this corner segment as the most suspicious leakage location.

[0082] The present invention is further configured such that S5 includes:

[0083] The risk score is calculated based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leakage score.

[0084] Risk levels are determined by threshold segmentation based on risk scores.

[0085] When the risk level is less than the threshold, it is judged as low risk, and the execution log is recorded.

[0086] When the risk score is greater than or equal to threshold one and less than threshold two, it is judged as medium risk, the execution log is recorded, an early warning message is generated and pushed to the cloud platform;

[0087] When the risk score is greater than or equal to threshold two, it is judged as high risk, triggering valve closure and shutdown operations, recording execution logs, generating alarm information, and pushing it to the cloud platform. Specifically, the risk score is a comprehensive quantitative indicator used to characterize the degree of leakage risk of a specific corner segment throughout the entire sampling period. It is a single scalar obtained by fusing the hydrogen concentration time series, maximum inversion flux, and thermoacoustic leakage score of the corner segment. The risk score is used to provide a quantifiable safety decision reference value, which facilitates the sorting and classification of leakage risks of different corner segments or patterns. The risk score calculation logic is as follows: the reciprocal of the hydrogen concentration time series of each corner segment is weighted and fused with the thermoacoustic leakage score and the maximum inversion flux to generate a single quantitative indicator. Because a high hydrogen concentration indicates normal hydrogen transmission in the pipeline, a low value indicates a leak, so the reciprocal should be taken. This risk score is then compared with two classification thresholds. If it is less than threshold one, it is judged as low risk; if it is between threshold one and threshold two, it is judged as medium risk and an early warning message is generated; if it is higher than threshold two, it is judged as high risk, triggering valve closure, shutdown, and alarm. The risk level, risk score, leakage location, and related log records are saved and pushed to the cloud platform. Based on subsequent sampling, the above steps are repeated to update the risk score in real time, thereby achieving continuous monitoring and security control. The default threshold is 0.4, and the default threshold is 0.7.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for monitoring leakage in a sealed connection structure used for hydrogen transportation, characterized in that, include: S1: Set up an annular negative pressure sampling chamber, extract gas from the chamber according to a preset pattern, use multimodal sensors to collect data from the chamber to calculate the observation scalar, collect machine status to calculate the comprehensive exposure intensity, and generate corner segment suspicion by maximizing operation. S2: Apply thermal pulses and sweep frequency acoustic waves to the outer wall of the annular cavity, collect the excitation response signals, calculate the excitation gain of each corner segment, and generate a thermoacoustic leakage score by combining the corner segment suspicion level. S3: Based on intracavitary data and acquired environmental data, after dynamic interference compensation, drift correction and cross-sensitivity compensation, multi-task signal fusion is performed by combining a lightweight long short-term memory network to obtain the time series of hydrogen concentration in the corner segment. S4: Calculate the comprehensive response index based on the observation scalar, calculate the path fusion concentration by combining the comprehensive exposure intensity and intracavitary data, calculate the inversion flux based on the gradient descent method and introduce the sparse-smoothing inversion objective function, and obtain the leak location by maximizing the inversion flux. S5: Calculate risk scores based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leakage score, classify them according to thresholds, and generate corresponding alarm information.

2. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 1, characterized in that, S1 includes: An annular negative pressure sampling chamber is set up on the outside of each monitored sealed connection structure, and a multimodal sensor group is installed in several corner channels of the sampling chamber; Gas is extracted from the cavity by a micro-pump-electronically controlled valve array according to a preset spatiotemporal coding pattern. The cavity data and machine status are collected simultaneously and preprocessed using a filter. The cavity data includes: original hydrogen concentration, cavity micro-pressure difference, and branch flow rate. The machine status includes: valve opening degree and pumping energy.

3. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 2, characterized in that, Based on the intracavity data of the same corner segment, a pattern-level observation scalar is synthesized by power transformation. The observation scalars of each corner segment are weighted and aggregated according to the acquired valve opening to generate the overall response. The overall exposure is calculated by statistically analyzing the power-weighted fusion values ​​of valve opening and pumping energy under all spatiotemporal coding patterns.

4. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 3, characterized in that, Sparse reconstruction calculations are performed based on the overall response, valve opening, and pumping energy. A compressed sensing iterative subspace algorithm is used to recover the corner segment coefficient representation. The corner segment suspicion degree, representing the sparse response intensity of each corner segment, is obtained by maximizing the calculation.

5. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 1, characterized in that, S2 includes: Based on the intracavity data collected by the multimodal sensor at the previous moment, the baseline energy value is calculated. Micro heating elements and miniature ultrasonic transducers are installed on the outer wall of the annular negative pressure sampling cavity. Thermal pulse signals and swept frequency sound wave signals are applied sequentially according to a preset cycle. A new excitation response signal is constructed by synchronously collecting intracavitary data using a multimodal sensor during the excitation process. The excitation response signal includes the following after excitation: excitation hydrogen concentration, intracavitary micro-pressure difference, and excitation branch flow rate.

6. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 5, characterized in that, The response energy is calculated by performing multi-resolution wavelet analysis and high-resolution spectral estimation on the excitation response signal, combined with the baseline energy. The excitation gain is calculated for each corner segment based on the difference between the baseline energy value and the response energy before and after excitation. The excitation gain at each corner position is weighted and fused with the corresponding corner position suspicion level to obtain a thermoacoustic leakage score that represents the possibility of leakage after thermoacoustic enhancement at each corner position.

7. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 1, characterized in that, S3 includes: Collect environmental datasets output from a multimodal environmental sensor array. The environmental datasets include: temperature, humidity, ambient pressure, and wind speed. The original hydrogen concentration in the cavity data is correlated with the machine's operating status, temperature, and humidity at the corresponding time. Using a compensation model based on historical interference patterns, the original hydrogen concentration under the evacuation state is corrected to the static hydrogen concentration.

8. The leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 7, characterized in that, By calling up historical data from the reference channel and comparing the hydrogen concentration ratio at adjacent sampling times, the internal drift factor, which characterizes the long-term drift of the sensor, is obtained. Based on the environmental dataset and the internal drift factor, the corner segment compensation signal after eliminating the cross-sensitivity effect is obtained by separating and calculating the component signals of different interference sources. By combining static hydrogen concentration and corner compensation signal, corner hydrogen concentration data of corresponding time series are generated.

9. A leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 2, characterized in that, S4 includes: The comprehensive response index is calculated based on the observed scalar values ​​of each corner segment and the valve opening degree in the machine status. The comprehensive transfer coefficient of each corner segment is obtained by linear weighting based on the comprehensive response index, comprehensive exposure intensity and pumping energy in the intracavitary data, and the fusion concentration is calculated in combination with the corner segment path length marked in the structural design drawing; Based on the fusion concentration, the inversion flux of each corner segment is obtained by using the gradient descent method and combining the inversion objective function with data consistency and circumferential smoothing constraints. Based on the magnitude of the inverted flux, the leak location is determined using the maximization function.

10. A leakage monitoring method for a sealed connection structure for hydrogen transportation according to claim 1, characterized in that, S5 includes: The risk score is calculated based on the time series of hydrogen concentration in the corner segment, the maximum inversion flux, and the thermoacoustic leakage score. Risk levels are determined by threshold segmentation based on risk scores. When the risk score is greater than the threshold, it is determined to be low risk, and the execution log is recorded. When the risk score is less than or equal to threshold one and greater than threshold two, it is judged as medium risk, the execution log is recorded, an early warning message is generated and pushed to the cloud platform; When the risk score is less than or equal to the threshold of 2, it is judged as high risk, triggering valve closure and shutdown operations, recording execution logs, generating alarm information, and pushing it to the cloud platform.

Images