Pressure vessel water pressure test anomaly early warning method based on multi-source image cross-validation

Abstract

The application discloses a pressure vessel water pressure test anomaly early warning method based on multi-source image cross verification, relates to the technical field of data analysis, and comprises the following steps: calling water pressure test records of a pressure vessel, extracting test time, fault types and fault levels of each fault position of the test records, and establishing a detection image of a maintenance robot; arranging sensors on the pressure vessel, determining target equipment for detection operation of the pressure vessel to be detected according to sensor data and the detection image; and respectively shooting pressure vessel images by the target equipment and workers, performing feature matching verification in the image content level, identifying abnormal behaviors and timely warning and prompting. The application can judge whether there is an anomaly in the test of the pressure vessel to be detected and timely prompt by linking the test records and multi-source image analysis, which is helpful to guarantee the authenticity and standardization of pressure vessel inspection work, and improve the reliability of water pressure test detection results and the on-site operation control level.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to an anomaly early warning method for pressure vessel hydrostatic tests based on multi-source image cross-validation. Background Technology

[0002] Pressure vessels can be used to hold liquid and gaseous pressurized media. After a period of use, a hydrostatic test is carried out to effectively ensure the quality of the pressure vessel and detect potential faults in the pressure vessel shell, welds and pipes in advance. Staff can then repair and handle the faults in a timely manner, thus preventing major safety accidents such as leaks or explosions from the source. The complete hydrostatic test process mainly includes pre-test preparation, graded pressurization, pressure stabilization and holding, and pressure release and finalization. Although standardized operating procedures have been established for hydrostatic testing of pressure vessels in China, there are still common problems in actual field operations, such as non-standard operation and difficulty in effectively verifying the authenticity of the entire test process. At the same time, there is a lack of intelligent image and data linkage anti-counterfeiting verification mechanisms, which cannot automatically identify and warn of various anomalies, seriously weakening the authenticity and standardization of pressure vessel inspection work and creating a huge safety supervision loophole. Summary of the Invention

[0003] The purpose of this invention is to provide an early warning method for abnormalities in hydrostatic testing of pressure vessels based on multi-source image cross-validation, so as to solve the problems mentioned in the prior art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for early warning of anomalies in hydraulic tests of pressure vessels based on multi-source image cross-validation includes the following steps: Retrieve historical hydrostatic test records of the pressure vessel. These records store information on various faults identified by the maintenance robot during the hydrostatic test. Extract the test time, fault type, and fault level corresponding to each fault location from the test records. Based on the detection results of the maintenance robot for various faults, establish a detection profile of the maintenance robot. The system acquires the pressure vessel currently undergoing hydrostatic testing, deploys various sensors on the pressure vessel, and determines the target equipment for the testing of the pressure vessel from among multiple maintenance robots based on the sensor data and the corresponding inspection profile of the maintenance robot. Monitoring cameras are installed on the target equipment. During the entire process of the hydrostatic test of the pressure vessel to be tested, the target equipment and personnel take pictures of the pressure vessel, so as to realize the overall inspection of the pressure vessel, perform feature matching and verification at the image content level, identify abnormal behavior and issue timely warnings.

[0005] Furthermore, establish a detection profile for the maintenance robot, including: Extract multiple test records corresponding to the maintenance robot Rot, filter out test samples with fault type L, and count the number of faults that actually exist and are correctly detected by the maintenance robot Rot (TP), the number of faults that actually exist but are not detected by the maintenance robot Rot (FN), and the number of faults that are actually not detected by the maintenance robot Rot (FP). Calculate the precision rate P = TP / (TP+FP) and recall rate R = TP / (TP+FN) of the maintenance robot Rot; Calculate the comprehensive detection score of the maintenance robot Rot for fault type L. , where β 2 It is the weighting coefficient, β 2 Greater than 1; It should be noted that pressure vessels are high-risk special equipment. If defects are missed during detection, they can easily lead to major safety accidents such as leaks and explosions, causing personal injury and property damage. False alarms, on the other hand, can be eliminated through secondary verification, making the economic and safety risks manageable. Therefore, this solution requires setting a weighting coefficient β when calculating the comprehensive inspection score. 2 And here we set β 2 >1 can make the evaluation indicators biased towards recall rate, which will have a stronger constraint and punishment effect on the failure to detect faults, and better meet the safety management and control needs of high-risk testing of pressure vessels.

[0006] Based on the comprehensive detection score for each fault type, a detection profile of the maintenance robot Rot is established.

[0007] Furthermore, the target equipment for the inspection of the pressure vessel to be inspected is identified, including: Retrieve the test record corresponding to fault type L, extract the fault severity parameter and test time from the test record, and take the time interval from the time the pressure vessel was manufactured to the time of this test as the target time period of the test record; Various sensors are installed on the pressure vessel to collect the corresponding sensor values. Set a fixed time window duration, divide the target time period into multiple time windows, and obtain the target sensor variable corresponding to the fault type L based on the fluctuation of the sensor values ​​within each time window, and construct the corresponding objective function. Obtain the time interval from the time the pressure vessel to be tested is manufactured to the current time, and calculate the estimated fault severity of the fault type L based on the target sensing variables and objective function. Calculate the estimated fault severity of each fault type in the pressure vessel to be inspected, combine the comprehensive inspection score of the corresponding fault in the inspection profile of each maintenance robot, multiply the comprehensive inspection score of each fault type by its estimated fault severity, and sum them up to obtain the capability value of each maintenance robot. The maintenance robot with the highest capability value was selected as the target equipment for this inspection operation.

[0008] First, in calculating the capability value of the maintenance robot, this application integrates the comprehensive detection score of the maintenance robot for various types of faults and the estimated fault severity for each type of fault. The comprehensive detection score is a subjective capability indicator that characterizes the robot's detection level, while the estimated fault severity is an objective fault indicator that reflects the severity of the fault. This design ensures that the detection advantage of the maintenance robot is only reflected in the overall reliability under corresponding high-risk fault scenarios, effectively eliminating the interference of low-risk irrelevant faults on the capability evaluation, making the final capability value more realistic and objective, and accurately reflecting the actual detection efficiency of the maintenance robot on site.

[0009] Furthermore, the target sensing variables corresponding to fault type L are obtained, and the corresponding objective function is constructed, including: The target time period is divided into several time windows. The normal value range of a certain sensor data T is obtained. The number of samples of sensor data T that exceed the normal value range within a certain time window is counted. The number of samples is normalized to obtain the target value of sensor data T in the time window. Calculate the target value of the sensor data T in each time window, and perform an accumulation operation on the target value to obtain the characteristic value of the sensor data T within the target time period; Obtain the fault level matched by multiple test records corresponding to fault type L and the feature value of sensor data T, fit a linear function of the fault level changing with the feature value, and solve the goodness of fit corresponding to the linear fit. Based on the linear functions and corresponding goodness-of-fit values ​​obtained by fitting fault type L with various types of sensor data, the sensor data corresponding to the largest goodness-of-fit value is selected and set as the target sensor variable for fault type L. At the same time, the linear fitting function corresponding to this sensor data is used as the objective function.

[0010] Furthermore, the steps to determine the normal value range are as follows: obtain the values ​​of the sensor data T of the pressure vessel at multiple operating times, calculate the average value µ and the standard deviation σ, and based on the 3σ principle, take the range [µ-3σ, µ+3σ] as the normal value range of the sensor data T.

[0011] The 3σ principle is an existing technology. Since the probability of data falling within the mean ± 3 times the standard deviation is 99.73%, and the probability of falling outside this range is only 0.27%, the method of determining the normal value range based on the 3σ principle in this solution can effectively cover the vast majority of test data under normal operating conditions, accurately distinguish normal operating data from abnormal distorted data caused by sensor noise, test disturbances or operating condition deviations, and ensure that the values ​​in the normal value range are representative and authentic.

[0012] Furthermore, the calculation of the current estimated fault level of fault type L includes: obtaining the time interval from the time the pressure vessel to be tested is manufactured to the current time, dividing the time interval into multiple time windows, calculating the target value of the target sensor variable in each time window based on the monitoring data of the target sensor variable in each time window, obtaining the characteristic value of the target sensor variable in the time interval, and substituting the characteristic value into the objective function to obtain the current predicted fault level of fault type L.

[0013] Furthermore, identifying abnormal behavior and providing timely early warnings includes: Extract the target components on the pressure vessel to be tested. The target components are those that have experienced failures in the past. Based on the historical test records, calculate the monitoring values ​​of each target component. Sort the target components in descending order of the monitoring values ​​and construct the corresponding monitoring value set. The number of images of each target component in the pressure vessel to be inspected, taken by the target equipment and personnel, is collected, and a set of image counts is constructed based on the sorting number of each target component. Calculate the cosine similarity between the set of monitored values ​​and the set of quantities. If the cosine similarity is lower than the preset similarity threshold, output a prompt message indicating insufficient regulatory efforts.

[0014] Furthermore, the monitoring values ​​of each target component are calculated, including: obtaining the statistical number of various types of failures in the history of a target component, and calculating the proportion of each type of failure based on the total number of failures; multiplying the proportion of each type of failure by the estimated failure severity of the corresponding type one by one and summing them up to obtain the monitoring value of the target component, and then obtaining the monitoring value of each target component.

[0015] Furthermore, if the cosine similarity is not lower than the preset similarity threshold, identifying abnormal behavior and issuing timely warnings also includes: before uploading fault information to the system, extracting the pressure vessel fault locations marked by the staff and the target equipment respectively, counting the number of intersection points and the number of union points of the two types of marked fault locations, calculating the ratio of the number of intersection points to the number of union points, and if the ratio is lower than the preset ratio threshold, triggering a warning and suspending the uploading of the fault data to the system.

[0016] This step involves a human-machine cross-verification process before uploading fault information to the system. By calculating the cross-comparison ratio (CCR) between the fault locations marked by personnel and equipment, the consistency between the two sets of judgments is quantified. When the CCR is lower than a preset threshold, it indicates a significant discrepancy in the fault location identification between the two parties, and the reliability of the fault location result is insufficient. At this point, an early warning is activated and data upload is suspended, urging on-site re-verification to confirm the true location of the fault. This prevents erroneous and distorted fault information from being stored in the management system, ensuring the accuracy of the pressure vessel fault log data and preventing safety and management issues such as maintenance omissions and ineffective operation and maintenance caused by disordered fault records.

[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides an anomaly early warning method for pressure vessel hydrostatic testing based on multi-source image cross-validation, including: retrieving the hydrostatic test record of the pressure vessel, extracting the test time, fault type and fault level of each fault location from the test record, and establishing a detection profile of the maintenance robot; deploying sensors on the pressure vessel, and determining the target equipment for the inspection operation of the pressure vessel to be inspected based on sensor data and the detection profile; taking images of the pressure vessel through the target equipment and personnel respectively, performing feature matching verification at the image content level, identifying abnormal behavior and providing timely early warning. This invention, by linking test records and multi-source image analysis, determines whether there are any anomalies during the test of the pressure vessel to be inspected and provides timely warnings, which helps to ensure the authenticity and standardization of pressure vessel inspection work, and improves the reliability of hydrostatic test results and the level of on-site operation control. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the abnormal early warning method for pressure vessel hydrostatic testing based on multi-source image cross-validation according to the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0021] Example: Figure 1 As shown, this invention provides a technical solution for an anomaly early warning method in pressure vessel hydrostatic testing based on multi-source image cross-validation, comprising the following steps: Retrieve historical hydrostatic test records of the pressure vessel. These records store information on various faults identified by the maintenance robot during hydrostatic testing. Extract the test time, fault type, and fault level corresponding to each fault location from the test records. Based on the maintenance robot's detection results for various faults, establish a detection profile for the maintenance robot, specifically: Extract multiple test records corresponding to the maintenance robot Rot, filter out test samples with fault type L, and count the number of faults that actually exist and are correctly detected by the maintenance robot Rot (TP), the number of faults that actually exist but are not detected by the maintenance robot Rot (FN), and the number of faults that are actually not detected by the maintenance robot Rot (FP). Calculate the precision rate P = TP / (TP+FP) and recall rate R = TP / (TP+FN) of the maintenance robot Rot; Calculate the comprehensive detection score of the maintenance robot Rot for fault type L. , where β 2 It is the weighting coefficient, β 2 Greater than 1; It should be noted that pressure vessels are high-risk special equipment. If defects are missed during detection, they can easily lead to major safety accidents such as leaks and explosions, causing personal injury and property damage. False alarms, on the other hand, can be eliminated through secondary verification, making the economic and safety risks manageable. Therefore, this solution requires setting a weighting coefficient β when calculating the comprehensive inspection score. 2 And here we set β 2 >1, which can make the evaluation index biased towards recall rate, creating a stronger constraint and penalty effect on missed fault detection, and better meeting the safety management requirements of high-risk testing in pressure vessel hydrostatic testing. In this embodiment, β 2 =2.

[0022] Based on the comprehensive detection score for each fault type, a detection profile of the maintenance robot Rot is established.

[0023] The process involves identifying the pressure vessel currently undergoing hydrostatic testing, deploying various sensors on the vessel, and determining the target device from among multiple maintenance robots to perform the testing operation. Retrieve the test record corresponding to fault type L, extract the fault severity parameter and test time from the test record, and take the time interval from the time the pressure vessel was manufactured to the time of this test as the target time period of the test record; Various sensors are installed on the pressure vessel to collect the corresponding sensor values. These sensors include pressure sensors, vibration sensors, temperature sensors, etc. It should be noted that the generation and expansion processes of different fault types are affected by different physical parameters. In other words, different physical quantities will cause different types of faults to pressure vessels. For example, continuous abnormal pressure fluctuations usually correspond to faults such as vessel leakage and the opening of microcracks in welds, while abnormal temperature deviations from the normal range can lead to problems such as localized corrosion, media leakage, and abnormal heat transfer. Because there are different inherent correlations between various faults and their corresponding sensing physical quantities, this invention fits the functional relationship between the feature values ​​of each sensor and the defect level for each single fault type. Based on the goodness of fit, the sensing quantity with the best correlation is selected as the target sensing variable to quantitatively characterize the degree of fault development. The specific implementation is as follows: A fixed time window duration is set, dividing the target time period into multiple time windows. Based on the fluctuation of sensor values ​​within each time window, the target sensor variable corresponding to fault type L is obtained, and the corresponding objective function is constructed. The steps include: The target time period is divided into several time windows. The normal value range of a certain sensor data T is obtained. The number of samples of sensor data T that exceed the normal value range within a certain time window is counted. The number of samples is normalized to obtain the target value of sensor data T in the time window. In this embodiment, the duration of the time window is 1000 hours, meaning 1000 hours is considered as a time window. For example, if the target time period is 5000 hours, then the target time period can be divided into 5 time windows. Taking the sensor data T as temperature as an example, the normal temperature range for a pressure vessel during normal operation is [t1, t2]. First, the number of samples where the sensor data T exceeds the normal range within a certain time window is obtained, and then normalization is performed to obtain the target value. In this embodiment, the normalization formula is sigmoid, where the larger the number of samples, the larger the target value.

[0024] The steps for determining the normal value range are as follows: obtain the sensor data T values ​​of the pressure vessel at multiple operating times, calculate the average value µ and standard deviation σ, and based on the 3σ principle, take the range [µ-3σ, µ+3σ] as the normal value range of the sensor data T.

[0025] Calculate the target value of the sensor data T in each time window, and perform an accumulation operation on the target value to obtain the characteristic value of the sensor data T within the target time period; It should be noted that the eigenvalue is obtained by accumulating the target value of the sensor data T in each time window. Therefore, the magnitude of the eigenvalue is mainly affected by two factors: the statistical duration and the fluctuation level of the sensor data T itself. Since all sensor data in this scheme adopt the same segmented window division method, that is, they all use the same total time duration, the time dimension conditions corresponding to each group of sensor data are unified. Under this premise, the difference in the eigenvalues ​​calculated by different sensor data is only determined by the degree of abnormal fluctuation of the sensor data itself. The higher the eigenvalue, the stronger the overall abnormal accumulation of the sensor data, and the greater the impact of the fault type L on the sensor data. This is reflected in the stronger linear correlation between the fault level and the eigenvalue, and the higher the goodness of fit of the linear function obtained by fitting the two. Accordingly, we can traverse all sensor data, select the sensor quantity corresponding to the maximum goodness of fit as the target sensor variable of fault type L, match and construct a dedicated objective function, and realize quantitative prediction of defect degree, as follows: Obtain the fault level matched by multiple test records corresponding to fault type L and the feature value of sensor data T, fit a linear function of the fault level changing with the feature value, and solve the goodness of fit corresponding to the linear fit. Based on the linear functions and corresponding goodness-of-fit values ​​obtained by fitting fault type L with various types of sensor data, the sensor data corresponding to the largest goodness-of-fit value is selected and set as the target sensor variable for fault type L. At the same time, the linear fitting function corresponding to this sensor data is used as the objective function.

[0026] Obtain the time interval from the manufacturing date to the current time for the pressure vessel under test. Based on the target sensor variable and objective function, calculate the estimated fault severity of fault type L. The steps include: The time interval from the time the pressure vessel to be tested is obtained to the current time. This time interval is divided into multiple time windows. Based on the monitoring data of the target sensor variable in each time window, the target value of the target sensor variable in each time window is calculated. Based on this, the characteristic value of the target sensor variable in the time interval is obtained. The characteristic value is substituted into the objective function to obtain the current predicted fault degree of fault type L.

[0027] The duration of the time window here is the same as described above. For the specific steps to obtain the target value and feature value, please refer to the steps above. They will not be elaborated on here.

[0028] Calculate the estimated fault severity of each fault type in the pressure vessel to be inspected, combine the comprehensive inspection score of the corresponding fault in the inspection profile of each maintenance robot, multiply the comprehensive inspection score of each fault type by its estimated fault severity, and sum them up to obtain the capability value of each maintenance robot. The maintenance robot with the highest capability value was selected as the target equipment for this inspection operation.

[0029] First, in calculating the capability value of the maintenance robot, this application integrates the comprehensive detection score of the maintenance robot for various types of faults and the estimated fault severity for each type of fault. The comprehensive detection score is a subjective capability indicator that characterizes the robot's detection level, while the estimated fault severity is an objective fault indicator that reflects the severity of the fault. This design ensures that the detection advantage of the maintenance robot is only reflected in the overall reliability under corresponding high-risk fault scenarios, effectively eliminating the interference of low-risk irrelevant faults on the capability evaluation, making the final capability value more realistic and objective, and accurately reflecting the actual detection efficiency of the maintenance robot on site.

[0030] Surveillance cameras are deployed on the target equipment. During the entire hydrostatic test of the pressure vessel under inspection, both the target equipment and personnel capture images of the pressure vessel. This allows for comprehensive inspection of the pressure vessel, feature matching verification at the image content level, identification of abnormal behavior, and timely early warning. Specifically: Extract the target components on the pressure vessel to be tested. The target components are those that have experienced failures in the past. Based on the historical test records, calculate the monitoring values ​​of each target component. Sort the target components in descending order of the monitoring values ​​and construct the corresponding monitoring value set. The calculation of monitoring values ​​for each target component includes: Obtain the statistical number of various types of failures in the history of a target component, and calculate the proportion of each type of failure based on the total number of failures; multiply the proportion of each type of failure by the estimated failure severity of the corresponding type one by one and sum them up to obtain the monitoring value of the target component, and then obtain the monitoring value of each target component.

[0031] The number of images of each target component in the pressure vessel to be inspected, taken by the target equipment and personnel, is collected, and a set of image counts is constructed based on the sorting number of each target component. Calculate the cosine similarity between the set of monitored values ​​and the set of quantities. If the cosine similarity is lower than the preset similarity threshold, output a prompt message indicating insufficient regulatory efforts.

[0032] It should be noted that cosine similarity is used to measure the consistency of the changing trends of two sets of data. When the cosine similarity between the set of monitored values ​​and the set of quantities is large, it indicates that the risk ranking trend of components and the shooting frequency ranking trend are consistent. High-risk components are indeed inspected more frequently, the inspection configuration is more reasonable, and the supervision is more effective. Therefore, if the cosine similarity is lower than the preset similarity threshold, it indicates that there is an imbalance where high-risk components are photographed less frequently, while low-risk components are photographed too often. There are control loopholes such as missed inspections of key components, perfunctory inspection arrangements, or inadequate performance of duties. This triggers an early warning of insufficient supervision, reminding managers to optimize the allocation of inspection frequency and make up for the shortcomings in the inspection of key areas.

[0033] If the cosine similarity is not lower than the preset similarity threshold, identifying abnormal behavior and issuing timely warnings also includes: before uploading fault information to the system, extracting the pressure vessel fault locations marked by the staff and the target equipment respectively, counting the number of intersection points and the number of union points of the two types of marked fault locations, calculating the ratio of the number of intersection points to the number of union points, and if the ratio is lower than the preset ratio threshold, triggering a warning and suspending the uploading of the fault data to the system.

[0034] This step involves a human-machine cross-verification process before uploading fault information to the system. By calculating the cross-comparison ratio (CCR) between the fault locations marked by personnel and equipment, the consistency between the two sets of judgments is quantified. When the CCR is lower than a preset threshold, it indicates a significant discrepancy in the fault location identification between the two parties, and the reliability of the fault location result is insufficient. At this point, an early warning is activated and data upload is suspended, urging on-site re-verification to confirm the true location of the fault. This prevents erroneous and distorted fault information from being stored in the management system, ensuring the accuracy of the pressure vessel fault log data and preventing safety and management issues such as maintenance omissions and ineffective operation and maintenance caused by disordered fault records.

[0035] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0036] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0037] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation, characterized in that, Includes the following steps: Retrieve historical hydrostatic test records of the pressure vessel. These records store information on various faults identified by the maintenance robot during the hydrostatic test. Extract the test time, fault type, and fault level corresponding to each fault location from the test records. Based on the detection results of the maintenance robot for various faults, establish a detection profile of the maintenance robot. The system acquires the pressure vessel currently undergoing hydrostatic testing, deploys various sensors on the pressure vessel, and determines the target equipment for the testing of the pressure vessel from among multiple maintenance robots based on the sensor data and the corresponding inspection profile of the maintenance robot. Monitoring cameras are installed on the target equipment. During the entire process of the hydrostatic test of the pressure vessel to be tested, the target equipment and personnel take pictures of the pressure vessel, so as to realize the overall inspection of the pressure vessel, perform feature matching and verification at the image content level, identify abnormal behavior and issue timely warnings.

2. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 1, characterized in that, Establish a detection profile for the maintenance robot, including: Extract multiple test records corresponding to the maintenance robot Rot, filter out test samples with fault type L, and count the number of faults that actually exist and are correctly detected by the maintenance robot Rot (TP), the number of faults that actually exist but are not detected by the maintenance robot Rot (FN), and the number of faults that are actually not detected by the maintenance robot Rot (FP). Calculate the precision rate P = TP / (TP+FP) and recall rate R = TP / (TP+FN) of the maintenance robot Rot; Calculate the comprehensive detection score of the maintenance robot Rot for fault type L. , where β 2 It is the weighting coefficient, β 2 Greater than 1; Based on the comprehensive detection score for each fault type, a detection profile of the maintenance robot Rot is established.

3. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 1, characterized in that, Identify the target equipment for the inspection of the pressure vessel to be inspected, including: Retrieve the test record corresponding to fault type L, extract the fault severity parameter and test time from the test record, and take the time interval from the time the pressure vessel was manufactured to the time of this test as the target time period of the test record; Various sensors are installed on the pressure vessel to collect the corresponding sensor values. Set a fixed time window duration, divide the target time period into multiple time windows, and obtain the target sensor variable corresponding to the fault type L based on the fluctuation of the sensor values ​​within each time window, and construct the corresponding objective function. Obtain the time interval from the time the pressure vessel to be tested is manufactured to the current time, and calculate the estimated fault severity of the fault type L based on the target sensing variables and objective function. Calculate the estimated fault severity of each fault type in the pressure vessel to be inspected, combine the comprehensive inspection score of the corresponding fault in the inspection profile of each maintenance robot, multiply the comprehensive inspection score of each fault type by its estimated fault severity, and sum them up to obtain the capability value of each maintenance robot. The maintenance robot with the highest capability value was selected as the target equipment for this inspection operation.

4. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 3, characterized in that, Obtain the target sensing variables corresponding to fault type L, and construct the corresponding objective function, including: The target time period is divided into several time windows. The normal value range of a certain sensor data T is obtained. The number of samples of sensor data T that exceed the normal value range within a certain time window is counted. The number of samples is normalized to obtain the target value of sensor data T in the time window. Calculate the target value of the sensor data T in each time window, and perform an accumulation operation on the target value to obtain the characteristic value of the sensor data T within the target time period; Obtain the fault level matched by multiple test records corresponding to fault type L and the feature value of sensor data T, fit a linear function of the fault level changing with the feature value, and solve the goodness of fit corresponding to the linear fit. Based on the linear functions and corresponding goodness-of-fit values ​​obtained by fitting fault type L with various types of sensor data, the sensor data corresponding to the largest goodness-of-fit value is selected and set as the target sensor variable for fault type L. At the same time, the linear fitting function corresponding to this sensor data is used as the objective function.

5. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 4, characterized in that, The steps to determine the normal value range are as follows: obtain the sensor data T values ​​of the pressure vessel at multiple operating times, calculate the average value µ and standard deviation σ, and based on the 3σ principle, take the range [µ-3σ, µ+3σ] as the normal value range of the sensor data T.

6. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 4, characterized in that, The calculation of the current predicted fault level of fault type L includes: obtaining the time interval from the time the pressure vessel under test is manufactured to the current time, dividing the time interval into multiple time windows, calculating the target value of the target sensor variable in each time window based on the monitoring data of the target sensor variable in each time window, obtaining the characteristic value of the target sensor variable in the time interval, and substituting the characteristic value into the objective function to obtain the current predicted fault level of fault type L.

7. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 6, characterized in that, Identify abnormal behavior and provide timely warnings, including: Extract the target components on the pressure vessel to be tested. The target components are those that have experienced failures in the past. Based on the historical test records, calculate the monitoring values ​​of each target component. Sort the target components in descending order of the monitoring values ​​and construct the corresponding monitoring value set. The number of images of each target component in the pressure vessel to be inspected, taken by the target equipment and personnel, is collected, and a set of image counts is constructed based on the sorting number of each target component. Calculate the cosine similarity between the set of monitored values ​​and the set of quantities. If the cosine similarity is lower than the preset similarity threshold, output a prompt message indicating insufficient regulatory efforts.

8. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 7, characterized in that, The monitoring values ​​for each target component are calculated, including: obtaining the statistical number of various types of failures in the history of a target component, and calculating the proportion of each type of failure based on the total number of failures; multiplying the proportion of each type of failure by the estimated failure severity of the corresponding type one by one and summing them up to obtain the monitoring value of the target component, and then obtaining the monitoring value of each target component.

9. The method for early warning of anomalies in pressure vessel hydrostatic tests based on multi-source image cross-validation according to claim 7, characterized in that, If the cosine similarity is not lower than the preset similarity threshold, identifying abnormal behavior and issuing timely warnings also includes: before uploading fault information to the system, extracting the pressure vessel fault locations marked by the staff and the target equipment respectively, counting the number of intersection points and the number of union points of the two types of marked fault locations, calculating the ratio of the number of intersection points to the number of union points, and if the ratio is lower than the preset ratio threshold, triggering a warning and suspending the uploading of the fault data to the system.

Images