Shot blasting storage tank safety monitoring method and system based on machine learning

Abstract

The invention relates to the technical field of safety monitoring, in particular to a shot blasting storage tank safety monitoring method and system based on machine learning, and the method comprises the steps: collecting pressure data and spray gun flow data in real time, and constructing a real-time pressure sequence and a real-time spray gun flow sequence; determining an affiliation cluster of the real-time pressure sequence according to the difference between the real-time pressure sequence and a historical pressure sequence corresponding to the clustering center of the pressure cluster, and adjusting the real-time spray gun flow sequence according to an error distribution state corresponding to the affiliation cluster to obtain a real-time shot blasting sequence; the similarity between the real-time shot blasting sequence and the real-time pressure sequence is calculated, and weighted summation is conducted on the similarity based on the error distribution state; and taking a result obtained after similarity weighted summation as change consistency, and judging the operation state of the storage tank based on the change consistency. The shot blasting storage tank safety monitoring system has the effect of improving the timeliness and comprehensiveness of shot blasting storage tank safety monitoring.

Description

Technical Field

[0001] This application relates to the technical field of safety monitoring, and particularly to a safety monitoring method and system for shot peening storage tanks based on machine learning. Background Art

[0002] Shot peening is a surface strengthening process that bombards workpieces with high-speed moving shot particles to implant compressive stress, thereby improving the overall strength of the material. During the operation of shot peening equipment, the shot peening is stored in a shot peening storage tank, and the shot peening is transported to the spray gun through compressed air and then ejected to achieve shot peening treatment on the surface of the workpiece. With the long-term use of shot peening equipment, safety hazards such as metal fatigue and corrosion damage may occur in the shot peening storage tank. If not detected and monitored in time, it will seriously affect the safe operation of the equipment. Traditional detection methods include manual visual inspection, which can detect problems such as cracks and corrosion on the shot peening storage tank. However, there is a certain lag in the actual monitoring process, and it is easy to miss detections. Therefore, with the development and progress of sensing technology, ultrasonic flaw detection is mainly used for detection in the current industry. Its main principle is: when there is a crack or leakage in the storage tank, an eddy current will be generated at the leakage hole, thereby generating an ultrasonic wave segment, and it is judged whether the storage tank leaks according to the waveform received by the ultrasonic probe.

[0003] Compared with the manual visual inspection method, the ultrasonic flaw detection method improves the detection accuracy to a certain extent and reduces the occurrence of missed detections. However, the ultrasonic flaw detection method can only detect cracks after they occur in the shot peening storage tank, and it cannot detect the early tiny changes in the cracks in time; moreover, the traditional ultrasonic flaw detection method is still based on the staff to monitor the shot peening storage tank, that is, the staff needs to manually use the ultrasonic detection device to monitor the shot peening storage tank, and its detection process has a certain lag. Therefore, the traditional detection methods for shot peening storage tanks cannot accurately and real-time monitor some potential risks of shot peening storage tanks, resulting in certain safety risks during the use of shot peening equipment. Summary of the Invention

[0004] To solve the problem that the potential risks of shot peening storage tanks cannot be accurately detected in real time in related technologies, this application provides a safety monitoring method and system for shot peening storage tanks based on machine learning.

[0005] In the first aspect, this application provides a safety monitoring method for shot peening storage tanks based on machine learning, adopting the following technical solutions: A safety monitoring method for shot peening storage tanks based on machine learning, which collects historical data and divides the historical data into multiple data sequences, where the data sequences include historical pressure sequences, historical spray gun flow sequences, and historical shot peening sequences; clusters the historical pressure sequences to construct multiple pressure clustering clusters; Obtain the flow error sequence corresponding to the historical pressure sequence in the pressure clustering cluster based on the historical spray gun flow sequence and the historical shot peening sequence; determine the error distribution state of the pressure clustering cluster based on the quantity and types of the flow error sequences in the pressure clustering cluster; Collect pressure data and spray gun flow data in real time, construct a real-time pressure sequence and a real-time spray gun flow sequence, determine the belonging clustering cluster of the real-time pressure sequence according to the difference between the real-time pressure sequence and the historical pressure sequence corresponding to the clustering center of the pressure clustering cluster, and adjust the real-time spray gun flow sequence according to the error distribution state corresponding to the belonging clustering cluster to obtain a real-time shot peening sequence; calculate the similarity between the real-time shot peening sequence and the real-time pressure sequence, and perform weighted summation on the similarity based on the error distribution state; use the result after weighted summation of the similarity as the change consistency, and judge the operation state of the storage tank based on the change consistency.

[0006] The beneficial effects are as follows: During the process of the shot peening process, there is a certain error between the shot peening flow and the spray gun flow detected by the sensor at the spray gun. Therefore, in this application, the pressure data, spray gun flow data, and shot peening data during the historical operation process of the shot peening equipment are collected to form a historical pressure sequence, a historical spray gun flow sequence, and a historical shot peening sequence. Analyze the historical data, calculate the error between the spray gun flow and the shot peening flow under different pressure conditions, and form a flow error sequence. Subsequently, during the real-time process of the shot peening process, collect pressure data in real time to form a real-time pressure sequence, compare the relationship between the real-time pressure sequence and the historical pressure sequence, and determine the state to which the real-time pressure sequence belongs. Adjust the real-time spray gun flow sequence according to the flow error sequence under different pressure states calculated from the historical data to obtain a real-time shot peening sequence, and judge the operation state of the shot peening storage tank according to the change consistency between the real-time shot peening sequence and the real-time pressure sequence, completing the detection of the shot peening storage tank. Compared with the method of detecting the shot peening storage tank by a staff member holding a detection device in the traditional technology, the method in this application can detect some potential defects that affect the pressure and shot peening flow of the shot peening storage tank, and the detection is more comprehensive. At the same time, with the increase of historical data and the deepening of machine learning, the detection of the shot peening storage tank is more accurate; moreover, in this application, data is collected in real time for detection, improving the timeliness of detecting the shot peening storage tank and reducing the safety risk during the process of the shot peening process.

[0007] Optionally, the step of clustering the historical pressure sequences to construct multiple pressure clustering clusters includes: calculating the sequence distance between any two historical pressure sequences, and clustering the historical pressure sequences based on the sequence distance between the two historical pressure sequences.

[0008] The beneficial effects are as follows: The sequence distance represents the similarity between two historical pressure sequences. The more similar the data in the two historical pressure sequences are, the smaller the distance between the two historical pressure sequences. Cluster the historical pressure sequences according to the sequence distance, so as to cluster the historical pressure sequences with similar data in the historical pressure sequences into a cluster.

[0009] Optionally, use the kmeans clustering method to cluster the historical pressure sequences.

[0010] Optionally, the steps of calculating the sequence distance between any two historical pressure sequences include: taking the square of the difference between the data at the same ordinal positions in the two historical pressure sequences as the local distance, and taking the sum of the local distances corresponding to each data in the historical pressure sequence as the sequence distance between the two historical pressure sequences.

[0011] The beneficial effects are as follows: Calculate the difference of each data in the historical pressure sequence, and accumulate the differences corresponding to each data in the historical pressure data, so as to reflect the sequence distance between the two historical pressure sequences.

[0012] Optionally, the steps of calculating the sequence distance between any two historical pressure sequences include: obtaining the mean value and variance of the pressure data in the historical pressure sequence; determining the sequence distance between the two historical pressure sequences through the difference between the means of the two historical pressure sequences and the difference between the variances.

[0013] The beneficial effects are as follows: Compare the differences in the means and variances of the two historical pressure sequences, so as to reflect the sequence distance between the two historical pressure sequences. This method pays more attention to the overall differences between the two historical pressure sequences, reduces the influence of individual data on the sequence distance, and improves the robustness of the calculation.

[0014] Optionally, the steps of obtaining the flow error sequence include calculating the flow difference between the data at the same ordinal positions in the historical spray gun flow sequence and the historical shot peening sequence, and multiple flow differences form the flow error sequence.

[0015] The beneficial effects are as follows: The historical spray gun flow sequence is detected by a sensor, and the historical shot peening sequence is reflected by the number or type of actual shot peening per unit time. The flow difference between the historical spray gun flow sequence and the shot peening sequence reflects the error between the flow rate detected by the sensor, that is, the spray gun flow rate, and the actual shot peening flow rate. Furthermore, multiple flow difference data form the flow error sequence.

[0016] Optionally, the steps of determining the error distribution state of the pressure clustering cluster based on the number and types of flow error sequences in the pressure clustering cluster include: clustering the corresponding flow error sequences in the pressure clustering cluster to obtain multiple error clustering clusters; obtaining the mean value of the data at the same ordinal positions of multiple flow error sequences in the error clustering cluster to form a representative error sequence; taking the ratio of the number of elements in the error clustering cluster to the total number of corresponding flow error sequences in the pressure clustering group as the probability of the representative error sequence; and obtaining the error distribution state corresponding to the pressure clustering cluster by obtaining the probabilities of multiple representative error sequences.

[0017] The beneficial effects are as follows: In the same pressure clustering cluster, the pressure fluctuations of multiple historical pressure sequences are similar, so one pressure clustering cluster represents a pressure state. Since there are multiple corresponding flow error sequences under a certain pressure condition, by clustering the flow error sequences, the probabilities of different errors under this pressure state can be obtained.

[0018] Optionally, the steps of adjusting the real-time spray gun flow sequence based on the flow error sequence in the pressure clustering cluster to obtain a real-time shot peening sequence include: adding the real-time spray gun flow sequence to the data at the same ordinal positions in the representative error sequence to obtain the real-time shot peening sequence.

[0019] The beneficial effects are as follows: After determining which pressure state the real-time pressure sequence belongs to according to the real-time pressure sequence, the real-time shot peening flow is calculated based on the flow error obtained from historical data analysis, and the real-time shot peening sequence is obtained.

[0020] Optionally, the steps of judging the operating state of the storage tank based on change consistency include setting an alarm threshold, and issuing an alarm in response to the change consistency being less than the alarm threshold.

[0021] In a second aspect, the present application provides a shot peening storage tank safety monitoring system based on machine learning, adopting the following technical solution: The shot peening storage tank safety monitoring system based on machine learning includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the shot peening storage tank safety monitoring method based on machine learning as described above is implemented.

[0022] The beneficial effects are as follows: The shot peening storage tank safety monitoring method based on machine learning as described above is generated into a computer program and stored in the memory to be loaded and executed by the processor. Thus, a system is made according to the memory and the processor, which is convenient to use.

[0023] The present application has the following technical effects: In this application, historical data is analyzed based on machine learning to calculate the difference between the detected spray gun flow rate and the actual shot peening flow rate. Then, according to the pressure data collected in real time, the possible error between the currently collected spray gun flow rate and the actual shot peening flow rate is determined. The real-time shot peening sequence is obtained by adjusting the real-time spray gun flow rate sequence collected in real time. According to the change consistency between the real-time shot peening sequence and the real-time pressure sequence, the running state of the current device is judged in real time. On the one hand, the timeliness of detecting the shot peening storage tank is improved. On the other hand, it is convenient to discover potential defects of the shot peening storage tank other than crack defects, and the safety risk during the shot peening process is reduced. Description of the Drawings

[0024] Figure 1 It is a flowchart of the method for safety monitoring of a shot peening storage tank based on machine learning in an embodiment of this application. Detailed Embodiments

[0025] An embodiment of this application discloses a method for safety monitoring of a shot peening storage tank based on machine learning. The historical operation data of the shot peening storage tank is collected and analyzed to obtain a flow error sequence between the historical spray gun flow rate sequence and the specific historical shot peening sequence. The flow error sequence represents the difference between the actual shot peening flow rate and the spray gun flow rate. By adjusting the data of the spray gun flow rate collected in real time through this difference, the real-time shot peening flow rate data can be obtained, and the shot peening storage tank is detected in real time according to the similarity between the shot peening flow rate and the pressure data. During the operation of the shot peening storage tank, the defects and potential defects of the shot peening storage tank cause certain fluctuations in the pressure data and the actual shot peening data. According to this fluctuation, the defects of the shot peening storage tank are analyzed. On the one hand, with the continuous increase of the historical data volume and the continuous deepening of machine learning, it is convenient to discover potential defects of the shot peening storage tank. On the other hand, the shot peening storage tank is detected in real time to improve the timeliness of detecting the shot peening storage tank.

[0026] Refer to Figure 1 , the method for safety monitoring of a shot peening storage tank based on machine learning includes steps S1 - S3.

[0027] S1: Collect historical data and divide the historical data into multiple data sequences, where the data sequences include a historical pressure sequence, a historical spray gun flow rate sequence, and a historical shot peening sequence; cluster the historical pressure sequence to construct multiple pressure clustering clusters.

[0028] Collect multiple operation data during the historical operation of the shot peening storage tank. In this embodiment, the collected operation data includes the working pressure of the shot peening storage tank, the gun flow rate, and the shot peening flow rate. The collection frequency in this embodiment is 1 Hz. The working pressure data is collected by a pressure sensor, and the gun flow rate is collected by a flow sensor installed on the gun. The shot peening flow rate refers to the flow rate of the actually ejected shot peening, and the shot peening flow rate can be reflected by the quantity or type of the actually ejected shot peening per unit time.

[0029] After the historical data is collected, a data sequence is formed. A segmentation window is set, and each time series is segmented into multiple data sequences based on the size of the segmentation window. Since multiple data are collected in this application, the data sequences formed after the historical data is segmented include a historical pressure sequence, a historical gun flow rate sequence, and a historical shot peening sequence. In this embodiment, the size of the segmentation window is 200, and the length of the corresponding historical pressure sequence is 200.

[0030] Calculate the distance between any two historical pressure sequences; In one embodiment, the steps of calculating two historical pressure sequences include: taking the square of the difference between the data at the same ordinal position in the two historical pressure sequences as the local distance, and taking the sum of the local distances corresponding to each data in the historical pressure sequence as the sequence distance between the two historical pressure sequences.

[0031] Specifically, the calculation formula for the sequence distance between two historical pressure sequences can be expressed as: ; where represents the sequence distance between the th historical pressure sequence and the th historical pressure sequence; represents the th pressure value in the th historical pressure sequence; represents the th pressure value in the th historical pressure sequence; represents the number of data in the historical pressure sequence.

[0032] represents the difference between two pressure values at the same ordinal position in the two historical pressure sequences, that is, the local distance between the two historical pressure sequences. Traverse the differences of all data in the historical pressure sequence to obtain the sequence distance between the two historical pressure sequences. The smaller the sequence distance between the two historical pressure sequences, the more similar the data fluctuations of the two historical pressure sequences are.

[0033] In another embodiment, the step of calculating two historical pressure sequences includes: obtaining the mean and variance of the pressure data in the historical pressure sequences; determining the sequence distance between the two historical pressure sequences based on the difference between the means of the two historical pressure sequences and the difference between the variances.

[0034] Specifically, the calculation formula for the sequence distance between two historical pressure sequences is: ; where represents the sequence distance between the th historical pressure sequence and the th historical pressure sequence; represents the mean of the th historical pressure sequence; represents the mean of the th historical pressure sequence; represents the variance of the th historical pressure sequence; represents the variance of the th historical pressure sequence.

[0035] In this method, the sequence distance between two historical pressure sequences is reflected by the differences in the means and variances of the overall historical pressure sequences, making the finally calculated sequence distance less susceptible to the influence of individual abnormal data and improving the accuracy and robustness of the final sequence distance calculation.

[0036] Cluster the historical pressure sequences based on the sequence distance between the historical pressure sequences, and multiple pressure clustering clusters are formed after clustering. The clustering method used in this embodiment is k-means clustering, and other clustering methods can also be used in other embodiments.

[0037] S2: Obtain the flow error sequences corresponding to the historical pressure sequences in the pressure clustering clusters based on the historical spray gun flow sequences and historical shot peening sequences; determine the error distribution state of the pressure clustering clusters based on the quantity and types of the flow error sequences in the pressure clustering clusters.

[0038] During the process of the shot peening process, since the shot peening is ejected based on the pressure of compressed air, the change of pressure data will cause an increase in the spray gun flow and the shot peening flow; there is a correlation between the pressure and the shot peening flow. However, during the monitoring process, with the fluctuation of the pressure, there are certain errors in the acquisition of the shot peening flow (i.e., the difference between the spray gun flow and the shot peening flow). At the same time, under different pressure change conditions, the errors in the acquisition of the shot peening flow are also different. Therefore, it is necessary to analyze the influence of the change of pressure on the difference between the spray gun flow and the shot peening flow.

[0039] In the pressure clustering clusters, each historical pressure sequence corresponds to a historical spray gun flow rate sequence and a shot peening sequence. The difference between the historical spray gun flow rate sequence and the shot peening sequence represents the difference between the shot peening flow rate data collected by the sensor at the spray gun and the actual flow rate of the finally ejected shot peening.

[0040] The historical spray gun flow rate sequence and the historical shot peening sequence have the same length. The flow rate in the historical spray gun flow rate sequence minus the flow rate at the same position in the historical shot peening sequence gives the flow rate difference. The flow rate differences between each data in the historical spray gun sequence and the corresponding flow rate data in the historical shot peening sequence together constitute the flow rate error sequence. The flow rate error sequence corresponds to the historical pressure sequence, and the data in the flow rate error sequence represents the error between the detected spray gun flow rate and the actual shot peening flow rate under a certain pressure state.

[0041] A pressure clustering group includes multiple historical pressure sequences, and each historical pressure sequence corresponds to a flow rate error sequence. Cluster the multiple flow rate error sequences corresponding to the pressure clustering clusters. The method of clustering the flow rate error sequences here is the same as the principle of clustering the historical pressure sequences, which will not be elaborated here.

[0042] After clustering the flow rate error sequences, multiple error clustering clusters are formed, that is, each pressure clustering cluster corresponds to multiple error clustering clusters. Calculate the mean of the errors at the same position in the multiple flow rate error sequences in the error clustering cluster to obtain the representative error sequence corresponding to the error clustering cluster. The number of flow rate error sequences in the error clustering cluster represents the frequency of occurrence of the representative error sequence. Take the ratio between this frequency and the total number of flow rate error sequences corresponding to the pressure clustering cluster as the probability of the representative error sequence. A pressure clustering cluster corresponds to multiple error clustering clusters, that is, a pressure clustering cluster corresponds to multiple representative error sequences. Calculate the probability of each representative error sequence to obtain the data of the error distribution state corresponding to the pressure clustering cluster.

[0043] Specifically, the calculation formula for the probability of the representative error sequence is: ; represents the th pressure clustering cluster, and the probability that the representative error sequence is ; represents the th pressure clustering cluster, and the th element number in the error clustering cluster corresponding to the representative error sequence; represents the th total number of flow rate error sequences corresponding to the pressure sequence.

[0044] For example, in a pressure clustering cluster, there are 8 flow error sequences corresponding to it. The 8 flow error sequences correspond to three error clustering clusters, which are defined as the first clustering cluster, the second clustering cluster, and the third clustering cluster respectively. The representative sequences corresponding to the first clustering cluster, the second clustering cluster, and the third clustering cluster are the first representative sequence, the second representative sequence, and the third representative sequence respectively. Among them, the first clustering cluster includes 3 flow error sequences; the second clustering cluster includes 2 flow error sequences; the third clustering cluster includes 3 flow error sequences. When the pressure sequence is in this pressure clustering cluster, the flow error sequence between the spray gun flow and the shot blasting flow under this pressure condition has a probability of being the first representative sequence, a probability of being the second representative sequence; a probability of being the third representative sequence.

[0045] S3: Real-time collect pressure data and spray gun flow data, construct a real-time pressure sequence and a real-time spray gun flow sequence, and determine the belonging clustering cluster of the real-time pressure sequence according to the difference between the real-time pressure sequence and the historical pressure sequence corresponding to the clustering center of the pressure clustering cluster.

[0046] During the shot blasting process, collect the real-time pressure data of the shot blasting storage tank and the spray gun flow data. Obtain the real-time pressure sequence and the real-time spray gun flow sequence. Obtain the pressure sequence corresponding to the clustering center of the pressure clustering cluster, and determine this pressure sequence as the central pressure sequence. Calculate the belonging distance between the real-time pressure sequence and different central pressure sequences. The calculation of the belonging distance is the same as the calculation method of the sequence distance between pressure sequences in step S2, which will not be elaborated here. Take the pressure clustering cluster corresponding to the central pressure sequence with the smallest belonging distance as the belonging clustering cluster of the real-time pressure sequence.

[0047] S4: Adjust the real-time spray gun flow sequence according to the error distribution state corresponding to the belonging clustering cluster to obtain a real-time shot blasting sequence; calculate the similarity between the real-time shot blasting sequence and the real-time pressure sequence, and perform weighted summation on the similarity based on the error distribution state; take the result after weighted summation of the similarity as the change consistency.

[0048] The belonging clustering cluster corresponds to multiple representative error sequences, and the real-time spray gun flow sequence is adjusted through the representative error sequences. Specifically, the flow data at the same ordinal position of the real-time spray gun flow sequence and different representative error sequences are added to obtain multiple real-time shot blasting sequences.

[0049] Calculate the similarity of the fluctuations between the real-time shot blasting sequence and the real-time pressure sequence, and perform weighted summation on the similarity based on the error distribution state. Take the result after weighted summation of the similarity as the change consistency, and judge the operating state of the storage tank based on the change consistency.

[0050] In this embodiment, the similarity is the Pearson correlation coefficient between the real-time shot peening sequence and the real-time pressure sequence.

[0051] Specifically, the calculation formula for the change consistency can be expressed as: ; represents the change consistency between the real-time shot peening sequence and the real-time pressure sequence; represents the th similarity between the real-time shot peening sequence and the real-time pressure sequence; represents the th probability of the representative error sequence corresponding to the real-time shot peening sequence.

[0052] After determining the belonging clustering cluster of the real-time pressure sequence, it shows that under this pressure condition, the error sequence formed by the error between the real-time spray gun flow rate and the real-time shot peening flow rate has multiple (i.e., multiple representative error sequences), and the probabilities corresponding to different representative sequences are different. Based on this probability, the similarities between the real-time shot peening sequence and the real-time pressure sequence are weighted and summed to obtain the change consistency between the real-time shot peening flow rate and the real-time pressure sequence, so as to further determine whether the change of the real-time shot peening flow rate with the change of the real-time pressure is relevant, and then determine the current working state of the shot peening storage tank.

[0053] S5: Judge the operating state of the storage tank based on the change consistency.

[0054] Set an alarm threshold, and issue an alarm in response to the change consistency being less than the alarm threshold, and judge that the current working state of the shot peening storage tank is abnormal.

[0055] The embodiment of the present application also discloses a shot peening storage tank safety monitoring system based on machine learning, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the shot peening storage tank safety monitoring method based on the present application is realized.

[0056] The above system also includes other components well known to those skilled in the art such as a communication bus and a communication interface, and their settings and functions are known in the art, so they will not be described in detail here.

[0057] The above are all the preferred embodiments of the present application. The protection scope of the present application is not limited by this. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A safety monitoring method for shot-peened storage tanks based on machine learning, characterized in that, Collect historical data and split the historical data into multiple data sequences, where the data sequences include a historical pressure sequence, a historical spray gun flow rate sequence, and a historical shot peening sequence; perform clustering on the historical pressure sequence to construct multiple pressure clustering clusters; Obtain a flow rate error sequence corresponding to the historical pressure sequence in the pressure clustering cluster based on the historical spray gun flow rate sequence and the historical shot peening sequence; Determine the error distribution state of the pressure clustering cluster based on the quantity and types of the flow rate error sequences in the pressure clustering cluster; Collect pressure data and spray gun flow rate data in real time, construct a real-time pressure sequence and a real-time spray gun flow rate sequence, and determine the belonging clustering cluster of the real-time pressure sequence according to the difference between the real-time pressure sequence and the historical pressure sequence corresponding to the clustering center of the pressure clustering cluster; Adjust the real-time spray gun flow rate sequence according to the error distribution state corresponding to the belonging clustering cluster to obtain a real-time shot peening sequence; calculate the similarity between the real-time shot peening sequence and the real-time pressure sequence, and perform weighted summation on the similarity based on the error distribution state; use the result after the weighted summation of the similarity as the change consistency; Judge the operating state of the storage tank based on the change consistency.

2. The method for safety monitoring of shot peened storage tanks based on machine learning according to claim 1, characterized in that, The steps of performing clustering on the historical pressure sequence to construct multiple pressure clustering clusters include: calculating the sequence distance between any two historical pressure sequences, and clustering the historical pressure sequences based on the sequence distance between the two historical pressure sequences.

3. The method for safety monitoring of shot-peened storage tanks based on machine learning according to claim 2, characterized in that, Use the kmeans clustering method to cluster the historical pressure sequence.

4. The method for safety monitoring of shot peened storage tanks based on machine learning according to claim 2, wherein The steps of calculating the sequence distance between any two historical pressure sequences include: taking the square of the difference between the data at the same position in the two historical pressure sequences as the local distance, and taking the sum of the local distances corresponding to each data in the historical pressure sequence as the sequence distance between the two historical pressure sequences.

5. The method for safety monitoring of shot peened storage tanks based on machine learning according to claim 2, characterized in that, The steps of calculating the sequence distance between any two historical pressure sequences include: obtaining the mean value and variance of the pressure data in the historical pressure sequence; determining the sequence distance between the two historical pressure sequences through the difference between the means of the two historical pressure sequences and the difference between the variances.

6. The method for safety monitoring of shot peened storage tanks based on machine learning according to claim 1, wherein, The steps of obtaining the flow rate error sequence include calculating the flow rate difference between the data at the same position in the historical spray gun flow rate sequence and the shot peening sequence, and multiple flow rate differences form the flow rate error sequence.

7. The method for safety monitoring of shot-peened storage tanks based on machine learning according to claim 1, characterized in that, The steps of determining the error distribution state of the pressure clustering cluster based on the quantity and types of the flow rate error sequences in the pressure clustering cluster include: clustering the corresponding flow rate error sequences in the pressure clustering cluster to obtain multiple error clustering clusters; obtaining the mean value of the data at the same position of the multiple flow rate error sequences in the error clustering cluster to form a representative error sequence; taking the ratio of the number of elements in the error clustering cluster to the total number of the corresponding flow rate error sequences in the pressure clustering group as the probability of the representative error sequence; obtaining the probabilities of multiple representative error sequences to obtain the error distribution state corresponding to the pressure clustering cluster.

8. The method for safety monitoring of shot peened storage tanks based on machine learning according to claim 1, characterized in that The steps of adjusting the real-time spray gun flow rate sequence according to the flow rate error sequence in the pressure clustering cluster to obtain a real-time shot peening sequence include: adding the data at the same position in the real-time spray gun flow rate sequence and the representative error sequence to obtain the real-time shot peening sequence.

9. The method for safety monitoring of shot-peened storage tanks based on machine learning according to claim 1, characterized in that, The steps of judging the operating state of the storage tank based on the change consistency include setting an alarm threshold, and in response to the change consistency being less than the alarm threshold, issuing an alarm.

10. A shot peening storage tank safety monitoring system based on machine learning, characterized in that, Include: A processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement the machine learning-based shot peening storage tank safety monitoring method according to any one of claims 1-9.

Images