Vacuum valve airtightness detection method and device

Abstract

The present application belongs to the technical field of valve air tightness detection, and provides a vacuum valve air tightness detection method and device, including the following steps: collecting event records of multiple times of positive pressure-negative pressure bidirectional detection on the vacuum valve from a historical detection database, and performing cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups; obtaining detection conditions of a real-time bidirectional detection event, obtaining a matched detection condition similar event group as a target analysis group through a matching algorithm, screening events with consistent detection sequences from the target analysis group, and respectively constructing a first detection sequence event set and a second detection sequence event set; respectively performing deviation analysis on the detection leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set to obtain a first leakage rate deviation sequence and a second leakage rate deviation sequence.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of valve air tightness detection, and particularly relates to a vacuum valve air tightness detection method and device. BACKGROUND

[0002] A vacuum valve is a device used to control the flow of gas in a vacuum system, which can work in a vacuum environment and adjust the pressure, flow rate and direction of gas. The vacuum valve is a key component in many industrial and scientific applications, such as semiconductor manufacturing, vacuum coating, aerospace, etc. The leakage of the vacuum valve will directly affect the performance of the vacuum system, such as vacuum degree, pumping rate, etc. Through air tightness detection, it can ensure that the vacuum valve can maintain good sealing performance under normal working conditions, thereby ensuring the overall performance of the system.

[0003] The vacuum valve needs to prevent both internal and external pressure difference leakage under some working conditions, i.e. it has a bidirectional sealing function. Therefore, during air tightness detection, it cannot only detect one-way leakage, but also needs to be bidirectionally detected to ensure that the vacuum valve can be effectively sealed under both positive and negative pressure conditions.

[0004] When bidirectional detection is performed, the sequence of bidirectional detection includes: first positive pressure and then negative pressure, or first negative pressure and then positive pressure. Because the stress relaxation or residual stress of the vacuum valve body or sealing element under different pressure directions may cause different detection results, the selection of different bidirectional detection sequences may affect the detection results. If the detection sequence is not reasonable, it may cause deviation in the detection results, or even lead to incorrect conclusions. For example, first positive pressure detection may cause deformation of the sealing element, which in turn affects the results of negative pressure detection.

[0005] Therefore, the application provides a vacuum valve air tightness detection method and device. SUMMARY

[0006] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background.

[0007] The technical solution adopted by the application to solve the technical problems is: a vacuum valve air tightness detection method, comprising the following steps:

[0008] Collecting event records of multiple times of positive pressure-negative pressure bidirectional detection of the vacuum valve from a historical detection database, and performing cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups;

[0009] Obtaining the detection conditions of the real-time bidirectional detection event, obtaining the matched detection condition similar event group as the target analysis group through a matching algorithm, and screening events with consistent detection sequences from the target analysis group to construct a first detection sequence event set and a second detection sequence event set, respectively.

[0010] respectively, and the first leakage rate deviation sequence and the second leakage rate deviation sequence are quantitatively evaluated, and the optimal detection sequence for the real-time bidirectional detection event is selected;

[0011] Based on the selected optimal detection sequence and the corresponding leakage rate deviation sequence, a leakage rate deviation compensation mechanism is set to correct the leakage rate in the real-time bidirectional detection event and optimize the accuracy of detection.

[0012] As a further scheme of the application, the process of the detection condition similar event group is:

[0013] The data obtained by the positive pressure-negative pressure bidirectional detection of the vacuum valve is cleaned and processed, and the K-Means++ clustering algorithm is used for clustering analysis of events with similar detection conditions, which is used to group the bidirectional detection events according to the detection conditions to form the detection condition similar event group.

[0014] As a further scheme of the application, the process of the target analysis group is:

[0015] The detection conditions of the real-time bidirectional detection event of the vacuum valve air tightness are obtained, and the detection condition data of the real-time bidirectional detection event is converted into a real-time event feature vector;

[0016] The Euclidean distance between the real-time event feature vector and all candidate cluster centers is calculated respectively, and the cluster corresponding to the center with the smallest Euclidean distance is extracted as the target analysis group.

[0017] As a further scheme of the application, the process of the candidate cluster center is:

[0018] The cosine similarity between the real-time event feature vector and all historical detection event feature vectors is calculated respectively, and the historical detection events with a cosine similarity greater than a similarity threshold are extracted, and the similar event group cluster centers to which they belong are obtained as candidate cluster centers.

[0019] As a further scheme of the application, the process of the first leakage rate deviation sequence and the second leakage rate deviation sequence is:

[0020] The detection leakage rate and the actual leakage rate of all events in the first detection sequence event set are obtained, the detection leakage rate and the actual leakage rate are calculated by difference, the leakage rate deviation is obtained, and the first leakage rate deviation sequence is constructed;

[0021] The detection leakage rate and the actual leakage rate of all events in the second detection sequence event set are obtained, the detection leakage rate and the actual leakage rate are calculated by difference, the leakage rate deviation is obtained, and the second leakage rate deviation sequence is constructed.

[0022] As a further scheme of the present application, the process of the optimal detection sequence is:

[0023] Extract events in the first leakage rate deviation sequence within the preset leakage rate deviation allowable range, and count the number of events, and then calculate the ratio with the total number of events in the first detection sequence event set to obtain the first detection event qualified rate;

[0024] Extract events in the second leakage rate deviation sequence within the preset leakage rate deviation allowable range, and count the number of events, and then calculate the ratio with the total number of events in the second detection sequence event set to obtain the second detection event qualified rate;

[0025] If the first detection event qualified rate is greater than the second detection event qualified rate, the first sequence is selected as the optimal detection sequence of the real-time bidirectional detection event;

[0026] If the first detection event qualified rate is less than the second detection event qualified rate, the second sequence is selected as the optimal detection sequence of the real-time bidirectional detection event.

[0027] As a further scheme of the present application, the leakage rate deviation compensation mechanism comprises:

[0028] Obtain the selected optimal detection sequence, and obtain the leakage rate deviation sequence corresponding to the optimal detection sequence as a compensation analysis sequence;

[0029] Calculate the mean of the leakage rate deviation in the compensation analysis sequence, and determine the adjustment direction based on the positive and negative of the mean. If the mean is positive, the detected leakage rate needs to be reduced by the mean during real-time detection. If it is negative, the mean is added.

[0030] As a further scheme of the present application, the leakage rate deviation compensation mechanism further comprises:

[0031] Calculate the coefficient of variation of the leakage rate deviation in the compensation analysis sequence. If the coefficient of variation is less than the coefficient of variation limit, use the mean of the leakage rate deviation in the compensation analysis sequence as the compensation value of the real-time bidirectional detection event;

[0032] If the coefficient of variation is greater than or equal to the coefficient of variation limit, a dynamic compensation model is constructed based on the key influence characteristics;

[0033] The real-time value of the key influence characteristics in the real-time bidirectional detection event is input into the dynamic compensation model, and the leakage rate deviation prediction value of the real-time detection event is output as the compensation value.

[0034] As a further scheme of the present application, the key influence characteristics are:

[0035] Screening detection condition features with high correlation with leakage rate deviation from historical bidirectional detection event data corresponding to the compensation analysis sequence;

[0036] Using Pearson correlation coefficient to judge correlation, extracting features with Pearson correlation coefficient greater than correlation limit value as key influence features.

[0037] A vacuum valve airtightness detection device comprises:

[0038] A historical detection event analysis module: collects event records of multiple times of positive pressure-negative pressure bidirectional detection of vacuum valves from a historical detection database, and performs cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups;

[0039] A real-time detection event matching module: obtains the detection conditions of a real-time bidirectional detection event, obtains a matched detection condition similar event group as a target analysis group through a matching algorithm, and screens events with consistent detection sequences from the target analysis group to construct a first detection sequence event set and a second detection sequence event set, respectively;

[0040] An optimal detection sequence selection module: performs deviation analysis on the leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set, respectively, obtains a first leakage rate deviation sequence and a second leakage rate deviation sequence, performs quantitative evaluation on the first leakage rate deviation sequence and the second leakage rate deviation sequence, and selects an optimal detection sequence for the real-time bidirectional detection event;

[0041] A detection deviation compensation module: based on the selected optimal detection sequence and the corresponding leakage rate deviation sequence, sets a leakage rate deviation compensation mechanism to correct the leakage rate in the real-time bidirectional detection event and optimize the accuracy of detection.

[0042] The beneficial effects of the present application are as follows:

[0043] The present application uses K-Means++ algorithm to cluster historical detection events by conditions to form similar event groups, so that real-time detection can quickly match the corresponding scene, and through the two-layer matching algorithm of cosine similarity and Euclidean distance, the target analysis group is matched with the current detection condition, laying a reliable data foundation for subsequent detection sequence selection and deviation compensation, effectively reducing the detection deviation caused by historical data matching error;

[0044] The application selects the optimal detection sequence by calculating the leakage rate deviation sequence of the first and second detection sequence event sets, taking the preset allowable range as the basis for statistical qualified rate, thereby reducing detection errors caused by improper sequence from the source. In addition, the sequential detection method is introduced for the case of insufficient number of event sets, and the rationality of sequence selection is further guaranteed through dynamic adjustment, which improves the accuracy of the detection result, optimizes the detection efficiency, and reduces the resource waste caused by blind attempts.

[0045] The application judges the data stability according to the coefficient of variation of the deviation sequence, adopts mean compensation when the coefficient of variation is small, and when the coefficient of variation is large, the key influence features are screened through the Pearson correlation coefficient, a dynamic compensation model is constructed by using the random forest model, and the comprehensive influence of multiple factors on the deviation is captured, which can correct the detection result more in line with the actual detection error. BRIEF DESCRIPTION OF DRAWINGS

[0046] The application will be further described below with reference to the accompanying drawings.

[0047] Figure 1 is a step flow chart of a vacuum valve air tightness detection method of the application;

[0048] Figure 2 is a target analysis group acquisition step flow chart in a vacuum valve air tightness detection method of the application;

[0049] Figure 3 is a vacuum valve air tightness detection device architecture diagram of the application. DETAILED DESCRIPTION

[0050] In order to make the technical means, creative features, purposes and effects realized by the application easy to understand, the application will be further described below with reference to the specific embodiments.

[0051] Embodiment 1:

[0052] Please refer to Figure 1 and Figure 2 , a vacuum valve air tightness detection method according to the embodiments of the application includes the following steps:

[0053] Step 1: Collect event records of multiple times of positive pressure-negative pressure bidirectional detection of vacuum valves from the historical detection database, and perform cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups;

[0054] In some embodiments, events of positive pressure-negative pressure bidirectional detection of vacuum valves are collected from historical detection data;

[0055] The positive pressure-negative pressure bidirectional detection event at least includes: detection time stamp, vacuum valve model, material, sealing type, service life; positive pressure detection pressure value, negative pressure detection pressure value; detection environment temperature, humidity, atmospheric pressure; detection sequence, positive pressure first or negative pressure first; detection leakage rate, actual leakage rate;

[0056] The obtained data is cleaned, including: filling missing values; outlier processing; feature standardization;

[0057] Events with similar detection conditions are clustered and analyzed, which are used to group bidirectional detection events according to detection conditions to form groups of events with similar detection conditions;

[0058] The K-Means++ algorithm is selected, including: randomly selecting a sample as the first centroid from the data set; calculating the shortest distance between each sample and the selected centroid, and constructing a probability distribution based on the square of the distance; randomly selecting the next centroid according to the probability distribution, ensuring that the distance between the centroids is as dispersed as possible; until K centroids are selected;

[0059] The distance can be calculated based on the Euclidean distance formula;

[0060] The distance between each sample and all centroids is calculated, and it is assigned to the nearest cluster; for each cluster, the mean of all samples within it is calculated as a new centroid; until the centroid position changes less than a preset threshold or reaches the maximum number of iterations;

[0061] The silhouette coefficient under different K values is calculated, and the K value corresponding to the maximum coefficient is selected;

[0062] The feature center of each cluster is calculated to identify groups of events with similar detection conditions;

[0063] For example, a conductor factory needs to perform clustering analysis on 1000 historical vacuum valve detection events, and the detection events contain the following features;

[0064] Valve properties: model, material, sealing type, service life; detection parameters: positive / negative pressure detection pressure, detection sequence; environmental conditions: temperature, humidity, atmospheric pressure; detection results: detection leakage rate, actual leakage rate;

[0065] For example, A type, metal, rubber, 2 years; positive pressure 0.1 MPa, negative pressure -0.09 MPa, positive pressure first, then negative pressure; 25°C, 60% RH, 101.3 kPa; 、 ;

[0066] Randomly select one sample as the first centroid (e.g., event ID = 100); calculate the Euclidean distance of other samples from the centroid, and select the next centroid according to the probability distribution (e.g., event ID = 500); after 10 iterations, if the centroid position changes less than a preset threshold, convergence is achieved; the contour coefficient method verifies that the coefficient is 0.72 when K = 4, confirming that the grouping is reasonable;

[0067] Cluster feature analysis: Cluster 1 (metal valve body + rubber seal + high temperature environment):

[0068] Characteristic center: positive pressure 0.12 MPa, negative pressure -0.09 MPa, temperature 35℃, service life 1.5 years;

[0069] Typical event: 85% use the "positive pressure first, then negative pressure" sequence, with a mean leakage rate deviation of -5%;

[0070] Cluster 2 (ceramic valve body + PTFE seal + low temperature environment):

[0071] Characteristic center: positive pressure 0.08 MPa, negative pressure -0.07 MPa, temperature 10℃, service life 4 years;

[0072] Typical event: 90% use the "negative pressure first, then positive pressure" sequence, with a mean leakage rate deviation of +3%;

[0073] This step clusters historical bidirectional detection events based on detection conditions, thereby providing a historical reference sample library for subsequent real-time detection of vacuum valve airtightness, enabling quick matching of similar detection scenarios during real-time bidirectional detection events, providing a solid data foundation for optimal detection order selection and leakage rate deviation compensation, enhancing the accuracy of vacuum valve airtightness detection, and also providing scientific basis and decision support for detection process optimization and quality control;

[0074] Step two: obtain the detection conditions of the real-time bidirectional detection event, match the detection conditions to obtain a similar event group as the target analysis group, and select events with consistent detection orders from the target analysis group to construct a first detection order event set and a second detection order event set;

[0075] In some embodiments, the detection conditions of the real-time bidirectional detection event of the vacuum valve airtightness are obtained;

[0076] For example, the vacuum valve model is A type, the material is metal, the seal type is rubber, the service life is 1 year, the detection pressure is positive pressure 0.11 MPa and negative pressure -0.08 MPa, and the environmental temperature is 30℃;

[0077] The feature matching algorithm is used to identify a similar event group matching the detection condition of the real-time bidirectional detection event, including two layers of matching of data feature similarity of the detection condition and Euclidean distance from the clustering centroid of the similar event group;

[0078] The detection condition of the real-time event is subjected to data cleaning processing, and the detection condition data of the real-time bidirectional detection event is converted into a real-time event feature vector;

[0079] For example, the numerical features are: positive pressure 0.11 MPa (0.1 after standardization), and ambient temperature 30°C (0.5 after standardization); the classification features are: A type valve body -> [1, 0, 0], metal -> [1, 0], and rubber sealing element -> [1, 0, 0]; and the combined vector is: [0.1, 0.5, 1, 0, 0, 1, 0, 1, 0, 0];

[0080] The cosine similarity of the real-time event feature vector and all historical detection event feature vectors is calculated respectively, and the historical detection events with a cosine similarity greater than a similarity threshold value are extracted, and the clustering centroid of the similar event group to which the historical detection events belong is obtained as a candidate clustering centroid;

[0081] The similarity threshold value is set by a person skilled in the art based on experience, and is generally set to 0.8, so that the pre-screened event and the real-time event are highly consistent in key features;

[0082] For example, if the real-time event is "metal valve body + rubber sealing element", only the events in the historical events that are also "metal valve body + rubber sealing element" combination are retained after pre-screening, and irrelevant events such as "ceramic + PTFE" are excluded;

[0083] The Euclidean distance between the real-time event feature vector and all candidate clustering centroids is calculated respectively, and the cluster corresponding to the centroid with the smallest Euclidean distance is extracted as the target analysis group;

[0084] Through the two layers of matching process, the matching accuracy of the real-time bidirectional detection event is improved, and a reliable data basis is provided for subsequent detection sequence selection and dynamic compensation;

[0085] It is worth noting that when the centroid with the smallest Euclidean distance is extracted, a minimum threshold value is also set, and the minimum threshold value is determined by a person skilled in the art according to the historical data distribution, and can be set to 1.5 times the average distance within the cluster;

[0086] If the minimum Euclidean distance is less than the minimum threshold value, the matching is valid, and if the minimum Euclidean distance is greater than or equal to the minimum threshold value, the matching is invalid; if the matching is invalid, it means that the similarity between the real-time bidirectional detection event and the historical detection event group is low, and there is no suitable target analysis group, and a new bidirectional detection event is formed after detection;

[0087] Based on the matching, a target analysis group is obtained;

[0088] All events in the target analysis group with a bidirectional detection sequence of positive pressure first and negative pressure second are extracted to construct a first detection sequence event set;

[0089] All events in the target analysis group with a bidirectional detection sequence of negative pressure first and positive pressure second are extracted to construct a second detection sequence event set;

[0090] This step matches the real-time bidirectional detection events with corresponding similar event groups through two layers of matching mechanisms, thereby increasing the accuracy of matching, splitting the matched similar event groups according to the detection sequence, providing a basis for comparative analysis of two sequences for real-time detection, reducing blind selection, and enabling the first and second detection sequence event sets constructed based on the target analysis group to focus the subsequent deviation analysis on the historical data that are truly related to the current detection condition, thereby improving the analysis accuracy;

[0091] Step three: deviation analysis is performed on the detection leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set, respectively, to obtain a first leakage rate deviation sequence and a second leakage rate deviation sequence, and quantitative evaluation is performed on the first leakage rate deviation sequence and the second leakage rate deviation sequence to select the best detection sequence for the real-time bidirectional detection events;

[0092] In some embodiments, the detection leakage rate and the actual leakage rate of all events in the first detection sequence event set are obtained, the detection leakage rate and the actual leakage rate are subjected to difference calculation to obtain a leakage rate deviation, and a first leakage rate deviation sequence is constructed;

[0093] The actual leakage rate is measured by a high-sensitivity detection technology recognized by the industry, for example, a helium mass spectrometry leak detection method;

[0094] A preset leakage rate deviation allowable range is set, events in the first leakage rate deviation sequence that are within the preset leakage rate deviation allowable range are extracted, and the number of the events is counted, and then a ratio calculation is performed on the number of the events and the total number of events in the first detection sequence event set to obtain a first detection event qualification rate;

[0095] The detection leakage rate and the actual leakage rate of all events in the second detection sequence event set are obtained, the detection leakage rate and the actual leakage rate are subjected to difference calculation to obtain a leakage rate deviation, and a second leakage rate deviation sequence is constructed;

[0096] Events in the second leakage rate deviation sequence that are within the preset leakage rate deviation allowable range are extracted, and the number of the events is counted, and then a ratio calculation is performed on the number of the events and the total number of events in the second detection sequence event set to obtain a second detection event qualification rate;

[0097] The first detection event qualification rate and the second detection event qualification rate are compared;

[0098] If the first detection event qualified rate is greater than the second detection event qualified rate, the first order is selected as the best detection order of the real-time bidirectional detection event;

[0099] If the first detection event qualified rate is less than the second detection event qualified rate, the second order is selected as the best detection order of the real-time bidirectional detection event;

[0100] If the first detection event qualified rate is equal to the second detection event qualified rate, other dimensions are compared, including but not limited to: leakage rate deviation mean, leakage rate deviation extreme value;

[0101] It is worth noting that when the number of events in a certain detection order event set is too small, the calculation of its detection qualified rate is not statistically meaningful. For example, the first detection order event set has 100 events, and the second detection order event set has only 10 events. If there are a few abnormal data (such as extremely high leakage rate deviation caused by measurement error) in the second detection order event set, the qualified rate may be greatly reduced. Conversely, if the 10 event data performs well, the qualified rate will be artificially high. Therefore, a threshold of the number of events is set. When the number of events in a certain event set is less than the threshold of the number of events, the best detection order is not directly selected according to the qualified rate, but other strategies are adopted;

[0102] Other strategies include but are not limited to: using sequential detection method: first detecting a small amount, and dynamically adjusting the detection order according to the preliminary results, for example: first detecting 5 times in two orders respectively, if the qualified rate of a certain order is significantly higher than the other (such as difference > 15%), continue to use this order; If the difference is not significant, increase the detection times to 15 times and make a judgment again;

[0103] This step quantitatively compares the first detection event qualified rate and the second detection event qualified rate, selects the best detection order of the real-time bidirectional detection event based on data, reduces the detection error caused by improper selection of detection order, and optimizes the detection process from the source;

[0104] The difference between the detected leakage rate and the actual leakage rate is calculated to construct a leakage rate deviation sequence, so as to identify the error existing in the detection process. The detection event qualified rate is calculated based on the preset leakage rate deviation allowable range, the detection result meeting the requirements is further screened out, the data with large error is effectively excluded, so as to improve the accuracy of the final detection result, and make the detection result more reflect the real air tightness performance of the vacuum valve;

[0105] After determining the best detection order, the subsequent detection can directly use the order, reducing unnecessary detection attempts and repeated operations, saving detection time and resources, and improving detection efficiency;

[0106] The results of the leakage rate deviation analysis are not only used for current detection, but also provide data support for the design, production and process improvement of vacuum valves. Through long-term accumulation and analysis of leakage rate deviation data of different vacuum valves under various detection conditions, potential problems existing in product design or production process can be found, so as to optimize and improve the product quality;

[0107] Step four: based on the selected optimal detection order and the corresponding leakage rate deviation sequence, set up a leakage rate deviation compensation mechanism to correct the leakage rate in real-time bidirectional detection events and optimize the accuracy of detection;

[0108] In some embodiments, the selected optimal detection order is obtained, and the leakage rate deviation sequence corresponding to the optimal detection order is obtained as a compensation analysis sequence;

[0109] The mean value of the leakage rate deviation in the compensation analysis sequence is calculated, and based on the positive or negative nature of the mean value, it is determined whether the optimal detection order leads to generally high or generally low detection results to determine the adjustment direction;

[0110] According to the positive or negative nature of the mean value, the adjustment direction is determined. If the mean value is positive, the detected leakage rate needs to be reduced by the mean value in real-time detection. If it is negative, the mean value is added;

[0111] The coefficient of variation of the leakage rate deviation in the compensation analysis sequence is calculated. If the coefficient of variation is less than the coefficient of variation limit, it means that the compensation analysis sequence is relatively stable, and the mean value of the leakage rate deviation in the compensation analysis sequence is used as the compensation value of the real-time bidirectional detection event;

[0112] If the coefficient of variation is greater than or equal to the coefficient of variation limit, a dynamic compensation model is constructed based on the key influencing features;

[0113] From the historical bidirectional detection event data corresponding to the compensation analysis sequence, detection condition features with high correlation with the leakage rate deviation are screened, including but not limited to: detection temperature, humidity, atmospheric pressure;

[0114] Specifically, Pearson correlation coefficient is used to judge the correlation, and features with Pearson correlation coefficient greater than the correlation limit are extracted as key influencing features;

[0115] The correlation limit is set by those skilled in the art, which can be set to 0.65. If the correlation coefficient of detection temperature and leakage rate deviation is 0.7, the detection temperature is considered as a key influencing feature;

[0116] A random forest model is constructed as a dynamic compensation model;

[0117] Those skilled in the art can understand that the random forest model is constructed, including model input layer, model structure layer, model training layer and model output layer;

[0118] The model input layer is the key influencing features; the example input feature vector is: [detection temperature (℃), relative humidity (%), atmospheric pressure (kPa), positive pressure detection pressure (MPa)];

[0119] The model structure layer is the basic unit and integration strategy: set the number of trees, use the CART algorithm, and minimize the mean square error (MSE) as the splitting criterion; each tree is randomly selected features (n is the total number of input features) to enhance model diversity; use Bootstrap sampling with replacement to generate training subsets, and use unsampled samples as out-of-bag (OOB) data for validation; aggregate the prediction results of all trees by majority voting (classification tasks) or averaging (regression tasks); limit the maximum depth of a single tree and set the minimum number of samples per leaf node;

[0120] The model training layer aims to establish a nonlinear mapping relationship between key influencing features and leakage rate deviations;

[0121] The model output layer is the leakage rate deviation prediction value of the real-time detection event;

[0122] The real-time values ​​of key influencing features in the real-time bidirectional detection event are input into the dynamic compensation model, and the leakage rate deviation prediction value of the real-time detection event is output as the compensation value;

[0123] This step sets up a leak rate deviation compensation mechanism based on the selected optimal detection sequence and its corresponding leak rate deviation sequence, thereby correcting the leak rate in real-time bidirectional detection events and further optimizing detection accuracy.

[0124] Based on the deviation characteristics (positive and negative mean values, coefficient of variation) in the optimal order, static mean compensation or dynamic model compensation is selected to make the compensation direction and amplitude more consistent with the actual detection error;

[0125] Data stability is determined by the coefficient of variation. When deviations fluctuate significantly, key influencing features are screened based on the Pearson correlation coefficient. A nonlinear dynamic compensation model is constructed using a random forest model to more accurately capture the combined impact of multiple factors on leakage rate deviations, rather than relying on a single fixed compensation value.

[0126] Relying on historical data clustering and real-time matching, the compensation mechanism is closely linked to the current detection conditions, reducing the compensation deviation caused by ignoring scene differences in general compensation methods.

[0127] Example 2:

[0128] Based on the same inventive concept as the vacuum valve airtightness detection method in the above embodiment, Figure 3As shown, the present application provides a vacuum valve airtightness detection device, which specifically comprises a historical detection event analysis module, a real-time detection event matching module, an optimal detection sequence selection module, and a detection deviation compensation module.

[0129] The historical detection event analysis module collects event records of multiple positive pressure-negative pressure bidirectional detections of the vacuum valve from a historical detection database, and performs cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups.

[0130] The historical detection event analysis module collects event records of multiple positive pressure-negative pressure bidirectional detections of the vacuum valve from a historical detection database, and performs cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups.

[0131] The real-time detection event matching module obtains the detection conditions of the real-time bidirectional detection event, obtains a matched detection condition similar event group as a target analysis group through a matching algorithm, and selects events with consistent detection sequences from the target analysis group to construct a first detection sequence event set and a second detection sequence event set.

[0132] After the detection conditions of the real-time bidirectional detection event are converted into a feature vector, the target analysis group is selected through a two-layer matching algorithm: first, the cosine similarity with historical events is calculated to preliminarily screen events with consistent key features; then, the Euclidean distance with candidate cluster centroids is calculated to select the cluster with the smallest distance and less than the minimum threshold, and then the event set is split according to the detection sequence to construct the "positive pressure first and negative pressure second" and "negative pressure first and positive pressure second" event sets, which provide comparison data for deviation analysis.

[0133] The optimal detection sequence selection module performs deviation analysis on the detection leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set to obtain a first leakage rate deviation sequence and a second leakage rate deviation sequence, quantitatively evaluates the first leakage rate deviation sequence and the second leakage rate deviation sequence, and selects the optimal detection sequence for the real-time bidirectional detection event.

[0134] The optimal detection sequence selection module performs deviation analysis on the detection leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set to obtain a first leakage rate deviation sequence and a second leakage rate deviation sequence, quantitatively evaluates the first leakage rate deviation sequence and the second leakage rate deviation sequence, and selects the optimal detection sequence for the real-time bidirectional detection event.

[0135] Detecting deviation compensation module: based on the selected best detection order and the corresponding leakage rate deviation sequence, set up the leakage rate deviation compensation mechanism, correct the leakage rate in real-time bidirectional detection event, optimize the accuracy of detection;

[0136] Based on the selected best detection order corresponding to the deviation sequence, calculate the mean of the deviation to determine the adjustment direction (if the mean is positive, subtract the mean, if the mean is negative, add the mean), judge the stability of the data through the coefficient of variation, if the coefficient of variation is less than the limit value, directly use the mean as the compensation value; if it is greater than the limit value, then screen the features with high correlation with the deviation (such as temperature, humidity), use the random forest model to build a dynamic compensation model, input the real-time feature value to predict the deviation compensation value.

[0137] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method of hermetic testing of a vacuum valve, characterized by: The method comprises the following steps: collecting records of multiple positive pressure-negative pressure bidirectional detection events of the vacuum valve from a historical detection database, and performing cluster analysis on events with similar detection conditions to form multiple event groups with similar detection conditions; obtaining the detection conditions of a real-time bidirectional detection event, obtaining a matched event group with similar detection conditions as a target analysis group through a matching algorithm, and selecting events with consistent detection sequences from the target analysis group to construct a first detection sequence event set and a second detection sequence event set respectively; the obtaining process of the target analysis group comprises: obtaining the detection conditions of the real-time bidirectional detection event of the vacuum valve air tightness, and converting the detection condition data of the real-time bidirectional detection event into a real-time event feature vector; calculating the Euclidean distance between the real-time event feature vector and all candidate cluster centers respectively, and extracting the cluster corresponding to the cluster center with the minimum Euclidean distance as the target analysis group; the obtaining process of the candidate cluster center comprises: calculating the cosine similarity between the real-time event feature vector and all historical detection event feature vectors respectively, extracting historical detection events with a cosine similarity greater than a similarity threshold, obtaining the cluster center of the similar event group to which the historical detection event belongs, and taking the cluster center as a candidate cluster center; performing deviation analysis on the detection leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set respectively to obtain a first leakage rate deviation sequence and a second leakage rate deviation sequence, quantitatively evaluating the first leakage rate deviation sequence and the second leakage rate deviation sequence, and selecting the best detection sequence for the real-time bidirectional detection event; the process of the best detection sequence comprises: extracting events in the first leakage rate deviation sequence within a preset leakage rate deviation allowable range, counting the number of the events, and performing ratio calculation on the number of the events and the total number of events in the first detection sequence event set to obtain a first detection event qualification rate; extracting events in the second leakage rate deviation sequence within a preset leakage rate deviation allowable range, counting the number of the events, and performing ratio calculation on the number of the events and the total number of events in the second detection sequence event set to obtain a second detection event qualification rate; if the first detection event qualification rate is greater than the second detection event qualification rate, the first sequence is selected as the best detection sequence for the real-time bidirectional detection event; if the first detection event qualification rate is less than the second detection event qualification rate, the second sequence is selected as the best detection sequence for the real-time bidirectional detection event; based on the selected best detection sequence and the corresponding leakage rate deviation sequence, a leakage rate deviation compensation mechanism is set to correct the leakage rate in the real-time bidirectional detection event and optimize the accuracy of detection.

2. The method of claim 1, wherein: the process of the event group with similar detection conditions comprises: processing the data obtained from the events of the positive pressure-negative pressure bidirectional detection of the vacuum valve, using a K-Means++ clustering algorithm to perform cluster analysis on events with similar detection conditions, grouping the bidirectional detection events according to the detection conditions, and forming the event group with similar detection conditions.

3. The method of claim 1, wherein: the processes of the first leakage rate deviation sequence and the second leakage rate deviation sequence comprise: obtaining the detection leakage rate and the actual leakage rate of all events in the first detection sequence event set, calculating the difference between the detection leakage rate and the actual leakage rate to obtain a leakage rate deviation, and constructing the first leakage rate deviation sequence; Obtain the detection leakage rate and the actual leakage rate of all events in the second detection sequence event set, and calculate the difference between the detection leakage rate and the actual leakage rate to obtain the leakage rate deviation, and construct a second leakage rate deviation sequence.

4. The method of claim 1, wherein: The leakage rate deviation compensation mechanism comprises: Obtain the selected optimal detection sequence, and obtain the leakage rate deviation sequence corresponding to the optimal detection sequence as a compensation analysis sequence; Calculate the mean value of the leakage rate deviation in the compensation analysis sequence, and determine the adjustment direction based on the positive and negative nature of the mean value. If the mean value is positive, the detection leakage rate needs to be reduced by the mean value in real-time detection. If it is negative, the mean value is added.

5. The method of claim 4, wherein: The leakage rate deviation compensation mechanism further comprises: Calculate the coefficient of variation of the leakage rate deviation in the compensation analysis sequence. If the coefficient of variation is less than the coefficient of variation limit, use the mean value of the leakage rate deviation in the compensation analysis sequence as the compensation value of the real-time bidirectional detection event; If the coefficient of variation is greater than or equal to the coefficient of variation limit, a dynamic compensation model is constructed based on the key influence characteristics; Input the real-time value of the key influence characteristics in the real-time bidirectional detection event into the dynamic compensation model, and output the leakage rate deviation prediction value of the real-time detection event as the compensation value.

6. The method of claim 5, wherein: The key influence characteristics are: From the historical bidirectional detection event data corresponding to the compensation analysis sequence, filter the detection condition characteristics with high correlation with the leakage rate deviation; Use the Pearson correlation coefficient to judge the correlation, and extract the characteristics with a Pearson correlation coefficient greater than the correlation limit as the key influence characteristics.

7. A vacuum valve hermeticity testing apparatus, characterized by, The device is used to execute the method of any one of claims 1-6, comprising: A historical detection event analysis module: collects event records of multiple positive pressure-negative pressure bidirectional detections of the vacuum valve from a historical detection database, and performs cluster analysis on events with similar detection conditions to form multiple detection condition similar event groups; A real-time detection event matching module: obtains the detection conditions of a real-time bidirectional detection event, obtains a matched detection condition similar event group as a target analysis group through a matching algorithm, and selects events with consistent detection sequences from the target analysis group to construct a first detection sequence event set and a second detection sequence event set, respectively; An optimal detection sequence selection module: performs deviation analysis on the detection leakage rate and the actual leakage rate in the first detection sequence event set and the second detection sequence event set, respectively, to obtain a first leakage rate deviation sequence and a second leakage rate deviation sequence, and selects the optimal detection sequence for the real-time bidirectional detection event through quantitative evaluation of the first leakage rate deviation sequence and the second leakage rate deviation sequence; A detection deviation compensation module: sets a leakage rate deviation compensation mechanism based on the selected optimal detection sequence and the corresponding leakage rate deviation sequence, and corrects the leakage rate in the real-time bidirectional detection event to optimize the accuracy of detection.

Images