Valve sealing leakage detection method based on sensor monitoring data

Abstract

The invention relates to the technical field of valve sealing leakage detection, in particular to a valve sealing leakage detection method based on sensor monitoring data. The method comprises the steps of obtaining a termination judgment index value of a second outer cycle, judging whether the termination judgment index value of the second outer cycle is larger than a preset termination threshold value or not, if not, continuing to obtain a termination judgment index value of a third outer cycle, judging whether the termination judgment index value of the third outer cycle is larger than the preset termination threshold value or not, and if yes, continuing to obtain the termination judgment index value of the third outer cycle. If yes, it is judged that the outer circulation is terminated, and a result of conducting STL decomposition on the monitoring data sequence is obtained; and performing sealing leakage detection identification on the to-be-detected valve according to a result of performing the STL decomposition on the monitoring data sequence. And the accuracy and the credibility of detecting the sealing leakage of the valve can be improved.

Description

Technical Field

[0001] This invention relates to the field of valve seal leakage detection technology, and specifically to a valve seal leakage detection method based on sensor monitoring data. Background Technology

[0002] Since valve sealing leaks can lead to production interruptions, resource waste, economic losses, and equipment damage, it is necessary to detect valve sealing leaks in order to avoid a series of negative impacts.

[0003] Furthermore, when using sensors and other data acquisition devices to collect data reflecting the presence of valve sealing leaks, these devices are often subject to interference from process fluctuations, operational disturbances, or noise. This interference data may overlap with abnormal signals caused by valve sealing leaks in amplitude or frequency, leading to detection errors. Therefore, to improve detection accuracy, the collected data sequence is usually decomposed using STL (Standard Time Series) to avoid the influence of interference data on the detection results. However, when performing STL decomposition on the time-series data reflecting the presence of valve sealing leaks, the number of iterations in the outer loop is usually set to an empirically fixed value. This iteration strategy may cause the real high-frequency pulses caused by the leak to be suppressed as noise, or the residuals may contain interference signals such as process fluctuations, operational disturbances, or noise. This can result in low accuracy when subsequently detecting valve sealing leaks based on the decomposition results. Therefore, how to adaptively adjust the outer loop iteration strategy to improve the accuracy of valve sealing leak detection when performing STL decomposition on the time-series data reflecting the presence of valve sealing leaks has become an urgent problem to be solved. Summary of the Invention

[0004] To address the above problems, this invention provides a valve sealing leakage detection method based on sensor monitoring data, the specific technical solution of which is as follows: One embodiment of the present invention provides a valve seal leakage detection method based on sensor monitoring data, comprising the following steps: The monitoring data sequence of the valve to be tested is acquired using sensors; Based on the differences between adjacent data in the monitoring data sequence and the time interval between adjacent maxima, regularity characterization values ​​are obtained; The monitoring data sequence undergoes a first and second outer loop. Based on the differences in output results and weight values ​​between the second and first outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value, a termination judgment index value for the second outer loop is obtained. It is then determined whether the termination judgment index value for the second outer loop is greater than a preset termination threshold. If it is not greater, a third outer loop is performed on the monitoring data sequence. The results of the third outer loop are then compared with... The differences in output results and weight values ​​between the second and third outer loops, the autocorrelation coefficient of the residual components output by the third outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value are used to obtain the termination judgment index value of the third outer loop. It is then determined whether the termination judgment index value of the third outer loop is greater than the preset termination threshold. If it is greater, the outer loop is determined to be terminated, and the result of STL decomposition of the monitoring data sequence is obtained. Based on the result of STL decomposition of the monitoring data sequence, the valve to be tested is identified for sealing leakage.

[0005] Beneficial Effects: This invention first obtains regularity characterization values ​​based on the differences between adjacent data and the time interval between adjacent maxima in the monitoring data sequence. Then, it performs a first and second outer loop on the monitoring data sequence. Based on the differences in output results and weight values ​​between the second and first outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization values, it obtains the termination judgment index value for the second outer loop. Then, it determines whether the termination judgment index value for the second outer loop is greater than a preset termination threshold. If it is not greater, the process continues. The monitoring data sequence undergoes a third external loop. Based on the differences in output results and weight values ​​between the second and third external loops, the autocorrelation coefficient of the residual components output by the third external loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characteristics, a termination judgment index value for the third external loop is obtained. The system then checks whether the termination judgment index value for the third external loop exceeds a preset termination threshold. If it does, the external loop is terminated, and the STL decomposition result of the monitoring data sequence is obtained. Finally, the sealing leakage detection of the valve to be tested is performed based on the STL decomposition result. Furthermore, this invention uses the termination judgment index value as the condition for terminating the external loop, which results in higher quality and more reliable STL decomposition results, thereby improving the accuracy and reliability of valve sealing leakage detection. Attached Figure Description

[0006] 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.

[0007] Figure 1 This is a flowchart of a valve sealing leakage detection method based on sensor monitoring data according to the present invention. Detailed Implementation

[0008] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the protection scope of the embodiments of the present invention.

[0009] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.

[0010] This embodiment provides a valve sealing leakage detection method based on sensor monitoring data, which is described in detail below: like Figure 1 As shown, the valve seal leakage detection method based on sensor monitoring data includes the following steps: Step S001: Use sensors to collect and acquire monitoring data sequences of the valve to be tested.

[0011] The purpose of this embodiment is to improve the accuracy of valve sealing leakage detection by adaptively adjusting the external loop iteration strategy when performing STL decomposition on time-series data reflecting whether valve sealing leakage exists. When a valve is fully closed or in a regulating state, if the sealing pair (valve disc and seat, valve stem packing, etc.) experiences wear, corrosion, impurity blockage, or poor assembly, the medium will flow unexpectedly from the high-pressure side to the low-pressure side (or leak into the environment). This undesigned medium flow is called valve sealing leakage. Therefore, based on the above description, this embodiment needs to first obtain time-series data reflecting valve sealing leakage and then perform STL decomposition on it. The specific process for obtaining time-series data reflecting valve sealing leakage is as follows: First, the valve to be tested for sealing leaks is identified and designated as the valve to be tested. Then, sensors are used to acquire various types of data related to valve sealing leaks. The time series sequence of the same type of data acquired within the monitoring period is designated as the initial data sequence. The initial data sequence is then preprocessed, and the subsequent sequence is designated as the monitoring data sequence for the valve to be tested. In specific applications, the implementer needs to set the monitoring period and acquisition frequency according to the actual situation such as valve type, usage frequency, and operating conditions. For example, in this embodiment, the monitoring period can be set to 24 hours and the acquisition frequency to 1 Hz. The data types related to valve sealing leaks or the data types reflecting whether the valve has sealing leaks include, but are not limited to, temperature, pressure, flow rate, vibration, and humidity. In addition, the preprocessing process of the initial data sequence in this embodiment includes, but is not limited to, filling missing values ​​and normalization. Filling missing values ​​and normalization are well-known technologies and will not be described in detail.

[0012] Therefore, this embodiment obtains the monitoring data sequence of the valve to be detected through the above process. Since the STL decomposition process or the outer loop iteration strategy for different types of monitoring data sequences is consistent in this embodiment, in order to facilitate understanding and description, this embodiment will be described in the following description using the outer loop iteration process of any monitoring data sequence. That is, the monitoring data sequences that appear in this embodiment are all the same monitoring data sequence.

[0013] Step S002: Based on the differences between adjacent data in the monitoring data sequence and the time interval between adjacent maxima, a regularity characterization value is obtained.

[0014] Because data acquisition devices such as sensors are often subject to interference from process fluctuations, operational disturbances, or noise, this interference data may overlap with abnormal signals caused by valve seal leakage in terms of amplitude or frequency, thus negatively impacting valve seal leakage detection. Currently, to improve detection accuracy, the acquired data sequence is generally decomposed using STL (Standard Transmission Time) to avoid the influence of interference data on the detection results, providing clearer, cleaner, more reliable, or trustworthy data for subsequent valve seal leakage detection. However, when performing STL decomposition on the monitoring data sequence, existing external loop iteration strategies may result in the real high-frequency pulses caused by leakage being suppressed as noise after iteration, or the residual after iteration being mixed with interference signals from process fluctuations, operational disturbances, or noise. However, after the iteration is completed, if real high-frequency pulses caused by leakage are suppressed as noise or interference signals are mixed into the residuals after the iteration, the accuracy and reliability of subsequent valve sealing leakage detection based on the decomposition results will be low. To avoid the problems caused by the existing external loop iteration strategy, this embodiment will terminate the external loop iteration based on the analysis results of the convergence and residual quality after each iteration, under the premise of the strength of the regularity of the original sequence. This will reduce the probability that real high-frequency pulses caused by leakage are suppressed as noise or interference signals are mixed into the residuals after the iteration, thereby improving the decomposition quality and the accuracy and reliability of subsequent valve sealing leakage detection based on the decomposition results.

[0015] Based on the above description, this embodiment needs to first analyze the regularity strength of the sequence before decomposition. In STL decomposition, the root cause of excessive or insufficient outer loop iteration is closely related to the difference in regularity of the data itself. In other words, the strength of regularity in the original sequence directly affects the convergence speed and component extraction accuracy of the decomposition. Similarly, the root cause of phenomena such as leakage after iteration leading to the suppression of real high-frequency pulses as noise or the introduction of interference signals into the residual is closely related to the strength of regularity in the original sequence. To reduce the probability of leakage causing real high-frequency pulses to be suppressed as noise or interference signals to be introduced into the residual, this embodiment needs to analyze the strength of regularity in the sequence before decomposition. That is, it needs to analyze and obtain the regularity characterization value of the monitoring data sequence. The specific process for obtaining the regularity characterization value of the monitoring data sequence is as follows: First, based on the differences between adjacent data points in the monitoring data sequence, the trend characteristic value of the monitoring data sequence is obtained. Then, based on the time interval between adjacent maxima in the monitoring data sequence, the periodic characteristic value of the monitoring data sequence is obtained. Finally, the mean of the trend characteristic value and the periodic characteristic value of the monitoring data sequence is calculated and used as the regularity characteristic value of the monitoring data sequence. The larger the trend characteristic value and the periodic characteristic value of the monitoring data sequence, the more obvious the trend and periodic characteristics of the monitoring data sequence are, and the stronger the regularity of the monitoring data sequence. Conversely, the smaller the trend characteristic value and the periodic characteristic value of the monitoring data sequence, the less obvious the trend and periodic characteristics are, and the weaker the regularity of the monitoring data sequence.

[0016] In this embodiment, the specific process of obtaining the trend characterization value of the monitoring data sequence based on the differences between adjacent data in the monitoring data sequence is as follows: First, the first and second differences of each monitoring data in the monitoring data sequence are obtained. The absolute value of the difference between the second difference of each monitoring data and the first difference of the corresponding monitoring data is calculated and recorded as the neighborhood difference of the corresponding monitoring data. In this embodiment, the neighborhood difference between the first and last monitoring data in the monitoring data sequence is not calculated. For the *a*-th monitoring data in the monitoring data sequence (where *a* is not equal to 1 or A), the result of subtracting the *a*-th monitoring data from the *a+1*-th monitoring data in the monitoring data sequence is the first difference of the *a*-th monitoring data. The result of subtracting the *a-1*-th monitoring data from the *a*-th monitoring data in the monitoring data sequence is the second difference of the *a*-th monitoring data. The absolute value of the difference between the second difference of the *a*-th monitoring data and the first difference of the *a*-th monitoring data is the neighborhood difference of the *a*-th monitoring data. A is the total number of monitoring data in the monitoring data sequence. Therefore, the specific calculation expression for the neighborhood difference of the *a*-th monitoring data is as follows: ,in, This refers to the neighborhood difference of the a-th monitoring data in the monitoring data sequence. For the (a+1)th monitoring data in the monitoring data sequence, For the a-th monitoring data in the monitoring data sequence, Let a be the (a-1)th monitoring data in the monitoring data sequence; then calculate the sum of the neighborhood differences of all monitoring data in the monitoring data sequence, normalize the sum of the neighborhood differences of all monitoring data, and then perform negative correlation mapping. The mapping result is recorded as the trend characterization value of the monitoring data sequence, and the specific calculation expression of the trend characterization value of the monitoring data sequence is as follows: , Where Q is the trend representation value of the monitoring data sequence, and Norm() is the normalization function, which makes the data more uniform. The value is in the range of 0 to 1, and the constant 1 is subtracted from the value in the above formula. In order to achieve The negative correlation mapping; and when The smaller the value, the more similar the magnitude of changes in the data within the monitoring data sequence, indicating a stronger or more pronounced trend in the data. Conversely, when... The smaller the value of Q, the larger the value of Q. Therefore, when the value of Q is larger, it indicates that the trend of the data in the monitoring data sequence is stronger or more obvious. When the trend of the data in the monitoring data sequence is stronger or more obvious, it means that the regularity of the monitoring data sequence is stronger. Conversely, when the value of Q is smaller, it indicates that the trend of the data in the monitoring data sequence is weaker or less obvious.

[0017] In this embodiment, the specific process of obtaining the periodic characterization value of the monitoring data sequence based on the time interval between adjacent maxima in the monitoring data sequence is as follows: First, all maxima in the monitoring data sequence are obtained, and the time sequence formed by all the obtained maxima is recorded as the maxima sequence according to the order of data acquisition; the process of obtaining the maxima in the sequence is a known technique; then, the neighborhood acquisition time interval difference value of each maxima in the maxima sequence is obtained. In this embodiment, the neighborhood acquisition time interval difference value between the first and last maxima in the maxima sequence is not calculated; and for the b-th maxima in the maxima sequence, b is not equal to 1 and B, the acquisition time interval between the (b+1)-th maxima and the b-th maxima in the maxima sequence is the first time difference value of the b-th maxima, the acquisition time interval between the b-th maxima and the (b-1)-th maxima in the maxima sequence is the second time difference value of the b-th maxima, and the absolute value of the difference between the first and second time differences of the b-th maxima is the neighborhood acquisition time interval difference value of the b-th maxima, where B is the total number of maxima in the maxima sequence. The specific calculation expression for the neighborhood acquisition time interval difference value of the b-th maxima is as follows: ,in, This represents the difference in the sampling time interval between the neighborhoods of the b-th maximum value. This represents the acquisition time of the (b+1)th maximum value in the maximum value sequence. The sampling time for the b-th maximum in the maximum value sequence. Let b be the acquisition time of the (b-1)th maximum in the maximum value sequence; then, calculate the sum of the differences in the neighborhood acquisition time intervals of all maximum values ​​in the maximum value sequence, normalize the sum of the differences in the neighborhood acquisition time intervals of all maximum values, and then perform negative correlation mapping. The mapping result is recorded as the periodic characterization value of the monitoring data sequence, and the specific calculation expression of the periodic characterization value of the monitoring data sequence is as follows: , Where P is the periodicity characteristic of the monitoring data sequence, and the normalization function in the formula is to make... The value is in the range of 0 to 1, and the constant 1 is subtracted from the value in the above formula. In order to achieve The negative correlation mapping; and when The smaller the value, the more uniform the distribution of the maxima in the monitoring data sequence, or the more regular the occurrence of the maxima, indicating a stronger or more pronounced periodicity in the monitoring data sequence. Conversely, when... The smaller the value of P, the larger the value of P. Therefore, a larger value of P indicates a stronger or more obvious periodicity in the data sequence, which in turn indicates a stronger regularity in the data sequence. Conversely, a smaller value of P indicates a weaker or less obvious periodicity in the data sequence. Additionally, there may be cases where the data sequence lacks a maximum value. In such cases, the periodicity value is not calculated, and the trend value is directly used as the regularity value.

[0018] Therefore, this embodiment obtains the regularity characterization value of the monitoring data sequence through the above process.

[0019] Step S003: Perform a first and second outer loop on the monitoring data sequence. Based on the differences in output results and weight values ​​between the second and first outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value, obtain the termination judgment index value for the second outer loop. Determine whether the termination judgment index value for the second outer loop is greater than a preset termination threshold. If it is not greater, continue performing a third outer loop on the monitoring data sequence. The differences in output results and weight values ​​between the first and second outer loops, the autocorrelation coefficient of the residual components output by the third outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value are used to obtain the termination judgment index value of the third outer loop. It is then determined whether the termination judgment index value of the third outer loop is greater than a preset termination threshold. If it is greater, the outer loop is terminated, and the result of STL decomposition of the monitoring data sequence is obtained. Based on the result of STL decomposition of the monitoring data sequence, the valve to be tested is identified for sealing leakage.

[0020] As described in step S002, this embodiment, under the premise of analyzing the strength of regularity in the original sequence, terminates the outer loop iteration by analyzing the decomposition convergence and residual component confidence results after each iteration. This avoids the problem of poor decomposition effect caused by existing outer loop iteration strategies, resulting in low accuracy and reliability of valve sealing leakage detection. Based on the above analysis, after obtaining the regularity characterization value, this embodiment needs to formulate an outer loop iteration strategy based on the weights and output results of each outer loop iteration on the monitoring data sequence, thereby obtaining the final STL decomposition result. That is, it determines whether to terminate the outer loop based on the weights and output results of each outer loop iteration on the monitoring data sequence, thus obtaining the final STL decomposition result. The STL decomposition result serves as the basis for subsequent determination of whether valve sealing leakage has occurred. The specific process of determining whether to terminate the outer loop and obtaining the final STL decomposition result based on the weights and output results of each outer loop iteration on the monitoring data sequence is as follows: First, the monitoring data sequence undergoes a first and second outer loop. Based on the differences in output and weight values ​​between the first and second outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the difference between the residual components output by the second outer loop and the cumulative distribution function of the standard normal distribution, and the regularity characteristics of the monitoring data sequence, a termination criterion value for the second outer loop is obtained. Then, it is determined whether the termination criterion value of the second outer loop is greater than a preset termination threshold. If the termination criterion value is greater than the preset termination threshold, the outer loop is terminated, and the STL decomposition result obtained after the second outer loop is used as the result of the STL decomposition of the monitoring data sequence. If the termination criterion value is not greater than the preset termination threshold, a third outer loop is performed on the monitoring data sequence. Based on the differences in output and weight values ​​between the third and second outer loops, and the autocorrelation coefficient of the residual components output by the third outer loop, the termination criterion value is determined. The autocorrelation coefficient, the difference between the residual components output by the third outer loop and the cumulative distribution function of the standard normal distribution, and the regularity characteristics of the monitoring data sequence are used to obtain the termination judgment index value of the third outer loop. The system then checks whether the termination judgment index value of the third outer loop is greater than a preset termination threshold. If the termination judgment index value of the third outer loop is greater than the preset termination threshold, the outer loop is terminated, and the STL decomposition result obtained after the third outer loop is used as the result of the STL decomposition of the monitoring data sequence. If the termination judgment index value of the third outer loop is not greater than the preset termination threshold, the system continues with the fourth outer loop, and the termination judgment index value of the fourth outer loop is obtained. This process continues until the obtained termination judgment index value is greater than the preset termination threshold, at which point the system terminates, and the result of the STL decomposition of the monitoring data sequence is obtained. In this embodiment, the outer loop refers to the outer loop in the STL decomposition. In addition, since the number of iterations of the outer loop is usually greater than 1 under normal circumstances, in order to reduce the amount of computation, this embodiment does not perform the outer loop termination judgment when the first outer loop of the monitoring data sequence ends, and only starts to perform the outer loop termination judgment from the second outer loop.

[0021] Furthermore, a larger termination criterion value indicates a higher quality of the residual components, and a greater degree of convergence in the decomposition based on the strength of regularity analysis, thus increasing the probability of the outer loop terminating. Conversely, a smaller termination criterion value indicates a lower quality of the residual components, or a smaller degree of convergence in the decomposition based on the strength of regularity analysis, thus decreasing the probability of the outer loop terminating. In practical applications, implementers need to set a preset termination threshold based on experimental statistics, the range of termination criterion values, and other actual conditions. For example, in this embodiment, the termination criterion value can be set to 0.7. That is, when the termination criterion value is greater than the preset termination threshold, it indicates a higher quality of the residual components, and the convergence requirement has been met based on the strength of regularity analysis of the monitored data sequence, at which point the iteration of the outer loop can be stopped.

[0022] Additionally, it should be noted that the method for obtaining the termination judgment index value of the 3rd or 4th outer loop is the same as the method for obtaining the termination judgment index value of the 2nd outer loop, so it will not be described in detail again; moreover, the normalized value range in this embodiment is 0 to 1.

[0023] In this embodiment, the specific process for obtaining the termination judgment index value of the second outer loop based on the differences in output results and weight values ​​between the first and second outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the difference between the residual components output by the second outer loop and the cumulative distribution function of the standard normal distribution, and the regularity characterization value of the monitoring data sequence is as follows: First, obtain the weight values ​​for the first and second outer loops of the monitoring data sequence. The weight value for the first outer loop refers to the weight value of each monitoring data point in the monitoring data sequence during the first outer loop, or the weight value of each monitoring data point in the monitoring data sequence during the first outer loop. The weight value for the second outer loop refers to the weight value of each monitoring data point in the monitoring data sequence during the second outer loop, or the weight value of the monitoring data sequence during the second outer loop. The weight values ​​of each monitoring data point are determined, and the specific process for obtaining the weight values ​​of each round of the outer loop for the monitoring data sequence is a known technique. Then, the output results of the first outer loop and the second outer loop for the monitoring data sequence are obtained, and the output results of each round of the outer loop include residual components, trend components, and seasonal components. Then, based on the difference in output results and weight values ​​between the second and first outer loops, the decomposition convergence of the second outer loop is obtained. The decomposition convergence of the second outer loop is an important basis for determining whether the second outer loop should be terminated. Next, the autocorrelation coefficient of the residual components output from the second external loop of the monitoring data sequence is obtained, as are the cumulative distribution function of the residual components output from the second external loop and the cumulative distribution function of the standard normal distribution. Then, based on the autocorrelation coefficient of the residual components output from the second external loop and the difference between the cumulative distribution function of the residual components output from the second external loop and the cumulative distribution function of the standard normal distribution, the confidence level of the residual components of the second external loop is obtained. The confidence level of the residual components of the second external loop is also an important basis for determining whether the second external loop should be terminated. The autocorrelation coefficient of the residual components output from the second external loop refers to the autocorrelation coefficient calculated on the residual component sequence output from the second external loop, and the specific calculation process is well-known. The specific calculation process of the cumulative distribution function of the residual components output from the second external loop and the cumulative distribution function of the standard normal distribution are well-known techniques, and the cumulative distribution function is also a probability distribution function. After obtaining the decomposition convergence and residual component confidence of the second outer loop, the termination judgment index value of the second outer loop is obtained based on the regularity characterization value of the monitoring data sequence and the decomposition convergence and residual component confidence of the second outer loop.In addition, the residual component sequence, trend component sequence, and seasonal component sequence output by performing the i-th outer loop on the monitoring data sequence refer to the residual component sequence composed of the residual components of each monitoring data in the monitoring data sequence output by performing the i-th outer loop on the monitoring data sequence, the trend component sequence composed of the trend components of each monitoring data in the monitoring data sequence output by performing the i-th outer loop on the monitoring data sequence, and the seasonal component sequence composed of the seasonal components of each monitoring data in the monitoring data sequence output by performing the i-th outer loop on the monitoring data sequence.

[0024] In this embodiment, the specific process of obtaining the decomposition convergence of the second outer loop based on the differences in output results and weight values ​​between the second and first outer loops is as follows: First, based on the difference between the weight values ​​of the second outer loop and the weight values ​​of the first outer loop for each monitoring data point in the monitoring data sequence, the weight change representation value between the second and first outer loops is obtained. Then, the normalized DTW distance between the trend component sequence output by the second outer loop and the trend component sequence output by the first outer loop is calculated and recorded as the trend difference representation value corresponding to the second outer loop. Finally, the seasonal component sequence output by the second outer loop and the seasonal component sequence output by the first outer loop are calculated. The normalized DTW distance between the seasonal component sequences output by the second outer loop is recorded as the seasonal difference representation value corresponding to the second outer loop. Then, the mean of the trend difference representation value and the seasonal difference representation value corresponding to the second outer loop is calculated and used as the output difference representation value between the second and first outer loops. Next, the mean of the output difference representation value between the second and first outer loops and the weight change representation value between the second and first outer loops are calculated and negatively correlated. The mapping result is used as the decomposition convergence of the second outer loop, and the specific expression for calculating the decomposition convergence of the second outer loop is as follows: Where R is the decomposition convergence of the second outer loop, U is the weight change representation value between the second and first outer loops, and V is the output difference representation value between the second and first outer loops. Since the values ​​of U and V in this embodiment are both from 0 to 1, this embodiment can use a constant 1 minus the mean of U and V as the result of negatively correlated mapping of the mean of U and V; as another implementation, it is also possible to use... As the convergence factor of the second outer loop, the exponential function exp() with base e is used to perform a negative correlation mapping on the mean values ​​of U and V. In addition, the smaller U and V are, the more convergent the second outer loop decomposition tends to be. The smaller U and V are, the larger R is. Therefore, when R is larger, it indicates that the second outer loop decomposition tends to be more convergent, and conversely, when R is smaller, it indicates that the second outer loop decomposition tends to be less convergent.

[0025] In this embodiment, the specific process for obtaining the weight change representation value between the second outer loop and the first outer loop is as follows: First, the weight value of the first outer loop for each monitoring data in the monitoring data sequence is recorded as the first weight value of each monitoring data in the monitoring data sequence, and the weight value of the second outer loop for each monitoring data in the monitoring data sequence is recorded as the second weight value of each monitoring data in the monitoring data sequence; then, based on the absolute value of the difference between the first weight value and the second weight value of the same monitoring data, the weight difference sequence between the second outer loop and the first outer loop is obtained, and the normalized result after accumulating all weight differences in the weight difference sequence between the second outer loop and the first outer loop is used as the weight change representation value between the second outer loop and the first outer loop. Here, the normalization function Norm() is also used to achieve normalization, and the f-th weight difference in the weight difference sequence between the second outer loop and the first outer loop is the absolute value of the difference between the second weight value and the first weight value of the f-th monitoring data in the monitoring data sequence. The smaller the weight change value between the second outer loop and the first outer loop, the more stable the outlier identification result is, and the less it needs to be adjusted. This also indicates that the decomposition of the second outer loop is more convergent. Conversely, the larger the weight change value between the second outer loop and the first outer loop, the less convergent the decomposition of the second outer loop is.

[0026] In this embodiment, the specific expression for calculating the difference in output results between the second outer loop and the first outer loop is as follows: , Where V represents the difference in output between the first and second outer loops, Norm() is the normalization function, T1 is the seasonal component sequence output by the first outer loop of the monitoring data sequence, and T2 is the seasonal component sequence output by the second outer loop of the monitoring data sequence. The DTW distance is defined as the seasonal component sequence output from the first outer loop of the monitoring data sequence and the seasonal component sequence output from the second outer loop of the monitoring data sequence. S1 represents the trend component sequence output from the first outer loop of the monitoring data sequence, and S2 represents the trend component sequence output from the second outer loop of the monitoring data sequence. The DTW distance between the trend component sequence output from the first outer loop of the monitoring data sequence and the trend component sequence output from the second outer loop of the monitoring data sequence is a known technique. Furthermore, when... and The smaller V is, the more similar the trend component sequences output by the second external cycle and the first external cycle are, and the more similar the seasonal component sequences output by the second external cycle and the first external cycle are. It also indicates that the trend and seasonal components output by the second external cycle tend to converge, or the decomposition of the second external cycle tends to converge. Conversely, when V is larger, it indicates that the trend and seasonal components output by the second external cycle are less likely to converge, or the decomposition of the second external cycle tends to converge.

[0027] In this embodiment, the specific process for obtaining the confidence level of the residual components of the second external loop based on the autocorrelation coefficient of the residual components output by the second external loop and the difference between the cumulative distribution function of the residual components output by the second external loop and the cumulative distribution function of the standard normal distribution is as follows: First, the absolute value of the difference between the cumulative distribution function of the residual components output by the second external loop and the cumulative distribution function of the standard normal distribution is calculated as the integral result over the entire domain. The integral result is then normalized and negatively correlated, and the mapping result is used as the first residual quality characterization value of the second external loop. Then, the autocorrelation coefficient of the residual components output by the second external loop is obtained, normalized, and negatively correlated, and used as the second residual quality characterization value of the second external loop. The mean of the first and second residual quality characterization values ​​is used as the confidence level of the residual components of the second external loop. The specific calculation expression for the confidence level of the residual components of the second external loop is as follows: , in, Let G be the confidence level of the residual components of the second outer loop, G be the integral of the absolute value of the difference between the cumulative distribution function of the residual components output by the second outer loop and the cumulative distribution function of the standard normal distribution over the entire domain, and ACF be the autocorrelation coefficient of the residual components output by the second outer loop; constant 1 minus or The purpose is to perform negative correlation mapping, while other real-time methods can also use a negative exponential function with a base of constant e to perform negative correlation mapping. This is to ensure that the value range is from 0 to 1.

[0028] The ultimate goal of STL decomposition is to obtain a residual useful for leak detection. Ideally, the residual should approximate the standard noise distribution when there is no valve seal leakage and contain clear anomalous components when there is valve seal leakage. In other words, ideally, after STL successfully separates the trend and periodic components, the residual will only contain unstructured information. That is, the more the interference signal is modeled as white noise that follows a normal distribution, the higher the quality of the residual component output by the decomposition. Or, the closer the output residual component is to the standard normal distribution, the higher the quality of the output residual component. When G is smaller, it indicates that the residual component output by the second outer loop is closer to the standard normal distribution, which also indicates that the quality of the output residual component is higher. Furthermore, to further improve the accuracy of residual component quality analysis, this embodiment also needs to consider the autocorrelation coefficient of the output residual components. The smaller the autocorrelation coefficient of the output residual components, the more it indicates that the model has fully extracted the trends and seasonality patterns in the time series, and the remaining fluctuations are close to random noise or interference signals, meeting the independence assumption that ideal residuals should possess. Therefore, the smaller the ACF, the higher the quality of the decomposed output residual components; since when G and ACF are smaller, The larger, therefore when A larger value indicates a higher quality residual component output by the second outer loop. A higher quality residual component means a greater probability that the STL algorithm can effectively separate the trend and seasonal components of the monitoring data sequence. It also means a greater probability that the remaining residual contains only random noise or interference signals. In other words, a lower probability that the real high-frequency pulses caused by leakage are suppressed as noise, and a lower probability that the residual contains interference signals such as process fluctuations, operational disturbances, or noise. Conversely, a smaller value indicates a lower quality residual. The smaller the value, the lower the quality of the residual component output by the second outer loop.

[0029] In this embodiment, the specific process of obtaining the termination judgment index value of the second outer loop based on the regularity characterization value of the monitoring data sequence and the decomposition convergence degree and residual component confidence degree of the second outer loop is as follows: First, based on the decomposition convergence of the second outer loop and the regularity characteristics of the monitoring data sequence, the component extraction confidence of the second outer loop is obtained. The process for obtaining the component extraction confidence of the second outer loop is as follows: the regularity characteristics are weighted using preset influence weights, then negative correlation mapping is performed, and this is recorded as the mapped regularity characteristics. The result of subtracting the mapped regularity characteristics from the decomposition convergence of the second outer loop and then performing linear normalization is used as the component extraction confidence of the second outer loop. The specific calculation expression for the component extraction confidence of the second outer loop is as follows: , Where K is the confidence level of component extraction in the second outer loop, and M() is the linear normalization function, which is used to normalize the data. Linear normalization is performed, and the normalized values ​​range from 0 to 1. R is the convergence degree of the decomposition in the second outer loop. To preset the influence weight, To monitor the regularity of data sequences.

[0030] Because the stronger the regularity of the monitored data sequence, that is... When the frequency is high, it is easy to over-decompose and suppress some of the real high-frequency pulses caused by leakage as noise. In other words, strong regularity is the main reason why real high-frequency pulses caused by leakage are suppressed as noise. To avoid this situation, this embodiment requires lower convergence requirements for highly regular monitoring data sequences to avoid unnecessary iterations. Conversely, the weaker the regularity of the monitoring data sequence, the lower the convergence requirement. The smaller the value, the more likely it is to introduce interference signals into the residuals. In other words, weak regularity is the main reason why process fluctuations, operational disturbances, or noise may be mixed into the residuals. To avoid this, this embodiment requires a higher convergence requirement for monitoring data sequences with weak regularity. Therefore, based on the above description, this embodiment needs to limit the decomposition convergence to the regularity characteristic value. That is, lower the convergence requirement for strong regularity and increase the convergence requirement for weak regularity. This embodiment achieves this through the formula above... To achieve the above requirements, This indicates the degree of difference between the convergence of the second outer loop and the required convergence index. A larger K value indicates a higher convergence than the required index, meaning the convergence is more closely matched to the strength of the regularity of the monitored data sequence, resulting in better convergence and more reliable component extraction in the second outer loop. Furthermore... The document specifies the required convergence index for the regularity representation value of the monitored data sequence. This convergence index is limited by the strength of the regularity of the monitored data sequence. The higher the regularity value of the monitored data sequence, that is, the stronger the regularity of the monitored data sequence, the lower the convergence requirement, thereby preventing excessive decomposition from suppressing some real high-frequency pulses caused by leakage as noise. The smaller the regularity value of the monitored data sequence, the higher the convergence requirement, thereby reducing the probability of interference signals such as process fluctuations, operational disturbances, or noise being mixed into the residuals. In this embodiment, the preset influence weight represents the impact of the strength of the regularity of the monitored data sequence on component extraction. The implementer needs to set the value of the preset influence weight according to the degree of influence of the regularity on component extraction, and the preset influence weight is set according to the principle that the smaller or larger the preset influence weight, the greater the influence on component extraction. For example, the preset influence weight can be set to 0.3. That is, if the preset influence weight is too large, it may over-amplify the influence of the regularity of the monitored data sequence on component extraction, resulting in too low a convergence requirement, which may lead to incomplete decomposition. If the preset influence weight is too small, it will lead to a high convergence requirement, resulting in over-convergence of the decomposition.

[0031] After obtaining the confidence scores of the component extraction in the second outer loop, the termination criterion value for the second outer loop is obtained based on the confidence scores of the component extraction and residual components of the second outer loop. In this embodiment, the termination criterion value for the second outer loop is obtained by adding the confidence scores of the component extraction and residual components of the second outer loop and then normalizing the result. This normalization is achieved using the normalization function Norm(). The larger the termination criterion value for the second outer loop, the higher the quality of the residual components and the better the convergence effect based on the regularity analysis results, thus increasing the probability of terminating the outer loop. Conversely, the smaller the termination criterion value for the second outer loop, the lower the quality of the residual components or the worse the convergence effect based on the regularity analysis results, thus decreasing the probability of terminating the outer loop.

[0032] Furthermore, since the confidence level of residual components reflects statistical characteristics rather than the component extraction state, and the outer loop iteration of STL aims to converge trend and seasonal components, relying solely on the quality of residual components cannot determine whether the iteration state truly satisfies the extraction of trend and seasonal components. In other words, the confidence level of residual components alone cannot indicate whether the iteration state truly satisfies the extraction of trend and seasonal components. For example, STL typically exhibits relatively stable signals such as pressure, flow, and temperature, which can easily lead to over-iteration. This causes abnormal signals (such as sudden pulses) to be absorbed into the trend or seasonal components. In this case, although the residual components show high quality, the anomalies have been excessively smoothed and lost. Therefore, the quality of residual components cannot be used as a direct basis for determining whether iteration can stop. A secondary judgment is needed regarding the extraction of trend and seasonal components, requiring a comprehensive assessment considering the inability to definitively determine whether the iteration state truly satisfies the extraction of trend and seasonal components. Moreover, the strength of the data's inherent regularity directly affects the convergence speed of the decomposition. To avoid over-decomposition and under-decomposition due to varying degrees of regularity in the sequence, this embodiment controls the convergence requirements of STL for trend and seasonal components based on the strength of regularity. Specifically, for data with strong regularity, the convergence requirements should be reduced to prevent the suppression of real high-frequency pulses caused by partial leakage as noise. Conversely, for data with weak regularity, the convergence requirements should be increased to remove noise or interference such as process fluctuations or trend drift. Therefore, this embodiment combines the confidence scores of residual components and regularity characteristics to obtain the component extraction confidence scores. After obtaining the component extraction confidence scores and residual component confidence scores, a termination judgment index value will be obtained through fusion analysis. This termination judgment index value is the basis for determining whether to perform outer loop iteration. Furthermore, based on the termination judgment index value, the probability of over-decomposition and under-decomposition can be minimized, resulting in higher decomposition quality and more reliable decomposition results.

[0033] Based on the above description, this embodiment will next perform sealing leakage detection and identification on the valve to be tested based on the results of STL decomposition of the monitoring data sequence. Since the results of STL decomposition of the monitoring data sequence are known, sealing leakage detection and identification on the valve to be tested is a well-known technique, and therefore will not be described in detail here. Furthermore, this embodiment analyzes the strength of regularity in the original monitoring data sequence and combines an external loop iteration strategy that considers the decomposition convergence and residual component confidence of each external loop iteration. This minimizes the probability that real high-frequency pulses caused by leakage are suppressed as noise or that interference signals are mixed into the residuals after the external loop iteration, thereby improving the effectiveness of STL decomposition of the monitoring data sequence and increasing the accuracy and reliability of subsequent valve sealing leakage detection based on the decomposition results.

[0034] Thus, this embodiment completes the detection and identification of leaks in the valve seal to be tested.

[0035] In summary, this embodiment first obtains a regularity characterization value based on the differences between adjacent data and the time interval between adjacent maxima in the monitoring data sequence. Then, it performs a first and second outer loop on the monitoring data sequence. Based on the differences in output results and weight values ​​between the second and first outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the differences between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value, it obtains a termination judgment index value for the second outer loop. Then, it determines whether the termination judgment index value for the second outer loop is greater than a preset termination threshold. If it is not greater, the process continues. The monitoring data sequence undergoes a third external loop. Based on the differences in output results and weight values ​​between the second and third external loops, the autocorrelation coefficient of the residual components output by the third external loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characteristics, a termination judgment index value for the third external loop is obtained. The system then checks whether the termination judgment index value for the third external loop exceeds a preset termination threshold. If it does, the external loop is terminated, and the STL decomposition result of the monitoring data sequence is obtained. Finally, based on the STL decomposition result of the monitoring data sequence, the valve to be tested is identified for sealing leakage. Furthermore, this embodiment uses the termination judgment index value as the condition for terminating the external loop, which ensures a higher quality and more reliable STL decomposition, thereby improving the accuracy and reliability of valve sealing leakage detection.

[0036] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A valve sealing leakage detection method based on sensor monitoring data, characterized in that, The method includes the following steps: The monitoring data sequence of the valve to be tested is acquired using sensors; Based on the differences between adjacent data in the monitoring data sequence and the time interval between adjacent maxima, regularity characterization values ​​are obtained; The monitoring data sequence undergoes a first and second outer loop. Based on the differences in output results and weight values ​​between the second and first outer loops, the autocorrelation coefficient of the residual components output by the second outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value, a termination judgment index value for the second outer loop is obtained. It is then determined whether the termination judgment index value for the second outer loop is greater than a preset termination threshold. If it is not greater, a third outer loop is performed on the monitoring data sequence. The results of the third outer loop are then compared with... The differences in output results and weight values ​​between the second and third outer loops, the autocorrelation coefficient of the residual components output by the third outer loop, the difference between the output residual components and the cumulative distribution function of the standard normal distribution, and the regularity characterization value are used to obtain the termination judgment index value of the third outer loop. It is then determined whether the termination judgment index value of the third outer loop is greater than the preset termination threshold. If it is greater, the outer loop is determined to be terminated, and the result of STL decomposition of the monitoring data sequence is obtained. Based on the result of STL decomposition of the monitoring data sequence, the valve to be tested is identified for sealing leakage.

2. The valve sealing leakage detection method based on sensor monitoring data as described in claim 1, characterized in that, Methods for obtaining regularity representation values ​​include: Based on the differences between adjacent data in the monitoring data sequence, a trend characterization value is obtained, and based on the time interval between adjacent maxima in the monitoring data sequence, a periodic characterization value is obtained. The average of the trend characterization value and the periodic characterization value is used as the regularity characterization value.

3. The valve sealing leakage detection method based on sensor monitoring data as described in claim 2, characterized in that, Methods for obtaining trend-related characterization values ​​include: For the a-th monitoring data in the monitoring data sequence, where a is not equal to 1 or A, the result of subtracting the a-th monitoring data from the (a+1)-th monitoring data in the monitoring data sequence is recorded as the first difference of the a-th monitoring data. The result of subtracting the (a-1)-th monitoring data in the monitoring data sequence from the a-th monitoring data is recorded as the second difference of the a-th monitoring data. The absolute value of the difference between the second difference and the first difference is recorded as the neighborhood difference of the a-th monitoring data. A is the total number of monitoring data in the monitoring data sequence. The result of normalizing the sum of the neighborhood differences of all monitoring data in the monitoring data sequence and then performing negative correlation mapping is recorded as the trend characterization value of the monitoring data sequence.

4. The valve sealing leakage detection method based on sensor monitoring data as described in claim 2, characterized in that, Methods for obtaining periodic representation values ​​include: The time series sequence consisting of all the maximum values ​​in the monitoring data sequence is denoted as the maximum value sequence; For the b-th maximum value in the maximum value sequence, where b is not equal to 1 and B, the acquisition time interval between the (b+1)-th maximum value and the b-th maximum value in the maximum value sequence is recorded as the first time difference of the b-th maximum value, the acquisition time interval between the b-th maximum value and the (b-1)-th maximum value in the maximum value sequence is recorded as the second time difference of the b-th maximum value, and the absolute value of the difference between the first time difference value and the second time difference value is recorded as the neighborhood acquisition time interval difference value of the b-th maximum value, where B is the total number of maximum values ​​in the maximum value sequence; The result of normalizing the sum of the differences in the neighborhood acquisition time intervals of all maxima in the maximum value sequence and then performing negative correlation mapping is recorded as the periodic characterization value of the monitoring data sequence.

5. The valve sealing leakage detection method based on sensor monitoring data as described in claim 1, characterized in that, The method for obtaining the termination judgment index value includes: The decomposition convergence of the second outer loop is obtained based on the difference in output results and weight values ​​between the second outer loop and the first outer loop. The confidence level of the residual components of the second outer loop is obtained based on the autocorrelation coefficient of the residual components output by the second outer loop and the difference between the output residual components and the cumulative distribution function of the standard normal distribution. Based on the regularity characterization value and the decomposition convergence and residual component confidence of the second outer loop, the termination judgment index value of the second outer loop is obtained; the method for obtaining the termination judgment index value of the third outer loop is the same as the method for obtaining the termination judgment index value of the second outer loop.

6. The signal data acquisition method for a battery central control motherboard as described in claim 5, characterized in that, The method for obtaining the convergence of the second outer loop decomposition includes: Based on the difference between the weight values ​​of the second outer loop and the weight values ​​of the first outer loop for each monitoring data in the monitoring data sequence, the weight change characterization value between the second outer loop and the first outer loop is obtained. The normalized DTW distance between the trend component sequence output by the second outer loop of the monitoring data sequence and the trend component sequence output by the first outer loop of the monitoring data sequence is calculated and recorded as the trend difference characterization value. The normalized DTW distance between the seasonal component sequence output by the second outer loop of the monitoring data sequence and the seasonal component sequence output by the first outer loop of the monitoring data sequence is calculated and recorded as the seasonal difference characterization value. The mean of the trend difference characterization value and the seasonal difference characterization value is used as the output result difference characterization value between the second outer loop and the first outer loop. The result of negatively mapping the mean of the output difference representation value and the weight change representation value is used as the decomposition convergence of the second outer loop.

7. The valve sealing leakage detection method based on sensor monitoring data as described in claim 6, characterized in that, The method for obtaining the weight change representation value between the second outer loop and the first outer loop includes: The weight value of the first outer loop for each monitoring data in the monitoring data sequence is recorded as the first weight value of each monitoring data in the monitoring data sequence, and the weight value of the second outer loop for each monitoring data in the monitoring data sequence is recorded as the second weight value of each monitoring data in the monitoring data sequence. The weight difference sequence between the second outer loop and the first outer loop is obtained, and the normalized result of accumulating all weight differences in the weight difference sequence is used as the weight change characterization value between the second outer loop and the first outer loop. The f-th weight difference in the weight difference sequence is the absolute value of the difference between the second weight value and the first weight value of the f-th monitoring data in the monitoring data sequence.

8. The valve sealing leakage detection method based on sensor monitoring data as described in claim 1, characterized in that, The method for obtaining the confidence level of the residual components in the second outer loop includes: The result of normalizing the integral of the absolute value of the difference between the cumulative distribution function of the residual component output by the second outer loop and the cumulative distribution function of the standard normal distribution over the entire domain, and then performing a negative correlation mapping, is used as the first residual quality characterization value of the second outer loop; the result of normalizing the autocorrelation coefficient of the residual component output by the second outer loop and then performing a negative correlation mapping is used as the second residual quality characterization value of the second outer loop; and the mean of the first residual quality characterization value and the second residual quality characterization value is used as the confidence level of the residual component of the second outer loop.

9. The valve sealing leakage detection method based on sensor monitoring data as described in claim 5, characterized in that, The method for obtaining the termination judgment index value of the second outer loop based on the regularity characterization value and the decomposition convergence and residual component confidence of the second outer loop includes: Based on the decomposition convergence of the second outer loop and the regularity characterization value, the component extraction confidence of the second outer loop is obtained. The result of normalizing the sum of the confidence scores of the residual components and the component extraction confidence scores of the second outer loop is used as the termination criterion value of the second outer loop.

10. The valve sealing leakage detection method based on sensor monitoring data as described in claim 9, characterized in that, The method for obtaining the confidence score of the component extraction in the second outer loop includes: The regularity representation value is weighted using a preset influence weight and then subjected to negative correlation mapping, and recorded as the mapped regularity representation value. The result of normalizing the decomposition convergence of the second outer loop by subtracting the mapped regularity representation value is used as the component extraction confidence of the second outer loop.

Images