Method for detecting the sealing of an ultra-high vacuum evacuation device

Abstract

The application provides a sealing detection method of an ultrahigh vacuum exhaust device, relates to the technical field of sealing detection, and comprises the following steps: performing leakage risk feature mining on the ultrahigh vacuum exhaust device, and constructing a detection point distribution; performing vacuum pumping smooth detection on the ultrahigh vacuum exhaust device, and generating a sealing detection starting signal according to a smooth detection result; performing sealing detection on the ultrahigh vacuum exhaust device according to the detection point distribution and a helium mass spectrometer leak detector, and obtaining a first device sealing detection result; and performing detection background interference correction, helium injection cross-influence compensation and material helium adsorption correction on the first device sealing detection result, and generating a fourth device sealing detection result. The application solves the technical problem that, due to the existence of various interference factors, misjudgment and omission are prone to occur, and the sealing detection precision is insufficient, the detection is performed after vacuum smoothing, the detection result is corrected, and the sealing detection effect of the ultrahigh vacuum exhaust device is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sealing detection, and particularly relates to a sealing detection method of an ultrahigh vacuum exhaust device. BACKGROUND

[0002] Sealing detection of an ultrahigh vacuum exhaust device is a key technology for ensuring the integrity and performance of an ultrahigh vacuum device. Since the ultrahigh vacuum device requires extremely low leakage rate, any penetration of external gas can affect the working efficiency and stability of the system. In order to detect the sealing property of the device, a helium mass spectrometer leak detector is usually used for leakage detection, by spraying helium into the inside of the device, the amount of leaked helium is detected by the helium mass spectrometer leak detector, so as to judge the sealing property of the device. However, when the helium mass spectrometer leak detector is used for sealing detection, the background gas in the external environment may interfere with the helium, affecting the accuracy of the helium mass spectrometer leak detector. In addition, the interference of the gas or the vacuum environment will cause the interference of the helium detection signal, and then produce false alarm or omission, affecting the accuracy of the sealing detection result.

[0003] In summary, in the prior art, due to the existence of various interference factors, false judgment and omission are prone to occur, resulting in the technical problem of insufficient sealing detection precision. SUMMARY

[0004] The purpose of the present application is to provide a sealing detection method of an ultrahigh vacuum exhaust device, in order to solve the technical problem of insufficient sealing detection precision in the prior art due to the existence of various interference factors, false judgment and omission are prone to occur.

[0005] In view of the above problems, the present application provides a sealing detection method of an ultrahigh vacuum exhaust device, wherein the sealing detection method of the ultrahigh vacuum exhaust device comprises: performing leakage risk feature mining on the ultrahigh vacuum exhaust device to construct a detection point distribution; performing vacuum pumping smooth detection on the ultrahigh vacuum exhaust device, and generating a sealing detection start signal according to the smooth detection result; based on the sealing detection start signal, performing sealing detection on the ultrahigh vacuum exhaust device according to the detection point distribution and a helium mass spectrometer leak detector, to obtain a first result of device sealing detection; performing detection background interference correction on the first result of device sealing detection, to obtain a second result of device sealing detection; performing helium spraying cross-influence compensation on the second result of device sealing detection, to obtain a third result of device sealing detection; performing material helium adsorption correction on the third result of device sealing detection, to generate a fourth result of device sealing detection.

[0006] Optionally, a device structure feature data set of the ultra-high vacuum exhaust device is obtained; multi-point leakage risk prediction is performed on the ultra-high vacuum exhaust device according to the device structure feature data set, a first feature point distribution greater than or equal to a predetermined leakage risk is obtained; leakage frequent feature mining is performed on the ultra-high vacuum exhaust device, a second feature point distribution is generated; the union of the first feature point distribution and the second feature point distribution is calculated, and the detection point distribution is generated.

[0007] Optionally, leakage event global retrieval is performed on the ultra-high vacuum exhaust device, and a leakage event retrieval set is obtained; leakage frequent degree evaluation is performed on a plurality of point positions of the ultra-high vacuum exhaust device according to the leakage event retrieval set, and a leakage frequent coefficient of each point position is obtained; based on the leakage frequent coefficient of each point position, the plurality of point positions are selected according to a leakage frequent threshold, and the second feature point distribution is obtained.

[0008] Optionally, vacuumizing treatment is performed on the ultra-high vacuum exhaust device according to a reference vacuum degree, and a real-time pressure data set of the ultra-high vacuum exhaust device is synchronously collected; stationarity evaluation is performed on the real-time pressure data set, and the stationarity detection result is obtained; when the real-time vacuum degree of the ultra-high vacuum exhaust device meets the reference vacuum degree and the stationarity detection result meets a predetermined stationarity, the sealing detection start signal is generated.

[0009] Optionally, when the real-time vacuum degree does not meet the reference vacuum degree and / or the stationarity detection result does not meet the predetermined stationarity, a sealing detection waiting signal is generated.

[0010] Optionally, based on the sealing detection start signal, helium injection is performed according to the detection point distribution, and helium concentration parameters are synchronously collected by the helium mass spectrometer leak detector, and helium detection data of each point position is obtained; the helium detection data of each point position is input into M sealing evaluation models, and a sealing evaluation sequence of each point position is obtained, M is a positive integer greater than 1; the sealing evaluation sequence of each point position is respectively subjected to central value calculation, a sealing coefficient of each point position is obtained, and the sealing coefficient of each point position is added to the device sealing detection first result.

[0011] Optionally, a detection background factor is set, the detection background factor includes a detection environment helium feature, a detection environment airflow feature, a detection environment other feature, and a detection equipment state feature; each point position detection background parameter collection is performed according to the detection background factor, and each point position detection background data is obtained; each detection background anomaly feature is obtained by performing anomaly identification according to the each point position detection background data; each background detection interference evaluation result is obtained by performing sealing detection interference evaluation according to the each detection background anomaly feature, and the device sealing detection second result is generated by correcting the device sealing detection first result according to the each background detection interference evaluation result.

[0012] Optionally, pre-injection helium information of each detection point is collected to obtain pre-injection helium characteristics of each point; cross-injection helium influence of each detection point is analyzed according to the pre-injection helium characteristics of each point to obtain cross-injection helium influence characteristics of each point; helium detection of each point is corrected according to the cross-injection helium influence characteristics of each point to obtain corrected detection data of each point; and the second sealing detection result of the device is corrected according to the corrected detection data of each point to generate a third sealing detection result of the device.

[0013] Optionally, a pre-injection helium characteristic sample set and a cross-injection helium influence characteristic sample set are obtained; a residual neural network is supervised trained according to the pre-injection helium characteristic sample set and the cross-injection helium influence characteristic sample set; the cross-injection helium influence analysis accuracy is obtained when the residual neural network is trained for a predetermined number of times; if the cross-injection helium influence analysis accuracy meets a predetermined analysis accuracy, a cross-injection helium influence analysis model is generated; and the pre-injection helium characteristics of each point are input into the cross-injection helium influence analysis model to obtain the cross-injection helium influence characteristics of each point.

[0014] Optionally, material characteristic data of each point is obtained; helium adsorption of each point is predicted according to the material characteristic data of each point to obtain helium adsorption characteristics of each point; sealing detection influence of each point is identified according to the helium adsorption characteristics of each point to obtain a helium adsorption correction factor of each point; and the third sealing detection result of the device is corrected according to the helium adsorption correction factor of each point to generate a fourth sealing detection result of the device.

[0015] The technical solutions provided in the present application have at least the following beneficial effects:

[0016] By mining the leakage risk characteristics of the ultra-high vacuum exhaust device, a detection point distribution is constructed; a stable detection is performed on the ultra-high vacuum exhaust device, and a sealing detection starting signal is generated according to the stable detection result; based on the sealing detection starting signal, the ultra-high vacuum exhaust device is subjected to sealing detection according to the detection point distribution and a helium mass spectrometer to obtain a first sealing detection result of the device; the first sealing detection result of the device is subjected to detection background interference correction to obtain a second sealing detection result of the device; the second sealing detection result of the device is subjected to cross-injection helium influence compensation to obtain a third sealing detection result of the device; and the third sealing detection result of the device is subjected to material helium adsorption correction to generate a fourth sealing detection result of the device. That is, by performing stable detection, it is ensured that the vacuum environment is stable before sealing detection, the background gas interference in the detection process is corrected, the influence of external gas is reduced, the helium adsorption effect of the material is corrected through cross-injection helium influence compensation, and the sealing detection effect of the ultra-high vacuum exhaust device is improved, which has high sensitivity, high robustness and self-adaptive multi-factor compensation capability.

[0017] The above description is only a summary of the technical solutions of the present application. In order to enable the technical means of the present application to be more clearly understood, and to be implemented according to the content of the description, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more apparent and easy to understand, the following specific embodiments of the present application are described. It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and other drawings can be obtained by those skilled in the art without creating any creative labor on the basis of the provided drawings.

[0019] Figure 1 The flowchart of the sealing detection method of the ultra-high vacuum exhaust device of the present application.

[0020] Figure 2 The flowchart of obtaining the third result of the sealing detection of the device in the sealing detection method of the ultra-high vacuum exhaust device of the present application. DETAILED DESCRIPTION

[0021] The present application provides a sealing detection method of an ultra-high vacuum exhaust device, which solves the technical problem of insufficient sealing detection precision due to multiple interference factors in the prior art, which easily leads to misjudgment and omission. The vacuum environment is stable before sealing detection by stable vacuum pumping detection, the background gas interference in the detection process is corrected, the influence of external gas is reduced, the helium adsorption effect of the material is corrected by helium cross-influence compensation, which has high sensitivity, high robustness and self-adaptive multi-factor compensation ability, and improves the sealing detection effect of the ultra-high vacuum exhaust device.

[0022] The technical solutions in the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creating any creative labor belong to the scope of protection of the present application. In addition, it should be noted that only parts related to the present application are shown in the drawings for convenience of description, not all.

[0023] Embodiment, please refer to the attached Figure 1The application provides a sealing detection method of an ultrahigh vacuum exhaust device, and the method specifically comprises the following steps.

[0024] S100: performing leakage risk feature mining on the ultrahigh vacuum exhaust device to construct a detection point distribution.

[0025] Further, the S100 of the application comprises:

[0026] S110: obtaining device structure feature data set of the ultrahigh vacuum exhaust device; S120: performing multi-point leakage risk prediction on the ultrahigh vacuum exhaust device according to the device structure feature data set to obtain a first feature point distribution greater than or equal to a predetermined leakage risk; S130: performing leakage frequent feature mining on the ultrahigh vacuum exhaust device to generate a second feature point distribution; S140: obtaining the union of the first feature point distribution and the second feature point distribution to generate the detection point distribution.

[0027] S131: performing leakage event global search on the ultrahigh vacuum exhaust device to obtain a leakage event search set; S132: performing leakage frequent degree evaluation on a plurality of point positions of the ultrahigh vacuum exhaust device according to the leakage event search set to obtain a leakage frequent coefficient of each point position; S133: selecting the plurality of point positions according to a leakage frequent threshold based on the leakage frequent coefficient of each point position to obtain the second feature point distribution.

[0028] Specifically, the size, material, connection mode and the like of each physical component of the ultrahigh vacuum exhaust device are obtained through design drawings and models of the ultrahigh vacuum exhaust device to obtain the device structure feature data set, which is helpful to analyze the performance and potential leakage points of each part of the device. According to the device structure feature data set, the ultrahigh vacuum exhaust device is subjected to multi-point leakage risk prediction, the structure features of each part of the ultrahigh vacuum exhaust device such as interfaces, welding positions and sealing rings are analyzed, a risk prediction model is used to evaluate the risk of each point, and the leakage probability of each point is predicted according to historical leakage data and material characteristics.

[0029] A model trained in a field closely related to ultra-high vacuum exhaust devices, especially in the prediction of leakage risks or the prediction of failures of similar equipment, such as leakage detection models of other vacuum systems, pressure vessels, is selected. The first few layers of it (the part used to extract features) are frozen, and then the last few layers of the model are retrained with the target field data (i.e. the device structure feature dataset of the ultra-high vacuum exhaust device). Through fine-tuning, the features extracted by the source field model are used on the target task while optimizing the specific part of the target task. First, the target field data (i.e. the device structure feature dataset of the ultra-high vacuum exhaust device) is standardized, cleaned, and filled, and it is divided into a training set and a validation set, usually 80% training set and 20& validation set, according to the requirements of transfer learning, the training set is input into the transfer learning model, and the validation set is used to evaluate the model trained after transfer learning, including accuracy, recall, precision and F1-score, etc. The convergence conditions of the model are set, such as the change of the validation set loss being less than 0.01 for 5 consecutive rounds or the training set accuracy reaching 95%. When the model reaches the convergence condition, the training is stopped, and the risk prediction model is obtained at this time.

[0030] The device structure feature dataset of the ultra-high vacuum exhaust device is input into the risk prediction model to predict the leakage probability of each point, that is, to perform risk analysis on multiple parts of the device. The predetermined leakage risk is a preset threshold used to determine whether the leakage risk of a certain point is high enough. Compare the leakage risk prediction values of multiple points with the predetermined leakage risk, and combine those that exceed the predetermined leakage risk to form the first feature point distribution.

[0031] Perform a global search for leakage events on the ultra-high vacuum exhaust device, that is, perform a comprehensive search and analysis of the leakage history of the entire ultra-high vacuum exhaust device to find all related leakage events. For example, the maintenance log shows that there have been 3 leakage events at the 10th joint and 2 leakage events at the 15th joint in the past year. Organize the leakage event search results into a dataset through simple statistical analysis, record the occurrence position (i.e. point) and frequency of each leakage event, and obtain the leakage event search set. Use the data in the leakage event search set to evaluate the leakage frequency coefficient of each point, calculate the number of times each point leaks, and obtain the leakage frequency of each point, i.e. the leakage frequency coefficient of each point.

[0032] A leakage frequency threshold is set to screen out point positions with high leakage frequency, so as to distinguish which point positions are frequent leakage areas and which point positions can be ignored. By screening point positions greater than or equal to the leakage frequency threshold, a final second feature point position distribution is obtained, which is an area that needs to be focused on. The union of the first feature point position distribution (point positions with high leakage risk) and the second feature point position distribution (point positions with high leakage frequency) is obtained, and a detection point position distribution is obtained, including all areas with high leakage risk or high leakage frequency, as the key positions for sealing detection. By accurately predicting and frequently mining high-risk leakage points, the detection work can be focused on key positions, important leakage points can be avoided, and high-risk and frequent leakage points can be accurately located, which significantly improves the detection efficiency and accuracy and reduces the risk of misjudgment and omission.

[0033] S200: performing stable detection on the ultra-high vacuum exhaust device, and generating a sealing detection start signal according to the stable detection result.

[0034] Further, the S200 of the present application comprises:

[0035] S210: performing vacuumizing treatment on the ultra-high vacuum exhaust device according to a reference vacuum degree, and synchronously collecting a real-time pressure data set of the ultra-high vacuum exhaust device; S220: performing stability evaluation according to the real-time pressure data set to obtain the stable detection result; S230: generating the sealing detection start signal when the real-time vacuum degree of the ultra-high vacuum exhaust device meets the reference vacuum degree and the stable detection result meets a predetermined stability; and S240: generating a sealing detection waiting signal when the real-time vacuum degree does not meet the reference vacuum degree and / or the stable detection result does not meet the predetermined stability.

[0036] Specifically, the reference vacuum degree refers to a standard vacuum degree that the ultra-high vacuum exhaust device should reach under normal working conditions, that is, the lowest expected vacuum degree set when the ultra-high vacuum exhaust device is designed, which is usually set by the equipment manufacturer or engineer according to the performance index of the equipment. The ultra-high vacuum exhaust device performs vacuumizing treatment to exhaust the gas inside until the set reference vacuum degree is reached. During the vacuumizing process, the pressure change of the ultra-high vacuum exhaust device is continuously detected using a vacuum pump, a vacuum sensor, etc., and a real-time pressure data set is collected, which records the pressure value of the equipment at each moment during the vacuumizing process. For example, during the vacuumizing process, the pressure data is recorded once per second, and the recording is continued until the reference vacuum degree is reached.

[0037] The standard deviation of the pressure fluctuation is calculated to evaluate the fluctuation degree of the pressure data, so as to obtain the smoothness detection result. The smaller the standard deviation is, the smaller the pressure fluctuation is, and the more stable the ultra-high vacuum exhaust device is. The stability evaluation refers to analyzing the real-time pressure data set to evaluate the stability of the pressure fluctuation. The evaluation criteria usually include the amplitude, frequency and trend of the pressure fluctuation, and the purpose is to judge whether the ultra-high vacuum exhaust device has reached a stable state suitable for subsequent sealing detection. For example, suppose the real-time pressure data set contains the following pressure values: P1=10 -5 Pa, P2=10 -5.1 Pa, P3=10 -5.2 Pa. The standard deviation is calculated as 1.51×10 -6 Pa, indicating that the pressure change is small.

[0038] The real-time vacuum degree must be less than or equal to the reference vacuum degree. If the real-time vacuum degree does not reach the reference vacuum degree, it means that the equipment is still in the process of vacuumizing and cannot be sealed for detection. At the same time, it is necessary to evaluate whether the pressure fluctuation is within the predetermined range, and if not, it indicates that the ultra-high vacuum exhaust device is unstable and cannot be sealed for detection.

[0039] When the real-time vacuum degree of the ultra-high vacuum exhaust device meets the reference vacuum degree, and the smoothness detection result meets the predetermined smoothness, it is considered that it has entered a stable state and is suitable for sealing detection, and a sealing detection start signal is generated. The predetermined smoothness is a predefined smoothness used to determine whether the ultra-high vacuum exhaust device has entered a stable state. If the real-time vacuum degree does not reach the reference vacuum degree, or the smoothness evaluation does not meet the predetermined standard, either of the two does not meet the requirement, a waiting signal is generated, indicating that the equipment needs to be further waited to reach an appropriate vacuum state before sealing detection. By ensuring that the real-time vacuum degree and the pressure smoothness meet the requirements, it avoids errors caused by unstable pressure or not meeting the reference vacuum degree, and ensures the accuracy of the detection result.

[0040] S300: based on the sealing detection start signal, according to the detection point distribution and the helium mass spectrometer leak detector, the ultra-high vacuum exhaust device is sealed for detection, and a device sealing detection first result is obtained.

[0041] Further, the present application S300 comprises:

[0042] S310: based on the sealing detection start signal, helium gas is sprayed according to the detection point position distribution, and the helium mass spectrometric leak detector is started to collect helium concentration parameters synchronously, to obtain helium detection data of each point position; S320: input the helium detection data of each point position into M sealing evaluation models to obtain a sealing evaluation sequence of each point position, M is a positive integer greater than 1; S330: respectively calculate the centralized value of the sealing evaluation sequence of each point position to obtain a sealing coefficient of each point position, and add the sealing coefficient of each point position to the first result of the device sealing detection.

[0043] Specifically, according to the sealing detection start signal, sealing detection is started. According to the detection point position distribution, helium gas is sprayed to each designated point position of the ultrahigh vacuum pumping device, simulating potential leakage paths, thereby detecting whether there is leakage at these point positions. Helium gas spraying refers to introducing helium gas into the ultrahigh vacuum pumping device, providing helium gas to each point position of the device through a spraying device. Helium gas is often used as a tracer gas in leak detection because of its small molecular weight and difficulty in reacting with other gases. The helium mass spectrometric leak detector is an instrument for detecting helium concentration, which can accurately measure the helium concentration inside the device. The helium mass spectrometric leak detector works synchronously at each detection point position, detects whether the sprayed helium gas leaks and records the amount of leakage, including the helium concentration value of each point position. For example, the helium concentration detection value of point position 1 is 0.08 ppm, the helium concentration detection value of point position 2 is 0.15 ppm, the helium concentration detection value of point position 3 is 0.06 ppm, the helium concentration detection value of point position 4 is 0.07 ppm, and the helium concentration detection value of point position 5 is 0.22 ppm.

[0044] The helium concentration data of each point is input into M sealability evaluation models, each model based on its specific algorithm and training data, outputting a sealability evaluation coefficient for each point, thus obtaining a sealability evaluation sequence for each point, each sealability evaluation sequence including M sealability evaluation coefficients. The M sealability evaluation models refer to inputting the helium concentration data of each point into multiple independent models, each model evaluating the sealability of the point according to different algorithms and feature analysis methods. For example, model 1 is trained according to support vector machines, model 2 is trained according to decision tree algorithms, and model 3 is trained according to neural networks. The training process of the models is similar. Historical helium concentration data and corresponding seal state labels (leak or seal) are collected, covering a variety of different operating conditions (such as different pressures, different temperatures, etc.), and containing helium leakage situations at different points. Remove missing values, outliers, and duplicate data to ensure data quality. Normalize the helium concentration of different points to train the data on the same scale. Select multiple evaluation models such as support vector machines, decision trees, neural networks, and random forests to build corresponding architectures. For each model, use the historical data set for training. For example, given the helium concentration data of a given point and the corresponding leak state label, train each model to predict the probability of leakage based on helium concentration. To prevent overfitting of the model, cross-validation is usually performed. By dividing the data set into several parts, the model is trained and validated in turn to evaluate the generalization ability of the model. Some standard evaluation indicators (such as accuracy, precision, recall, and F1 score) are used to evaluate the performance of each model. Based on the evaluation results, the hyperparameters of the model are optimized through grid search or random search, such as the kernel function parameters of SVM, the tree depth of decision tree, and the learning rate of neural network. Set the convergence conditions of the model, such as the validation set loss changing by less than 0.01 for 5 consecutive rounds or the training set accuracy reaching 95%. When the model meets the convergence condition, stop training and obtain M sealability evaluation models.

[0045] Exemplarily, the helium concentration data of points 1 to 5 is input into M sealability evaluation models, assuming there are three models, and the obtained sealability evaluation coefficients are shown in Table 1:

[0046] Table 1 Sealability evaluation coefficients

[0047] Detection point Helium concentration data (ppm) Sealing evaluation coefficient of model 1 Sealing evaluation coefficient of model 2 Sealing evaluation coefficient of model 3 Leakage or not (1 is yes, 0 is no) 1 0.08 0.89 0.90 0.88 0 2 0.15 0.31 0.32 0.30 1 3 0.06 0.91 0.93 0.92 0 4 0.07 0.88 0.92 0.91 0 5 0.22 0.23 0.25 0.24 1

[0048] The central value of each point sealability evaluation sequence is calculated, that is, for each point, the average value of the M sealability evaluation coefficients of the point is calculated to obtain the sealability coefficient of each point. Central value calculation is a process of processing multiple evaluation results to obtain a representative value, usually using the average value method for central calculation, or using the median, weighted average, etc., depending on the accuracy of the model.

[0049] The sealing coefficient of each point is added to the first result of the device sealing detection to form a device sealing detection first result. The accuracy of the ultra-high vacuum exhaust device sealing detection is further improved by multi-model evaluation and central value calculation. By combining the data of helium injection and the helium mass spectrometer with the output of the multi-model, the diversity and comprehensiveness of the detection are enhanced, and the final evaluation result is further optimized by central value calculation, thereby improving the reliability and precision of the detection.

[0050] S400: correcting the device sealing detection first result according to the detection background interference to obtain a device sealing detection second result.

[0051] Further, the S400 of the present application comprises:

[0052] S410: setting a detection background factor, the detection background factor comprising a detection environment helium characteristic, a detection environment airflow characteristic, a detection environment other characteristic, and a detection equipment state characteristic; S420: collecting a detection background parameter of each point according to the detection background factor to obtain a detection background data of each point; S430: identifying an anomaly according to the detection background data of each point to obtain a detection background anomaly characteristic; S440: evaluating a sealing detection interference according to the detection background anomaly characteristic to obtain a background detection interference evaluation result, and correcting the device sealing detection first result according to the background detection interference evaluation result to generate the device sealing detection second result.

[0053] Specifically, the detection background factor, i.e. the external environment and equipment factors affecting the sealing detection result, will interfere with the detection performance of the mass spectrometer, including a detection environment helium characteristic, a detection environment airflow characteristic, a detection environment other characteristic, and a detection equipment state characteristic. The detection environment helium characteristic is the helium concentration, gas distribution, airflow condition, etc. in the detection environment; the detection environment airflow characteristic is the air flow pattern in the environment, such as airflow speed, direction, temperature, etc., which affects the diffusion of helium and the detection result; the detection environment other characteristic includes temperature, humidity, pressure, and other environmental characteristics that affect gas distribution and detection result; the detection equipment state characteristic includes the working state of the helium mass spectrometer and other equipment, such as equipment calibration, sensor accuracy, equipment running time, etc.

[0054] According to the set detection background factor, the background parameters of each detection point are collected, that is, the background data such as helium concentration, gas flow rate, environmental temperature and humidity, and equipment state at each detection point are collected. The real-time detection data is recorded synchronously by the helium mass spectrometer leak detector, and is stored together with the background data. According to the abnormal characteristics of each detection background, the abnormal parts in the detection background data are identified, which may interfere with the sealing detection and need to be corrected. That is, the detection background data collected at each point is statistically analyzed to identify data points that are significantly different from normal conditions, and the abnormal characteristics of each detection background are obtained. For example, when the helium concentration fluctuates suddenly and greatly or the gas flow direction changes dramatically, it can be determined as abnormal.

[0055] According to the identified abnormal characteristics of each detection background, the sealing detection interference is evaluated, the influence of the interference on the sealing detection result is evaluated, and the sealing detection first result is corrected according to the evaluation. A regression model is established by historical data to predict the influence of background characteristics on sealing detection results. The historical background abnormal characteristics and the corresponding sealing detection results are collected, and the collected data set is divided into a training set and a validation set. The training set is used for model training, and the validation set is used for evaluating the performance of the model. During the training process, the model adjusts the weights and parameters by minimizing the error (such as least squares method) to fit the best relationship between the background characteristics and the detection results. The trained regression model is used to predict the abnormal characteristics of each detection background, and the interference of the abnormal characteristics of the background on the sealing detection result is evaluated. The sealing detection result of each point under different background conditions is predicted using the model, and the difference between the actual detection result and the predicted result is calculated, which is the deviation caused by the background interference.

[0056] Based on the sealing detection interference evaluation result, the sealing detection first result of the device is corrected to eliminate the influence of the background abnormality on the detection result, and a more accurate sealing detection second result of the device is obtained. According to the background interference evaluation result of each detection point, a correction factor is applied to adjust the sealing detection first result, to ensure that the sealing evaluation of each detection point is closer to the true situation, and the correction method can be a correction amount, or a more complex weighted average or dynamic adjustment, which usually depends on the accuracy requirement of the sealing detection. For example, for a detection point, the helium concentration is abnormally high, and the correction factor is 0.8 through interference evaluation, indicating that the sealing score of the point needs to be reduced (or the detection result is adjusted), and the final corrected sealing evaluation value is more accurate.

[0057] By setting the detection background factor and identifying the abnormality, the interference caused by the external environment and the equipment state is eliminated, thereby improving the accuracy and stability of the sealing detection of the ultra-high vacuum exhaust device, and overcoming the influence of the background interference on the detection result.

[0058] S500: helium sparging cross-influence compensation is performed on the device sealing detection second result to obtain a device sealing detection third result.

[0059] Further, as shown in the accompanying drawings, Figure 2 S500 of the present application comprises:

[0060] S510: pre-helium sparging information of each detection point is collected to obtain pre-helium sparging characteristics of each point; S520: according to the pre-helium sparging characteristics of each point, helium sparging cross-influence of each detection point is analyzed to obtain helium sparging cross-influence characteristics of each point; S530: according to the helium sparging cross-influence characteristics of each point, helium detection of each point is corrected to obtain corrected detection data of each point; S540: according to the corrected detection data of each point, the device sealing detection second result is corrected to generate the device sealing detection third result.

[0061] Specifically, when performing sealing detection, pre-helium sparging information is collected from each point of the ultra-high vacuum exhaust device, including but not limited to adjacent detection point helium sparging start and end time, helium sparging flow rate, helium sparging duration, relative distance between spouting space position and current point, gas flow direction and flow rate parameters. That is, during the detection process, helium is sparged at each point, and the start and end time, helium sparging flow rate (unit: L / min), helium sparging duration, spatial position distance from the spouting port to the current point, gas flow direction and flow rate of the spouting port, etc. are recorded. The helium sparging start and end time is the start and end time of helium sparging at each detection point, which is usually recorded by a timing device or a sensor synchronously at the start and end time of each point; the helium sparging flow rate is recorded by a flow meter in real time, which is usually the helium flow rate (unit: L / min); the helium sparging duration is calculated according to the start and end time; the relative distance between the spouting space position and the current point is measured by a spatial coordinate system (such as a three-dimensional coordinate system) to obtain the relative position; the gas flow direction and flow rate parameters are measured by a gas flow speed sensor or CFD (computational fluid dynamics) simulation during the helium sparging process. The collected data is processed to ensure that the pre-helium sparging information of each detection point is complete, including the start and end time, flow rate, duration, relative position, gas flow direction and flow rate of the helium sparging.

[0062] The pre-spray helium characteristics of each detection point are stored to obtain the pre-spray helium characteristics of each point. For example, the helium spray start time of point A is 10:00, the helium spray end time is 10:05, the helium spray flow rate is 3L / min, the helium spray duration is 5 minutes, the relative distance between the nozzle space position and the current point is 2m, the airflow direction is A to B, and the flow rate parameter is 0.5m / s; the helium spray start time of point B is 10:03, the helium spray end time is 10:08, the helium spray flow rate is 4L / min, the helium spray duration is 5 minutes, the relative distance between the nozzle space position and the current point is 5m, the airflow direction is A to B, and the flow rate parameter is 0.7m / s.

[0063] According to the pre-spray helium characteristics of each point, the cross-influence of helium spray on each detection point is analyzed to analyze the degree of influence of each point on the helium gas flow, and the cross-influence characteristics of each point are obtained. According to the pre-spray helium characteristics, the influence of helium spray on adjacent detection points is calculated, considering factors such as helium flow rate, airflow speed, direction and distance, to estimate the influence range and intensity of helium spray in space. Cross-influence characteristics refer to the influence data generated by these cross effects at each detection point, usually involving factors such as helium diffusion path, airflow interference, concentration change, etc. By analyzing the helium spray characteristics of each detection point, it is predicted how helium gas diffuses from one point to another, considering factors such as airflow direction, spatial position, helium flow rate, etc. According to the analysis results, the helium spray cross-influence characteristics of each point are generated, reflecting the mutual influence between points due to helium spray, such as helium interference concentration, cross-interference duration, airflow change, etc.

[0064] According to the helium spray cross-influence characteristics, the helium detection of each point is corrected to eliminate or reduce cross-interference and improve the accuracy of the detection results. That is, according to the helium spray cross-influence characteristics, the original helium detection data of each point is corrected. If a certain point is affected by the helium spray of an adjacent point, its detection data will be adjusted to eliminate this influence. For example, if point A has a strong cross-influence on point B, the correction module can reduce the helium concentration in the detection results of point B to compensate for the influence of point A helium spray on point B.

[0065] According to the corrected helium detection data of each point, the second result of the device sealing detection is corrected to eliminate the helium concentration fluctuations caused by the cross-influence of helium spray, making the detection data more accurate, and obtaining the third result of the device sealing detection. By analyzing the cross-influence of pre-spray helium and correcting the detection data, the influence of adjacent point helium spray on the detection results is reduced or eliminated, ensuring the accuracy of the measurement results. The corrected detection data is more consistent with the actual leakage situation, improving the accuracy of the sealing detection, and thus improving the reliability and safety of the ultra-high vacuum exhaust device.

[0066] Further, the present application also includes the following steps:

[0067] S521: obtaining a pre-injection helium feature sample set and a helium cross-influence feature sample set; S522: supervising training of a residual neural network according to the pre-injection helium feature sample set and the helium cross-influence feature sample set, and obtaining a helium cross-influence analysis precision each time a predetermined number of training is performed; S523: generating a helium cross-influence analysis model if the helium cross-influence analysis precision meets a predetermined analysis precision; S524: inputting the pre-injection helium features of each point into the helium cross-influence analysis model to obtain the helium cross-influence features of each point.

[0068] Specifically, sufficient sample data is required to train the residual neural network to ensure that the model can capture the cross-influence features between different points during the helium injection process. Therefore, the pre-injection helium feature sample set and the helium cross-influence feature sample set are obtained. The pre-injection helium feature sample set is collected from the pre-injection helium information of each detection point in actual testing, including information such as start and end time, flow rate, duration, airflow direction, and distance. The helium cross-influence feature sample set is obtained by simulating the helium flow and analyzing the cross-influence between different detection points. These influences are due to the diffusion and flow of helium in space, especially when the helium injection of one point crosses with the helium injection of another point, the cross-influence will change the distribution of helium in space and its diffusion process. The helium cross-influence feature sample set includes parameters such as the amplitude, duration, and spatial propagation of the cross-influence.

[0069] The pre-injection helium feature sample set and the helium cross-influence feature sample set are used as input data to train the residual neural network. According to the input feature data, the corresponding helium cross-influence features are predicted. The residual neural network is a deep neural network architecture that uses residual connections to avoid the gradient vanishing problem during training, which can make the network more efficient when processing complex tasks. Supervised training is a machine learning training method that compares the model's prediction results with the true labeled target, adjusts the model's parameters, and makes the model gradually learn the relationship between input and output.

[0070] During each training, the network calculates the loss based on the difference between the input features and the actual helium cross-influence results, updates the model weights using the backpropagation algorithm, and gradually minimizes the prediction error of the model. After a predetermined number of training (such as several epochs), the model is evaluated. The evaluation index can be mean squared error, which calculates the difference between the true value and the predicted value of the model. When the helium cross-influence analysis precision of the model meets the analysis precision, the training is stopped, and a final helium cross-influence analysis model is obtained. The analysis precision is determined by setting a target precision (such as MSE less than a certain threshold) to determine whether the desired effect is achieved.

[0071] The pre-spraying helium characteristics of each detection point are input into the trained spraying helium cross-influence analysis model, which will predict the degree of spraying helium cross-influence according to the input characteristics, such as the propagation and concentration distribution of helium between points. Through the training of pre-spraying helium characteristics and cross-influence characteristics, the model can identify and analyze the cross-influence of spraying helium between points, thereby correcting the second result of device sealing detection, making the sealing detection result more accurate.

[0072] S600: Material helium adsorption correction is performed on the third result of device sealing detection to generate a fourth result of device sealing detection.

[0073] Further, the present application S600 comprises:

[0074] S610: Obtain material characteristic data of each point; S620: Perform helium adsorption prediction according to the material characteristic data of each point to obtain helium adsorption characteristics of each point; S630: Identify sealing detection influence according to the helium adsorption characteristics of each point to obtain helium adsorption correction factors of each point; S640: Correct the third result of device sealing detection according to the helium adsorption correction factors of each point to generate the fourth result of device sealing detection.

[0075] Specifically, the material characteristic data of each point is obtained, i.e. the material information used in each detection point of the ultra-high vacuum exhaust device, including the type, density, surface structure, porosity, and adsorption capacity of the material. That is, the material characteristic data of each detection point is collected and recorded, including the chemical composition, surface roughness, specific surface area, porosity, and adsorption heat of the material. Different materials have different side effects on helium. For example, if aluminum material is used at a certain point, the higher the surface roughness of the aluminum material, the higher the probability of physical adsorption of helium molecules in the microscopic recesses or pores, resulting in a higher reading of helium concentration during detection (false positive leakage signal).

[0076] According to the material characteristic data of each point, a physical model (such as Langmuir adsorption model or BET model) is used to predict the helium adsorption amount of each point. By inputting the material characteristic data, the amount of helium that can be adsorbed on the surface of the material is calculated. The Langmuir adsorption model is a classic model that describes the adsorption behavior of gas molecules on the surface of a solid, assuming that each position on the surface where a gas molecule is adsorbed is independent and each adsorption site can only accommodate one gas molecule (single-layer adsorption). The key assumption of the Langmuir model is that the adsorption sites are uniform and have no interaction, and each adsorption site can only accommodate one molecule. The BET model is an extension of the Langmuir model and is used to describe multi-layer adsorption phenomena. It is widely used in the measurement of specific surface area (BET surface area) and explains the adsorption isotherm by considering the adsorption of multiple layers of gas molecules on the surface. The BET model is suitable for analyzing the adsorption of gas molecules in multiple layers, especially in porous materials, and can more accurately reflect the adsorption behavior of helium and other gases. For example, if a point uses a porous material with high porosity, the helium adsorption amount of this material may be high, so according to the adsorption model, it is predicted that this point may adsorb more helium during detection, resulting in incorrect detection data.

[0077] Based on the prediction results of helium adsorption, the influence of helium adsorption on the detection results of the helium mass spectrometer is analyzed. The adsorption of helium will affect the accuracy of the sealing detection. By comparing the experimental data and the predicted data, it is identified which points have significant adsorption effects that affect the detection results. For these points, corrections must be made. According to the characteristics of helium adsorption, a correction factor is derived, which is usually a proportional coefficient, used to adjust the actual detection data. The correction factor is usually calculated based on the relationship between the adsorption characteristics (such as adsorption amount, adsorption rate, etc.) and the actual detection error. For each point, a correction factor is calculated and generated according to its specific adsorption characteristics.

[0078] According to the helium adsorption correction factor of each point, the detection data of each point is corrected, usually by multiplying the device sealing detection third result by the correction factor to obtain the device sealing detection fourth result, thereby improving the accuracy of the detection results. Through helium adsorption prediction and correction, the influence of material adsorption on helium is eliminated, avoiding false positives or false negatives, and improving the accuracy of the sealing detection. According to the adsorption characteristics of different materials, the detection results of each point are individually corrected, improving the sealing detection accuracy of the ultra-high vacuum exhaust device, eliminating errors caused by material adsorption effects, and ensuring that the detection results are more reliable.

[0079] In summary, the sealing detection method for the ultra-high vacuum exhaust device provided in the present application has the following beneficial effects:

[0080] By excavating the leakage risk characteristics of the ultrahigh vacuum exhaust device, a detection point distribution is constructed; the ultrahigh vacuum exhaust device is subjected to vacuum pumping smooth detection, and a sealing detection start signal is generated according to the smooth detection result; based on the sealing detection start signal, the ultrahigh vacuum exhaust device is subjected to sealing detection according to the detection point distribution and the helium mass spectrometer leak detector, and a device sealing detection first result is obtained; the device sealing detection first result is subjected to detection background interference correction, and a device sealing detection second result is obtained; the device sealing detection second result is subjected to helium injection cross-influence compensation, and a device sealing detection third result is obtained; the device sealing detection third result is subjected to material helium adsorption correction, and a device sealing detection fourth result is generated. That is, by performing vacuum pumping smooth detection, it is ensured that the vacuum environment has been stabilized before sealing detection, the background gas interference in the detection process is corrected, the influence of external gas is reduced, the helium injection cross-influence compensation is performed, the helium adsorption effect of the material is corrected, and the sealing detection effect of the ultrahigh vacuum exhaust device is improved.

[0081] The above description of disclosed embodiments enables one of ordinary skill in the art to make or use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0082] Obviously, for those skilled in the art, some improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. A method of detecting a seal of an ultra-high vacuum exhaust device, characterized by, The method comprises the following steps: Leakage risk characteristics of an ultra-high vacuum exhaust device are excavated, and a detection point distribution is constructed; A vacuum stable detection is performed on the ultra-high vacuum exhaust device, and a sealing detection start signal is generated according to the stable detection result; Based on the sealing detection start signal, the ultra-high vacuum exhaust device is subjected to sealing detection according to the detection point distribution and a helium mass spectrometer, and a device sealing detection first result is obtained; The device sealing detection first result is subjected to detection background interference correction, and a device sealing detection second result is obtained; The device sealing detection second result is subjected to helium injection cross-influence compensation, and a device sealing detection third result is obtained; The device sealing detection third result is subjected to material helium adsorption correction, and a device sealing detection fourth result is generated; Leakage risk characteristics of an ultra-high vacuum exhaust device are excavated, and a detection point distribution is constructed, comprising: A device structure feature data set of the ultra-high vacuum exhaust device is obtained; The ultra-high vacuum exhaust device is subjected to multi-point leakage risk prediction according to the device structure feature data set, and a first feature point distribution equal to or greater than a predetermined leakage risk is obtained; Second feature point distribution is generated by leakage frequency feature excavation of the ultra-high vacuum exhaust device; The union of the first feature point distribution and the second feature point distribution is calculated, and the detection point distribution is generated; Second feature point distribution is generated by leakage frequency feature excavation of the ultra-high vacuum exhaust device, comprising: Leakage event global search is performed on the ultra-high vacuum exhaust device, and a leakage event search set is obtained; The leakage frequency of a plurality of points of the ultra-high vacuum exhaust device is evaluated according to the leakage event search set, and a leakage frequency coefficient of each point is obtained; Based on the leakage frequency coefficient of each point, the plurality of points are selected according to a leakage frequency threshold, and the second feature point distribution is obtained.

2. The method of claim 1, wherein the method is performed in an ultrahigh vacuum evacuation device. Based on the sealing detection start signal, the ultra-high vacuum exhaust device is subjected to sealing detection according to the detection point distribution and a helium mass spectrometer, and a device sealing detection first result is obtained, comprising: Based on the sealing detection start signal, helium injection is performed according to the detection point distribution, and the helium mass spectrometer is started synchronously to collect helium concentration parameters, and helium detection data of each point is obtained; The helium detection data of each point is input into M sealing evaluation models, and a sealing evaluation sequence of each point is obtained, M being a positive integer greater than 1; The sealing evaluation sequence of each point is subjected to central value calculation respectively, and a sealing coefficient of each point is obtained, and the sealing coefficient of each point is added to the device sealing detection first result.

3. The method of claim 1, wherein the method is used for detecting a leak in an ultra-high vacuum exhaust apparatus. The device sealing detection first result is subjected to detection background interference correction, and a device sealing detection second result is obtained, comprising: A detection background factor is set, the detection background factor comprising a detection environment helium feature, a detection environment airflow feature, a detection environment other feature, and a detection equipment state feature; Each point detection background parameter is collected according to the detection background factor, and each point detection background data is obtained; Each detection background anomaly feature is obtained by performing anomaly identification according to the each point detection background data; According to the characteristics of each detection background anomaly, a sealing detection interference evaluation is performed to obtain a background detection interference evaluation result, and the first sealing detection result of the device is corrected according to the background detection interference evaluation result to generate a second sealing detection result of the device.

4. The method of detecting a leak in an ultrahigh vacuum exhaust apparatus according to Claim 1, wherein The second sealing detection result of the device is subjected to helium sparging cross-influence compensation to obtain a third sealing detection result of the device, including: Collecting pre-helium sparging information of each detection point to obtain pre-helium sparging characteristics of each point; According to the pre-helium sparging characteristics of each point, the helium sparging cross-influence of each detection point is analyzed to obtain helium sparging cross-influence characteristics of each point; According to the helium sparging cross-influence characteristics of each point, a helium detection correction is performed on each point to obtain corrected detection data of each point; According to the corrected detection data of each point, the second sealing detection result of the device is corrected to generate the third sealing detection result of the device.

5. The method of claim 4, wherein the method is performed in a chamber of an ultra-high vacuum evacuation device. According to the pre-helium sparging characteristics of each point, the helium sparging cross-influence of each detection point is analyzed to obtain helium sparging cross-influence characteristics of each point, including: Obtaining a pre-helium sparging characteristic sample set and a helium sparging cross-influence characteristic sample set; According to the pre-helium sparging characteristic sample set and the helium sparging cross-influence characteristic sample set, a residual neural network is supervised trained, and the helium sparging cross-influence analysis accuracy is obtained every predetermined number of times of training; If the helium sparging cross-influence analysis accuracy meets the predetermined analysis accuracy, a helium sparging cross-influence analysis model is generated; The pre-helium sparging characteristics of each point are input into the helium sparging cross-influence analysis model to obtain the helium sparging cross-influence characteristics of each point.

6. The method of detecting a leak in an ultrahigh vacuum exhaust apparatus according to Claim 1, wherein The third sealing detection result of the device is subjected to material helium adsorption correction to generate a fourth sealing detection result of the device, including: Obtaining material characteristic data of each point; According to the material characteristic data of each point, helium adsorption prediction is performed to obtain helium adsorption characteristics of each point; According to the helium adsorption characteristics of each point, sealing detection influence identification is performed to obtain helium adsorption correction factors of each point; According to the helium adsorption correction factors of each point, the third sealing detection result of the device is corrected to generate the fourth sealing detection result of the device.

7. The method of detecting a leak in an ultrahigh vacuum exhaust apparatus according to Claim 1, wherein The ultra-high vacuum exhaust device is subjected to vacuum pumping stability detection, and a sealing detection start signal is generated according to the stability detection result, including: According to the reference vacuum degree, the ultra-high vacuum exhaust device is subjected to vacuum pumping treatment, and real-time pressure data sets of the ultra-high vacuum exhaust device are synchronously collected; According to the real-time pressure data sets, a stability evaluation is performed to obtain the stability detection result; When the real-time vacuum degree of the ultra-high vacuum exhaust device meets the reference vacuum degree and the stability detection result meets the predetermined stability, the sealing detection start signal is generated.

8. The method of claim 7, wherein the method is performed in an ultrahigh vacuum evacuation device. When the real-time vacuum degree does not meet the reference vacuum degree and / or the stability detection result does not meet the predetermined stability, a sealing detection waiting signal is generated.

Images