A helium mass spectrometer leak detection method and system using interception sampling method

Abstract

The invention discloses a helium mass spectrometer leak detection method and system using an interception sampling method, belonging to the technical field of leak detection. The leak detection method has the following specific steps: initializing a helium mass spectrometer and intercepting and sampling; preprocessing and classifying data; establishing a virtual model; optimizing the model and parameters; predicting leakage trends; verifying results; analyzing data and determining the location and severity of the leakage. The invention can improve the adaptability and detection accuracy of the detection system in a complex environment, thereby effectively reducing the false alarm rate and the missed alarm rate, enhancing the comprehensiveness and reliability of the system, reducing the need for human intervention, improving the continuity and stability of leak detection, making the detection system more intelligent and autonomous, being able to accurately identify different leakage modes from complex helium mass spectrometer detection data, improving the credibility of the detection system, helping operators to take preventive measures in a timely manner, being able to quickly respond to the occurrence and changes of leakage events, and improving the real-time and sensitivity of the detection system.

Description

Technical Field

[0001] The present invention relates to the technical field of leakage detection, and in particular to a helium mass spectrometry leak detection method and system using an intercept sampling method. Background Art

[0002] With the development of industrial technology and the improvement of safety awareness, gas leak detection technology plays an increasingly important role in real-time monitoring and prevention of environmental pollution. In industrial and laboratory environments, gas leakage is a common but serious safety hazard. Especially for closed systems or containers, such as chemical plants, pipeline systems or laboratory facilities, leakage may not only cause environmental pollution, but also endanger the health and safety of workers. Therefore, it is crucial to detect and accurately locate leaks as soon as possible. The helium mass spectrometry leak detection method has become one of the commonly used leak detection technologies due to its high sensitivity and accuracy. The traditional helium mass spectrometry leak detection method usually involves introducing helium into the system under test and using a mass spectrometer to detect changes in helium concentration in the external environment of the system to locate the leak. However, a single detection method may not meet the actual needs under different environmental conditions, so it is necessary to combine multiple technologies and algorithms to improve the efficiency and accuracy of detection.

[0003] The existing helium mass spectrometry leak detection method and system have low adaptability and detection accuracy in complex environments, which increases the false alarm rate and missed alarm rate of the system, and requires more human intervention, which reduces the continuity and stability of leak detection; in addition, the existing helium mass spectrometry leak detection method and system cannot identify different leakage patterns from complex helium mass spectrometry detection data, which reduces the credibility of the detection system and cannot help operators take preventive measures in time; for this reason, we propose a helium mass spectrometry leak detection method and system using interception sampling method. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a helium mass spectrometry leak detection method and system using an interception sampling method.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A helium mass spectrometer leak detection method using an intercept sampling method, the specific steps of the leak detection method are as follows:

[0007] S1, initializing the helium mass spectrometer settings and performing intercept sampling;

[0008] S2, preprocessing and classifying the mass spectrometry data;

[0009] S3. Establish a virtual model of the container or equipment under test to simulate leakage;

[0010] S4, optimizing the virtual model and iteratively updating the detection parameters;

[0011] S5. Analyze and predict multiple detection data and identify potential leakage patterns and trends;

[0012] S6. Verify the virtual model and actual test results;

[0013] S7. Analyze and classify all test data and determine the location and severity of the leak.

[0014] As a further solution of the present invention, the specific steps of interception sampling described in S1 are as follows:

[0015] Step 1: Clean the surface of the container or device to be tested to ensure that there are no contaminants that interfere with the test results. Then place the container or device to be tested in a stable environment to avoid the influence of external airflow and temperature changes on the test.

[0016] Step 2: Seal all interfaces of the container or equipment that do not need to be tested, install the sampling tube on the interface of the helium mass spectrometer, and perform pressure test or vacuum test to preliminarily confirm the sealing of all interfaces;

[0017] Step 3: Use a helium bottle to fill helium into the container or device under test through the sampling tube, and control the flow and pressure of the helium to make the helium evenly distributed in the system. Use the helium concentration sensor to monitor the helium concentration in the container under test in real time;

[0018] Step 4: Select one or more sampling points according to the structure of the container or equipment to be tested and the requirements for leak detection, install the sampling tube at the selected sampling point, ensure that the connection between the sampling tube and the container or equipment to be tested is completely sealed, start the vacuum pump, extract a preset amount of gas sample through the sampling tube, and introduce the gas sample obtained by interception sampling into the sample chamber of the helium mass spectrometer;

[0019] Step 5: Start the helium mass spectrometer, detect the introduced gas sample, measure the concentration of helium, and record the detection data of the helium mass spectrometer. If the helium concentration is detected to exceed the leakage threshold, it is preliminarily determined that there is a leak at the sampling point.

[0020] As a further solution of the present invention, the specific steps of mass spectrum data classification and identification in S2 are as follows:

[0021] Step 1: Extract key features related to the leak from the preprocessed data, including helium concentration changes, pressure changes, time series features, frequency analysis features, spatial distribution features, leak duration and environmental conditions, and decentralize and standardize each feature;

[0022] Step 2: Count the number of feature data after preprocessing, calculate the covariance between each feature, and construct a set of n×n covariance matrices , for the covariance matrix Perform eigenvalue decomposition and obtain the corresponding eigenvector and eigenvalues , for each set of eigenvalues ​​obtained Sort by largest to smallest and select the first feature vectors, where is the number of principal components required;

[0023] Step 3: Construct a projection matrix through the selected eigenvectors, project the original data set into the new feature space, and calculate the samples in the new feature space based on the original data set and the projection matrix. Then, annotate the existing historical data and experimental data to determine the data in normal state and the data in leaked state.

[0024] Step 4: Divide the labeled data set into a training set and a test set. Use the sample features and labels in the training set to calculate the distance between each sample through the classification model, store the calculated distance in the model, and use the data samples in the test set to verify the performance of the trained classification model. Then select the optimal K value through cross-validation.

[0025] Step 5: Calculate the Euclidean distance between each current mass spectrometry data and all samples in the training set. According to the calculated distance, sort the samples in the training set from small to large distance, select multiple neighbors with the smallest distance, vote according to their labels, and select the label with the highest frequency as the classification result of the mass spectrometry data;

[0026] Step 6: Based on the classification results and the information of the sampling points, determine the specific leakage location, record the classification results, sample characteristics, classification labels, and confidence levels, compare the classification results with the actual situation, analyze the classification errors and the causes of the errors, and then regularly update the training and test sets, and use new test data and results to continuously train and optimize the classification model.

[0027] As a further solution of the present invention, the specific calculation formula of the covariance in step 2 is as follows:

[0028] ;

[0029] In the formula, as well as Respectively represent Samples in the feature and Features The value of as well as Respectively represent characteristics and Features The mean of Represents the total number of samples in the dataset.

[0030] As a further solution of the present invention, the specific steps of constructing the virtual model in S3 are as follows:

[0031] Step ①: Collect the pre-processed mass spectrometry data and preliminary test results, build a geometric model using the modeling software ANSYS according to the actual geometric structure of the container or equipment being tested, and simplify the geometric model according to actual needs;

[0032] Step ②: Manually set the corresponding material properties, boundary conditions and environmental conditions for the model, then divide the geometric model into a limited number of grid cells, adjust the density and quality of the grid division, refine the grid in the key area, and set the calculation parameters of the time step and convergence criterion;

[0033] Step 3: Use modeling software to perform numerical simulation calculations to simulate the flow and diffusion process of helium in the container or equipment being tested, monitor the calculation process in real time, display the calculation results in the form of graphics or animations, analyze the calculation results in detail, extract key physical quantities and characteristics, and compare and verify the calculation results with the preliminary test data;

[0034] Step ④: Analyze the error between the calculation results and the preliminary test data. According to the results of the error analysis, adjust the model parameters and recalculate the adjusted model until the calculation results match the test data. Confirm the leakage location and type through the optimized virtual model. According to the calculation results of the virtual model, formulate specific leak detection and repair plans, and record the results of the model calculation and the analysis process.

[0035] As a further solution of the present invention, the specific steps of virtual model optimization in S4 are as follows:

[0036] Step I: Randomly generate an initial set of antibodies , represents different combinations of model parameters, where Represents the antibody population, Representative Antibodies, Represents the number of antibodies, and then the fitness of each antibody is evaluated based on the difference between the calculation results of the virtual model and the preliminary test data;

[0037] Step II: Select antibodies with high fitness according to the fitness value for cloning, clone the selected antibodies, generate multiple similar antibodies, and construct the corresponding cloned antibody set. ,in, Represents the antibody subpopulation after cloning, Representative The first clones, represents the number of clones;

[0038] Step III: Basis The clones are mutated to simulate the mutation process in the biological immune system, where represents the variation strength, Represents a random number from a standard normal distribution, and then recalculates the fitness value of the mutated clone to evaluate the effect of the mutation operation;

[0039] Step IV: Select clones with fitness values ​​higher than the selection threshold as new antibody populations based on the fitness values, and preset a similarity threshold , calculate each antibody Similarity to other antibodies in the population , and filter out to construct antibodies Overly similar antibody collections;

[0040] Step V: If the size of the similar antibody set exceeds the preset length, the antibodies therein are suppressed, the one with the highest fitness is retained, and the similar antibodies are deleted or the fitness of the similar antibodies is reduced. After deleting or reducing the fitness of the similar antibodies, the antibody population is updated;

[0041] Step VI: If the number of iterations reaches the preset value, the antibody with the highest fitness in the current population is output as the optimal solution, the optimal parameter combination is applied to the virtual model, the model parameters are updated, the numerical simulation calculation is re-performed using the optimized virtual model, and the result is compared and verified with the actual test data.

[0042] A helium mass spectrometer leak detection system using an intercept sampling method, comprising a helium mass spectrometer, a sensor, an acquisition and processing module, an extraction and detection module, a virtual modeling module, a numerical simulation module, an optimization and verification module, an analysis and prediction module, a simulation and decision module, a real-time monitoring module, a storage management module, and a report generation module;

[0043] The helium mass spectrometer is used to monitor and record the changes in helium concentration in real time;

[0044] The sensor is used to measure various key data of the container or equipment;

[0045] The acquisition and processing module is used to collect raw data from the helium mass spectrometer and each sensor, and clean the raw data;

[0046] The extraction and detection module is used to extract key features from the preprocessed data, classify and identify the feature data, and preliminarily determine the leakage location and type;

[0047] The virtual modeling module constructs a geometric model according to the actual geometric structure of the container or device under test;

[0048] The numerical simulation module is used to perform numerical simulation on the model and calculate the flow and diffusion process of helium in the container or device under test;

[0049] The optimization verification module is used to optimize the virtual model parameters, and apply the optimized model parameters to the virtual model to re-perform numerical simulation;

[0050] The analysis and prediction module is used to analyze and predict multiple detection data to identify potential leakage patterns and trends;

[0051] The simulation decision module is used to evaluate the effectiveness of different detection strategies and continuously adjust the detection strategies;

[0052] The real-time monitoring module is used to monitor various key data in real time, and when an abnormality or leakage is detected, the alarm system is triggered to notify the staff to take measures;

[0053] The storage management module is used to store the collected and processed data in a database;

[0054] The report generation module is used to automatically generate a test report based on the test results and numerical simulation data.

[0055] As a further solution of the present invention, the specific steps of identifying potential leakage patterns and trends of the analysis and prediction module are as follows:

[0056] Step 1: Collect the leakage status of the container or equipment under test and build a leakage status set , where each state Represents a leak state or a no-leak state, and collects helium concentration, pressure, and temperature data from multiple tests to construct a time series observation data set , where each observation Represents a test result;

[0057] Step 2: Collect data based on leak status With observation data set Construct the initial state transition probability matrix, observation probability matrix and initial state probability vector, use the Baum-Welch algorithm to calculate the forward probability and backward probability respectively, and update the HMM model parameters according to the calculation results;

[0058] Step 3: Use Viterbi algorithm to decode, initialized at time 1 state And observed The maximum probability of , and then recursively calculate to obtain the final maximum probability and the corresponding state sequence, and then backtrack to determine the state path with leakage to obtain the corresponding hidden state sequence, where Represents time Status is And observed The maximum probability of Represents the state Transfer to The probability of Represents in state Observed below The probability of

[0059] Step 4: Identify different leakage patterns through the decoded latent state sequence, and analyze the changing trend of leakage based on the latent state sequence to predict future leakage behavior. Display the identified patterns and trends through charts and synchronize them to the virtual model for demonstration.

[0060] As a further solution of the present invention, the specific steps of adjusting the detection strategy of the simulation decision module are as follows:

[0061] Step 1: Build a set of empty decision trees, collect all detection strategies, and each detection strategy includes detection order, detection frequency, and detection point location detection information, and correspond each strategy to a node or a path in the decision tree;

[0062] Step 2: Add the initial state of the start of detection as the root node to the decision tree. Its state includes the initial configuration and parameters of the detection system. Starting from the root node, calculate the UCB1 value of each child node using the UCB1 formula, select the child node with the highest UCB1 value, and select the strategy path along the tree downward;

[0063] Step 3: When a leaf node is detected, stop selecting and check whether the leaf node is not fully expanded. If so, introduce a new detection strategy or detection step and add it to the decision tree as a new child node to expand the new child node. Start from the expanded new node, perform random simulation, apply the current strategy to perform helium mass spectrometry detection, and generate detection results and calculate cumulative rewards during the simulation. If the leaf node has been fully expanded, perform random simulation directly;

[0064] Step 4: Trace back the cumulative rewards of the simulation results to all the nodes passed, update the number of visits and cumulative rewards of these nodes, and adjust the selection strategy based on the cumulative rewards and number of visits to optimize the detection path.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. The intercept sampling helium mass spectrometry leak detection method randomly generates a set of initial antibody sets, representing different model parameter combinations, and then evaluates the fitness of each antibody based on the difference between the calculation results of the virtual model and the preliminary detection data, selects antibodies with high fitness for cloning based on the fitness value, clones the selected antibodies, generates multiple similar antibodies, and constructs a corresponding clone antibody set, mutates each clone antibody to simulate the mutation process in the biological immune system, and then recalculates the fitness value of the mutated clones, evaluates the effect of the mutation operation, and selects clones with fitness higher than the selection threshold as new antibodies based on the fitness value. The antibody population is suppressed and updated. If the number of iterations reaches the preset value, the antibody with the highest fitness in the current population is output as the optimal solution. The optimal parameter combination is applied to the virtual model, the model parameters are updated, and the numerical simulation calculation is performed again using the optimized virtual model. The numerical simulation calculation is then compared and verified with the actual detection data. This can improve the adaptability and detection accuracy of the detection system in complex environments, thereby effectively reducing the false alarm rate and missed alarm rate, enhancing the comprehensiveness and reliability of the system, reducing the need for human intervention, improving the continuity and stability of leak detection, and making the detection system more intelligent and autonomous.

[0067] 2. The present invention collects various leakage states of the container or equipment under test, constructs a leakage state set, collects helium concentration, pressure and temperature data from multiple tests, constructs a time series observation data set, constructs an initial state transition probability matrix, an observation probability matrix and an initial state probability vector according to the leakage state set and the observation data set, uses the Baum-Welch algorithm to perform forward probability calculation and backward probability calculation respectively, and updates the HMM model parameters according to the calculation results, uses the Viterbi algorithm to decode, and continuously performs recursive calculations, and then obtains the final maximum probability and the corresponding state sequence, backtracks to determine the state path where the leakage exists, so as to obtain the corresponding hidden state sequence, identifies different leakage modes through the decoded hidden state sequence, and analyzes the leakage change trend according to the hidden state sequence, predicts future leakage behavior, displays the identified modes and trends through charts, and synchronizes them to the virtual model for demonstration, can accurately identify different leakage modes from complex helium mass spectrometry detection data, improve the credibility of the detection system, help operators to take preventive measures in time, can quickly respond to the occurrence and changes of leakage events, and improve the real-time and sensitivity of the detection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0069] Figure 1A flowchart of a helium mass spectrometer leak detection method using an intercept sampling method proposed by the present invention;

[0070] Figure 2 The present invention provides a system block diagram of a helium mass spectrometer leak detection system using an interception sampling method. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Example 1

[0072] Reference Figure 1 This embodiment discloses a helium mass spectrometer leak detection method using interception sampling method. The specific steps of the leak detection method are as follows:

[0073] Initialize the helium mass spectrometer setup and perform shutoff sampling.

[0074] Specifically, clean the surface of the container or device to be tested to ensure that there are no pollutants that interfere with the test results. Then place the container or device to be tested in a stable environment to avoid the influence of external airflow and temperature changes on the test. Seal all interfaces of the container or device that do not need to be tested, and install the sampling tube on the interface of the helium mass spectrometer. At the same time, perform a pressure test or a vacuum test to preliminarily confirm the sealing of all interfaces. Use a helium bottle to fill helium into the container or device to be tested through the sampling tube. At the same time, control the flow and pressure of the helium to make the helium evenly distributed in the system. The helium concentration sensor is used to measure the helium concentration in real time. Monitor the helium concentration in the container to be tested. According to the structure of the container or equipment to be tested and the requirements of leak detection, select one or more sampling points, install the sampling tube at the selected sampling point, ensure that the connection between the sampling tube and the container or equipment to be tested is completely sealed, start the vacuum pump, extract a preset amount of gas sample through the sampling tube, and introduce the gas sample obtained by interception sampling into the sample chamber of the helium mass spectrometer, start the helium mass spectrometer, detect the introduced gas sample, measure the helium concentration, and record the detection data of the helium mass spectrometer. If the helium concentration is detected to exceed the leakage threshold, it is preliminarily determined that there is a leak at the sampling point.

[0075] Preprocess and classify mass spectrometry data.

[0076] Specifically, extract the key features related to the leak from the preprocessed data, where the key features include helium concentration changes, pressure changes, time series features, frequency analysis features, spatial distribution features, leakage duration and environmental conditions. Decentralize and standardize each feature, count the number of feature data after preprocessing, calculate the covariance between the features, and construct a set of n×n covariance matrices. , for the covariance matrix Perform eigenvalue decomposition and obtain the corresponding eigenvector and eigenvalues , for each set of eigenvalues ​​obtained Sort by largest to smallest and select the first feature vectors, where is the number of principal components required. The projection matrix is ​​constructed by the selected eigenvectors, and the original data set is projected into the new feature space. The samples in the new feature space are obtained by calculating the original data set and the projection matrix. The existing historical data and experimental data are then labeled, and the data in the normal state and the data in the leakage state are determined. The labeled data set is divided into a training set and a test set. The sample features and labels in the training set are used to calculate the distance between each sample through the classification model, and the calculated distance is stored in the model. The data samples in the test set are used to verify the performance of the trained classification model. After that, the optimal K value is selected through cross-validation. Calculate the Euclidean distance between each current mass spectrometry data and all samples in the training set. According to the calculated distance, sort the samples in the training set from small to large distance, select multiple neighbors with the smallest distance, vote according to their labels, and select the label with the highest frequency as the classification result of the mass spectrometry data. According to the classification result and the information of the sampling point, determine the specific leakage location, record the classification result, sample characteristics, classification label and confidence, compare the classification result with the actual situation, analyze the classification error and the cause of the error, and then regularly update the training set and test set, and then use the new detection data and results to continuously train and optimize the classification model.

[0077] It should be further explained that the specific calculation formula of covariance is as follows:

[0078] ;

[0079] In the formula, as well as Respectively represent Samples in the feature and Features The value of as well as Represents characteristics and Features The mean of Represents the total number of samples in the dataset.

[0080] Create a virtual model of the container or equipment under test to simulate leakage.

[0081] Specifically, the pre-processed mass spectrometry data and preliminary test results are collected, and a geometric model is constructed using the modeling software ANSYS according to the actual geometric structure of the container or device under test. The geometric model is simplified according to actual needs, and the corresponding material properties, boundary conditions and environmental conditions are manually set for the model. The geometric model is then divided into a limited number of grid units, the density and quality of the grid division are adjusted, and the grid is refined in key areas. The time step and convergence criterion calculation parameters are set, and numerical simulation calculations are performed using the modeling software to simulate the flow and diffusion process of helium in the container or device under test. At the same time, the calculation process is monitored in real time, and the calculation results are displayed in the form of graphics or animations. The calculation results are analyzed in detail, key physical quantities and features are extracted, and the calculation results are compared and verified with the preliminary test data, and the errors between the calculation results and the preliminary test data are analyzed. According to the results of the error analysis, the various parameters of the model are adjusted, and the adjusted model is recalculated until the calculation results are consistent with the test data. The leak location and type are confirmed by the optimized virtual model, and a specific leak detection plan and repair plan are formulated according to the calculation results of the virtual model, and the results of the model calculation and the analysis process are recorded.

[0082] Optimize the virtual model and iteratively update the detection parameters.

[0083] Specifically, a set of initial antibody sets is randomly generated , represents different combinations of model parameters, where Represents the antibody population, Representative Antibodies, Represents the number of antibodies. Then, the fitness of each antibody is evaluated based on the difference between the calculation results of the virtual model and the preliminary test data. According to the fitness value, antibodies with high fitness are selected for cloning. The selected antibodies are cloned to generate multiple similar antibodies to construct the corresponding cloned antibody set. ,in, Represents the antibody subpopulation after cloning, Representative The first clones, Represents the number of clones, based on The clones are mutated to simulate the mutation process in the biological immune system, where represents the variation strength, Represents a random number of standard normal distribution, and then recalculates the fitness value of the mutated clones to evaluate the effect of the mutation operation. According to the fitness value, clones with fitness higher than the selection threshold are selected as the new antibody population, and a similarity threshold is preset. , calculate each antibody Similarity to other antibodies in the population , and filter out to construct antibodies with For an overly similar antibody set, if the size of the similar antibody set exceeds the preset length, the antibodies therein are suppressed, the one with the highest fitness is retained, and similar antibodies are deleted or the fitness of similar antibodies is reduced. After deleting or reducing the fitness of similar antibodies, the antibody population is updated. If the number of iterations reaches the preset value, the antibody with the highest fitness in the current population is output as the optimal solution. The obtained optimal parameter combination is applied to the virtual model, the model parameters are updated, and the numerical simulation calculation is re-performed using the optimized virtual model, which is then compared and verified with the actual test data.

[0084] Analyze and predict multiple inspection data and identify potential leakage patterns and trends.

[0085] Verify the virtual model and actual test results.

[0086] All inspection data is analyzed and classified to determine the leak location and severity. Example 2

[0087] Reference Figure 2 The present embodiment discloses a helium mass spectrometry leak detection system using an interception sampling method, including a helium mass spectrometer, a sensor, an acquisition and processing module, an extraction and detection module, a virtual modeling module, a numerical simulation module, an optimization and verification module, an analysis and prediction module, a simulation and decision-making module, a real-time monitoring module, a storage management module, and a report generation module.

[0088] The helium mass spectrometer is used to monitor and record the changes in helium concentration in real time; the sensor is used to measure the key data of the container or equipment being tested.

[0089] In this embodiment, the sensor specifically includes a pressure sensor, a temperature sensor, a gas sensor, a flow sensor, a humidity sensor, a displacement sensor, an acceleration sensor and an acoustic sensor.

[0090] The acquisition and processing module is used to collect raw data from the helium mass spectrometer and various sensors, and clean the raw data; the extraction and detection module is used to extract key features from the preprocessed data, classify and identify the feature data, and preliminarily determine the leakage location and type.

[0091] The virtual modeling module constructs a geometric model according to the actual geometric structure of the container or device under test; the numerical simulation module is used to perform numerical simulation on the model and calculate the flow and diffusion process of helium in the container or device under test.

[0092] The optimization and verification module is used to optimize the virtual model parameters, apply the optimized model parameters to the virtual model, and re-perform numerical simulation; the analysis and prediction module is used to analyze and predict multiple detection data to identify potential leakage patterns and trends.

[0093] Specifically, collect the leakage status of the container or equipment under test and construct a leakage status set , where each state Represents a leak state or a no-leak state, and collects helium concentration, pressure, and temperature data from multiple tests to construct a time series observation data set , where each observation Represents a test result, based on the leakage status set With observation data set Construct the initial state transition probability matrix, observation probability matrix and initial state probability vector, use the Baum-Welch algorithm to calculate the forward probability and the backward probability respectively, and update the HMM model parameters according to the calculation results, use the Viterbi algorithm to decode, and initialize the state at time 1 to And observed The maximum probability of , and then recursively calculate to obtain the final maximum probability and the corresponding state sequence, and then backtrack to determine the state path with leakage to obtain the corresponding hidden state sequence, where Represents time Status is And observed The maximum probability of Represents the state Transfer to The probability of Represents in state Observed below The probability of leakage is obtained by decoding the hidden state sequence, identifying different leakage patterns, and analyzing the leakage change trend based on the hidden state sequence, predicting future leakage behavior, displaying the identified patterns and trends through charts, and synchronizing them to the virtual model for demonstration.

[0094] The simulation decision module is used to evaluate the effectiveness of different detection strategies and continuously adjust the detection strategies.

[0095] Specifically, a set of empty decision trees are constructed to collect all detection strategies, and each detection strategy includes detection order, detection frequency and detection point location detection information. Each strategy corresponds to a node or a path in the decision tree, and the initial state of the start of detection is added to the decision tree as the root node, and its state includes the initial configuration and parameters of the detection system. Starting from the root node, the UCB1 value of each child node is calculated by the UCB1 formula, and the child node with the highest UCB1 value is selected, and the strategy path is selected along the tree downward. When a leaf node is detected, the selection is stopped, and it is detected whether the leaf node is not fully expanded. If so, a new detection strategy or detection step is introduced and added to the decision tree as a new child node to expand the new child node, and random simulation is performed starting from the expanded new node, and the current strategy is applied to perform helium mass spectrometry detection. During the simulation, the detection results are generated and the cumulative rewards are calculated. If the leaf node has been fully expanded, random simulation is performed directly, and the cumulative rewards of the simulation results are traced back to all the nodes passed, and the number of visits and cumulative rewards of these nodes are updated. According to the cumulative rewards and the number of visits, the selection strategy is adjusted to optimize the detection path.

[0096] The real-time monitoring module is used to monitor various key data in real time, and when an abnormality or leakage is detected, it triggers the alarm system to notify the staff to take measures; the storage management module is used to store the collected and processed data in the database; the report generation module is used to automatically generate a test report based on the test results and numerical simulation data.

Claims

1. A helium mass spectrometer leak detection method using intercept sampling method, characterized in that: The specific steps of the leak detection method are as follows: S1, initializing the helium mass spectrometer settings and performing intercept sampling; S2, preprocessing and classifying the mass spectrometry data; S3. Establish a virtual model of the container or equipment under test to simulate leakage; S4, optimizing the virtual model and iteratively updating the detection parameters; S5. Analyze and predict multiple detection data and identify potential leakage patterns and trends; S6. Verify the virtual model and the actual test results; S7. Analyze and classify all test data and determine the leak location and severity; The specific steps of mass spectrum data classification and identification in S2 are as follows: Step 1: Extract key features related to the leak from the preprocessed data, including helium concentration changes, pressure changes, time series features, frequency analysis features, spatial distribution features, leak duration and environmental conditions, and decentralize and standardize each feature; Step 2: Count the number of feature data after preprocessing, calculate the covariance between each feature, and construct a set of n×n covariance matrices , for the covariance matrix Perform eigenvalue decomposition and obtain the corresponding eigenvector and eigenvalues , for each set of eigenvalues ​​obtained Sort by largest to smallest and select the first feature vectors, where is the number of principal components required; Step 3: Construct a projection matrix through the selected eigenvectors, project the original data set into the new feature space, and calculate the samples in the new feature space based on the original data set and the projection matrix. Then, annotate the existing historical data and experimental data to determine the data in normal state and the data in leaked state. Step 4: Divide the labeled data set into a training set and a test set. Use the sample features and labels in the training set to calculate the distance between each sample through the classification model, store the calculated distance in the model, and use the data samples in the test set to verify the performance of the trained classification model. Then select the optimal K value through cross-validation. Step 5: Calculate the Euclidean distance between each current mass spectrometry data and all samples in the training set. According to the calculated distance, sort the samples in the training set from small to large distance, select multiple neighbors with the smallest distance, vote according to their labels, and select the label with the highest frequency as the classification result of the mass spectrometry data; Step 6: According to the classification results and the information of the sampling points, determine the specific leakage location, record the classification results, sample characteristics, classification labels and confidence levels, compare the classification results with the actual situation, analyze the classification errors and the causes of the errors, and then regularly update the training set and test set, and then use the new test data and results to continuously train and optimize the classification model; The specific steps of constructing the virtual model described in S3 are as follows: Step ①: Collect the pre-processed mass spectrometry data and preliminary test results, build a geometric model using the modeling software ANSYS according to the actual geometric structure of the container or equipment being tested, and simplify the geometric model according to actual needs; Step ②: Manually set the corresponding material properties, boundary conditions and environmental conditions for the model, then divide the geometric model into a limited number of grid cells, adjust the density and quality of the grid division, refine the grid in the key area, and set the calculation parameters of the time step and convergence criterion; Step 3: Use modeling software to perform numerical simulation calculations to simulate the flow and diffusion process of helium in the container or equipment being tested, monitor the calculation process in real time, display the calculation results in the form of graphics or animations, analyze the calculation results in detail, extract key physical quantities and characteristics, and compare and verify the calculation results with the preliminary test data; Step ④: Analyze the error between the calculation results and the preliminary test data. According to the results of the error analysis, adjust the parameters of the model and recalculate the adjusted model until the calculation results match the test data. Confirm the leakage location and type through the optimized virtual model. According to the calculation results of the virtual model, formulate a specific leakage detection plan and repair plan, and record the results of the model calculation and the analysis process. The specific steps of virtual model optimization described in S4 are as follows: Step I: Randomly generate an initial set of antibodies , represents different combinations of model parameters, where Represents the antibody population, Representative Antibodies, Represents the number of antibodies, and then the fitness of each antibody is evaluated based on the difference between the calculation results of the virtual model and the preliminary test data; Step II: Select antibodies with high fitness according to the fitness value for cloning, clone the selected antibodies, generate multiple similar antibodies, and construct the corresponding cloned antibody set. ,in, Represents the antibody subpopulation after cloning, Representative The first clones, represents the number of clones; Step III: Basis The clones are mutated to simulate the mutation process in the biological immune system, where represents the variation strength, Represents a random number from a standard normal distribution, and then recalculates the fitness value of the mutated clone to evaluate the effect of the mutation operation; Step IV: Select clones with fitness values ​​higher than the selection threshold as new antibody populations based on the fitness values, and preset a similarity threshold , calculate each antibody Similarity to other antibodies in the population , and filter out to construct antibodies Overly similar antibody collections; Step V: If the size of the similar antibody set exceeds the preset length, the antibodies therein are suppressed, the one with the highest fitness is retained, and the similar antibodies are deleted or the fitness of the similar antibodies is reduced. After deleting or reducing the fitness of the similar antibodies, the antibody population is updated; Step VI: If the number of iterations reaches the preset value, the antibody with the highest fitness in the current population is output as the optimal solution, the optimal parameter combination is applied to the virtual model, the model parameters are updated, the numerical simulation calculation is re-performed using the optimized virtual model, and the result is compared and verified with the actual test data.

2. The helium mass spectrometry leak detection method using interception sampling method according to claim 1, characterized in that: The specific steps of interception sampling described in S1 are as follows: Step 1: Clean the surface of the container or device to be tested to ensure that there are no contaminants that interfere with the test results. Then place the container or device to be tested in a stable environment to avoid the influence of external airflow and temperature changes on the test. Step 2: Seal all interfaces of the container or equipment that do not need to be tested, install the sampling tube on the interface of the helium mass spectrometer, and perform pressure test or vacuum test to preliminarily confirm the sealing of all interfaces; Step 3: Use a helium bottle to fill helium into the container or device under test through the sampling tube, and control the flow and pressure of the helium to make the helium evenly distributed in the system. Use the helium concentration sensor to monitor the helium concentration in the container under test in real time; Step 4: Select one or more sampling points according to the structure of the container or equipment to be tested and the requirements for leak detection, install the sampling tube at the selected sampling point, ensure that the connection between the sampling tube and the container or equipment to be tested is completely sealed, start the vacuum pump, extract a preset amount of gas sample through the sampling tube, and introduce the gas sample obtained by interception sampling into the sample chamber of the helium mass spectrometer; Step 5: Start the helium mass spectrometer, detect the introduced gas sample, measure the concentration of helium, and record the detection data of the helium mass spectrometer. If the helium concentration is detected to exceed the leakage threshold, it is preliminarily determined that there is a leak at the sampling point.

3. The helium mass spectrometer leak detection method using interception sampling method according to claim 1, characterized in that: The specific calculation formula of the covariance in step 2 is as follows: ； In the formula, as well as Respectively represent Samples in the feature and Features The value of as well as Respectively represent characteristics and Features The mean of Represents the total number of samples in the dataset.

4. A helium mass spectrometer leak detection system using intercept sampling method, used to implement the method described in any one of claims 1 to 3, characterized in that: It includes a helium mass spectrometer, a sensor, an acquisition and processing module, an extraction and detection module, a virtual modeling module, a numerical simulation module, an optimization and verification module, an analysis and prediction module, a simulation and decision-making module, a real-time monitoring module, a storage management module, and a report generation module; The helium mass spectrometer is used to monitor and record the changes in helium concentration in real time; The sensor is used to measure various key data of the container or equipment; The acquisition and processing module is used to collect raw data from the helium mass spectrometer and each sensor, and clean the raw data; The extraction and detection module is used to extract key features from the preprocessed data, classify and identify the feature data, and preliminarily determine the leakage location and type; The virtual modeling module constructs a geometric model according to the actual geometric structure of the container or device under test; The numerical simulation module is used to perform numerical simulation on the model and calculate the flow and diffusion process of helium in the container or device under test; The optimization verification module is used to optimize the virtual model parameters, and apply the optimized model parameters to the virtual model to re-perform numerical simulation; The analysis and prediction module is used to analyze and predict multiple detection data to identify potential leakage patterns and trends; The simulation decision module is used to evaluate the effectiveness of different detection strategies and continuously adjust the detection strategies; The real-time monitoring module is used to monitor various key data in real time, and when an abnormality or leakage is detected, the alarm system is triggered to notify the staff to take measures; The storage management module is used to store the collected and processed data in a database; The report generation module is used to automatically generate a test report based on the test results and numerical simulation data.

5. The helium mass spectrometer leak detection system using interception sampling method according to claim 4, characterized in that: The specific steps of potential leakage mode and trend identification of the analysis and prediction module are as follows: Step 1: Collect the leakage status of the container or equipment under test and build a leakage status set , where each state Represents a leak state or a no-leak state, and collects helium concentration, pressure, and temperature data from multiple tests to construct a time series observation data set , where each observation Represents a test result; Step 2: Collect data based on leak status With observation data set Construct the initial state transition probability matrix, observation probability matrix and initial state probability vector, use the Baum-Welch algorithm to calculate the forward probability and backward probability respectively, and update the HMM model parameters according to the calculation results; Step 3: Use Viterbi algorithm to decode, initialized at time 1 state And observed The maximum probability of , and then recursively calculate to obtain the final maximum probability and the corresponding state sequence, and then backtrack to determine the state path with leakage to obtain the corresponding hidden state sequence, where Represents time Status is And observed The maximum probability of Represents the state Transfer to The probability of Represents in state Observed below probability; Step 4: Identify different leakage patterns through the decoded latent state sequence, and analyze the changing trend of leakage based on the latent state sequence to predict future leakage behavior. Display the identified patterns and trends through charts and synchronize them to the virtual model for demonstration.

6. The helium mass spectrometer leak detection system using interception sampling method according to claim 4, characterized in that: The specific steps of adjusting the detection strategy of the simulation decision module are as follows: Step 1: Build a set of empty decision trees, collect all detection strategies, and each detection strategy includes detection order, detection frequency, and detection point location detection information, and correspond each strategy to a node or a path in the decision tree; Step 2: Add the initial state of the start of detection as the root node to the decision tree. Its state includes the initial configuration and parameters of the detection system. Starting from the root node, calculate the UCB1 value of each child node using the UCB1 formula, select the child node with the highest UCB1 value, and select the strategy path along the tree downward; Step 3: When a leaf node is detected, stop selecting and check whether the leaf node is not fully expanded. If so, introduce a new detection strategy or detection step and add it to the decision tree as a new child node to expand the new child node. Start from the expanded new node, perform random simulation, apply the current strategy to perform helium mass spectrometry detection, and generate detection results and calculate cumulative rewards during the simulation. If the leaf node has been fully expanded, perform random simulation directly; Step 4: Trace back the cumulative rewards of the simulation results to all the nodes passed, update the number of visits and cumulative rewards of these nodes, and adjust the selection strategy based on the cumulative rewards and number of visits to optimize the detection path.

Images