PDS Ball Valve Condition Monitoring Method and System Based on Multi-Sensor Fusion

Abstract

This invention relates to the field of ball valve condition monitoring, and particularly to a PDS ball valve condition monitoring method and system based on multi-sensor fusion. The method includes: acquiring and preprocessing ball valve operating data through multiple sensors; fusing an initial matrix and performing PCA decomposition; calculating the correlation between abnormal feature sequences and principal component score sequences; quantifying the anomaly sensitivity of each principal component based on the average anomaly level of the abnormal feature sequences; calculating and ranking the importance of each principal component based on the anomaly sensitivity and the variance contribution rate of each principal component; and denoising the reconstructed data by selecting the optimal number of principal components to retain. The reconstruction error is calculated on the denoised operating data, and the improved LOF algorithm is used to calculate local anomaly factors to determine the ball valve operating status. This invention achieves efficient denoising by optimizing principal component selection, and enhances the reliability of anomaly detection by combining the improved LOF algorithm with reconstruction error, thereby improving the real-time performance, response speed, and stability of the monitoring system.

Description

Technical Field

[0001] This invention relates to the field of ball valve condition monitoring. In particular, it relates to a PDS ball valve condition monitoring method and system based on multi-sensor fusion. Background Technology

[0002] Ball valves are devices used in industrial applications, and their performance and condition are crucial to the stability and safety of the entire production process. The importance of ball valve condition monitoring lies in its ability to promptly detect potential faults, preventing production interruptions and economic losses due to equipment failure. Effective condition monitoring allows for advance maintenance planning, extending equipment lifespan while simultaneously improving production efficiency and product quality.

[0003] Existing PDS ball valve condition monitoring methods typically employ single-sensor data acquisition, such as vibration or pressure data, to monitor the valve's condition and determine if any abnormalities have occurred. However, industrial environments are complex and subject to significant noise interference. This noise can cause deviations in the data acquired by single sensors, leading to misjudgments or missed diagnoses during data analysis. A misjudgment occurs when a normally operating state is identified as an abnormal state, resulting in unnecessary maintenance work, wasted resources, and reduced production efficiency. A missed diagnose occurs when potential faults are not detected in time, allowing them to persist and potentially worsen, ultimately leading to more serious equipment damage. Summary of the Invention

[0004] To address the problem that in existing PDS ball valve condition monitoring methods, data collected by a single sensor is easily affected by industrial environmental noise, leading to misjudgments or omissions in data analysis, resulting in unnecessary maintenance or more serious equipment damage and production accidents, this invention provides solutions in the following aspects.

[0005] In the first aspect, the PDS ball valve status monitoring method based on multi-sensor fusion includes: acquiring ball valve operation data based on multiple sensors and preprocessing it; fusing the preprocessed operation data into an initial matrix and performing PCA decomposition to obtain principal components and score sequences; calculating the correlation between the abnormal feature sequences of the operation data sequences of each sensor and the score sequences of the principal components; quantifying the abnormality sensitivity of each principal component based on the average abnormality level of the abnormal feature sequences; calculating the importance of each principal component based on the abnormality sensitivity and the variance contribution rate of each principal component; sorting the principal components according to their importance; traversing different numbers of retained principal components and calculating the denoising effect of different numbers of principal components; selecting the principal component number corresponding to the largest denoising effect; reconstructing the operation data to complete denoising; calculating the reconstruction error of each sampling point for the denoised operation data; improving the distance between sampling points in the LOF algorithm based on the reconstruction error to obtain a weighted distance; calculating the local anomaly factor of each sampling point based on the weighted distance; and judging the operation status of the PDS ball valve based on the local anomaly factor.

[0006] Preferably, the calculation method of the abnormal feature sequence includes:

[0007] In the initial matrix, each row represents a sample and each column represents the data dimension of a sensor. For the operating data sequence of each sensor in the initial matrix, the mean of the operating data sequence of each sensor is calculated, the ratio of each sampling point data to the mean is calculated, the absolute value of 1 minus the ratio is taken as the relative deviation of each sampling point, the information entropy value of the operating data sequence of each sensor is calculated, and the information entropy value is multiplied by the relative deviation to obtain the abnormal features of the data of each sampling point, thus constructing the abnormal feature sequence of the operating data sequence.

[0008] Preferably, the method for obtaining the score sequence includes the following steps:

[0009] Calculate the mean of each column of the initial matrix, subtract the mean from each feature to obtain the decentralized initial matrix, calculate the covariance initial matrix of the decentralized initial matrix, perform eigenvalue decomposition on the covariance initial matrix to obtain eigenvalues ​​and corresponding eigenvectors, arrange all eigenvectors into the initial matrix, and project the decentralized initial matrix onto the eigenvector initial matrix to obtain the initial matrix of principal components, where each column of the initial matrix of principal components represents the score sequence of each principal component.

[0010] Preferably, the method for calculating the abnormal sensitivity of the principal component includes:

[0011] Using the score sequence of any principal component as the target sequence, calculate the Pearson correlation coefficient between the target sequence and each abnormal feature sequence. Take the absolute value of the Pearson correlation coefficient and combine it with the mean of each abnormal feature to take the absolute value of the Pearson correlation coefficient for weighted summation. Calculate the sum of the means of all abnormal features for normalization processing to obtain the abnormality sensitivity of the principal component.

[0012] Preferably, the method for calculating the importance of each principal component includes:

[0013] The proportion of each principal component's eigenvalue to the sum of all eigenvalues ​​is used as the variance contribution rate, and the product of the variance contribution rate and the anomaly sensitivity is used as the importance of each principal component.

[0014] Preferably, the calculation method for the denoising effect of different principal component numbers includes:

[0015] Extract the score sequence of retained principal components from the initial matrix, construct the reconstruction matrix, calculate the reconstruction error of each sampling point in the initial matrix and the reconstruction matrix, calculate the ratio between the sum of anomaly sensitivities corresponding to the number of retained principal components and the sum of anomaly sensitivities corresponding to the number of all principal components, use the reconstruction error as the influence weight, and process the reconstruction error using a normalization function to obtain the denoising effect corresponding to different numbers of retained principal components.

[0016] Preferably, the weighted distance is obtained in the following ways:

[0017] Calculate the Euclidean distance between any two sampling points in the denoised running data, calculate the reconstruction error for each sampling point, select the maximum value of the reconstruction error, and weight the Euclidean distance by the ratio between the maximum value of the reconstruction error and the mean of the reconstruction errors of all sampling points to obtain the improved weighted distance.

[0018] Secondly, a PDS ball valve condition monitoring system based on multi-sensor fusion includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned PDS ball valve condition monitoring method based on multi-sensor fusion is implemented.

[0019] The present invention has the following effects:

[0020] 1. This invention calculates and sorts the importance of each principal component, then iterates through different numbers of retained principal components to select the number of principal components with the best denoising effect, thus reconstructing and denoising the operating data. This removes noise interference from the data, retains the most valuable information for anomaly detection, and reduces the complexity and computational load of data processing. It improves data processing efficiency, enabling faster analysis and judgment of the ball valve's operating status, providing timely and accurate results, enhancing the real-time performance and response speed of the monitoring system, and better meeting the real-time requirements for equipment status monitoring in industrial production.

[0021] 2. This invention calculates the reconstruction error of each sampling point and improves the distance between sampling points in the LOF algorithm by incorporating the reconstruction error, obtaining a weighted distance. Based on the weighted distance, the local anomaly factor of each sampling point is calculated to determine the operating status of the ball valve. This enhances the reliability of anomaly detection, enabling the monitoring system to more stably and accurately determine the operating status of the ball valve when facing different types of faults and complex operating environments. It reduces misjudgments and missed judgments caused by factors such as noise interference, and improves the adaptability and stability of the monitoring system. Attached Figure Description

[0022] Figure 1 This is a flowchart of steps S1-S4 in the PDS ball valve condition monitoring method based on multi-sensor fusion according to an embodiment of the present invention.

[0023] Figure 2 This is a structural block diagram of the PDS ball valve condition monitoring system based on multi-sensor fusion according to an embodiment of the present invention. Detailed Implementation

[0024] 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 some embodiments of the present invention, but not all embodiments.

[0025] Reference Figure 1 The PDS ball valve condition monitoring method based on multi-sensor fusion includes steps S1-S4, as follows:

[0026] S1: Acquire ball valve operating data based on multiple sensors and perform preprocessing.

[0027] Multiple types of sensors, including vibration sensors, acoustic emission (AE) sensors, pressure sensors, and temperature sensors, are deployed at key locations on the ball valve of the PDS (Products Discharge System) to comprehensively monitor the operating status of the ball valve.

[0028] The PDS ball valve's vibration, acoustic emission, pressure, and temperature signals are collected in real time via sensors at set intervals (e.g., 1 second) during operation. The collection interval can be adjusted according to actual monitoring needs and equipment operating characteristics to ensure the timeliness and completeness of the data.

[0029] For the collected data from different dimensions, precise alignment is performed based on their timestamps. This ensures that all data points have a consistent time base across the time series, enabling comparison and analysis at the same points in time. For data with inconsistent timestamps, interpolation or other time alignment techniques are used to keep the data synchronized across the time dimension.

[0030] Normalize the data for each dimension separately to eliminate the influence of different units and ranges. Employ methods such as min-max normalization and Z-score normalization, selecting the appropriate normalization strategy based on the data characteristics and analytical needs.

[0031] In industrial environments, the data collected by various sensors during the operation of PDS ball valves often contains a large amount of noise, which severely interferes with the accurate identification of the ball valve's status. To effectively remove noise, existing technologies typically employ PCA (Principal Component Analysis) decomposition and reconstruction methods for data denoising. However, traditional PCA methods primarily seek the set of projection directions that maximizes the variance after projection during optimization, and select principal components based on the variance contribution rate (eigenvalues), thus retaining the directions with the greatest variation in the data. While this approach can capture the main information in the data, it has certain limitations. Because the characteristics of data in different dimensions are not entirely the same, some low-variance directions may contain important anomalies, such as weak impact signals or local leakage signals. These are often ignored in traditional PCA methods, leading to unsatisfactory denoising results and affecting the accuracy of subsequent anomaly detection.

[0032] To overcome the shortcomings of traditional PCA methods, this paper, based on the variance-based ranking of principal components, further introduces anomalous features from the operational data within the principal components. This enhances the sensitivity to fault signals, thereby improving denoising performance and the accuracy of subsequent anomaly detection. Before implementing this scheme, the degree of anomalous features in the operational data needs to be quantified. The specific steps are as follows:

[0033] S2: After preprocessing the running data, fuse the initial matrix and perform PCA decomposition to obtain principal components and score sequences. Calculate the correlation between the abnormal feature sequences of the running data sequences of each sensor and the score sequences of the principal components. Combine the average abnormality level of the abnormal feature sequences to quantify the abnormality sensitivity of each principal component. Based on the abnormality sensitivity and the variance contribution rate of each principal component, calculate the importance of each principal component.

[0034] In the initial matrix, each row represents a sample and each column represents the data dimension of a sensor. For the operating data sequence of each sensor in the initial matrix, the mean of the operating data sequence of each sensor is calculated, the ratio of each sampling point data to the mean is calculated, the absolute value of 1 minus the ratio is taken as the relative deviation of each sampling point, the information entropy value of the operating data sequence of each sensor is calculated, and the information entropy value is multiplied by the relative deviation to obtain the abnormal features of the data of each sampling point, thus constructing the abnormal feature sequence of the operating data sequence.

[0035] Specifically, the abnormal feature sequence satisfies the following relationship:

[0036] ;

[0037] In the formula, Indicates the first The data collected by the first sensor Abnormal characteristics of data from each sampling point Indicates the first Operational data sequence collected by each sensor The information entropy value reflects the complexity of the data distribution in this operational data sequence. A higher information entropy value indicates greater fluctuation in the operational data during the operation of the PDS ball valve, and a higher probability of anomalies. Represents the initial matrix The Middle The data collected by the first sensor Data from each sampling point, Represents the initial matrix The Middle Sensor operating data sequence Mean and ratio of all data in The closer it is to 1, the more it indicates the first... The data collected by the first sensor The closer the data from each sampling point is to the average level, the less likely an anomaly is to occur; conversely, the further the ratio deviates from 1, the less likely an anomaly is to occur. The data collected by the first sensor The further the data from each sampling point deviates from the average level, the greater the likelihood of anomalies.

[0038] By introducing information entropy and mean ratio to quantify the degree of anomaly, data fluctuations and deviations during the operation of PDS ball valves can be effectively captured. This approach not only considers the complexity and volatility of the data but also measures the deviation of each sampling point from the overall average level using the mean ratio. Specifically, a higher information entropy value indicates greater volatility in the data sequence and a higher probability of anomalies; conversely, a greater deviation of the mean ratio from 1 indicates more anomalous data at the sampling point. By combining the deviation of information entropy and mean ratio, this invention can more accurately identify potential fault signals, thereby significantly enhancing the sensitivity of PCA to fault signals. Unlike traditional PCA, which only focuses on directions with high energy, quantifying anomaly features avoids losing crucial anomaly information, thus effectively improving denoising performance and the accuracy of subsequent anomaly detection.

[0039] The steps to obtain the score sequence are as follows:

[0040] Calculate the mean of each column of the initial matrix, subtract the mean from each feature to obtain the decentralized initial matrix, calculate the covariance initial matrix of the decentralized initial matrix, perform eigenvalue decomposition on the covariance initial matrix to obtain eigenvalues ​​and corresponding eigenvectors, arrange all eigenvectors into the initial matrix, and project the decentralized initial matrix onto the eigenvector initial matrix to obtain the initial matrix of principal components, where each column of the initial matrix of principal components represents the score sequence of each principal component.

[0041] Using the score sequence of any principal component as the target sequence, calculate the Pearson correlation coefficient between the target sequence and each abnormal feature sequence. Take the absolute value of the Pearson correlation coefficient and combine it with the mean of each abnormal feature to take the absolute value of the Pearson correlation coefficient for weighted summation. Calculate the sum of the means of all abnormal features for normalization processing to obtain the abnormality sensitivity of the principal component.

[0042] Specifically, the anomaly sensitivity of the principal components satisfies the following relationship:

[0043] ;

[0044] In the formula, Indicates the first Abnormal sensitivity of principal components Indicates abnormal characteristics The total number, Indicates the first The score sequence of each principal component With the A sequence of abnormal features The Pearson correlation coefficient is used to quantify the correlation between two factors, and the range of values ​​for the Pearson correlation coefficient is... Here, the Pearson correlation coefficient is used to characterize the correlation without considering whether it is positive or negative. Therefore, the absolute value of the Pearson correlation coefficient is used for calculation. The closer the absolute value is to 1, the stronger the correlation, and the closer the absolute value is to 0, the weaker the correlation. Indicates the first One abnormal feature The mean.

[0045] In other words, the anomaly sensitivity of each principal component is quantified by multiplying the correlation strength between the anomaly feature sequence of each sensor and the score sequence of the principal component by the average anomaly feature of the corresponding sensor, and then summing the results over all sensors. This comprehensively considers the correlation between the principal components and the anomaly features of each sensor, as well as the average anomaly level of each sensor.

[0046] Simply considering the correlation between principal components and sensor data may not be sufficient to fully reflect the sensitivity of principal components to anomalies, because some sensors may also exhibit high anomalous characteristics under normal conditions, but these characteristics may not be significant. By multiplying by the average anomalous characteristics, the contribution of each sensor to the anomaly sensitivity of the principal components can be measured more accurately.

[0047] If a sensor's anomalous feature sequence is highly correlated with principal components but has a low average anomalous feature, then the sensor's contribution to principal component anomalous sensitivity may be small. Conversely, if a sensor's anomalous feature sequence is low correlated with principal components but has a high average anomalous feature, then the sensor may still make a significant contribution to principal component anomalous sensitivity.

[0048] By calculating the anomaly sensitivity of each principal component and combining it with variance (eigenvalues) to determine the importance of the principal components, the priority of the principal components can be dynamically adjusted. This approach considers not only the variance contribution of the principal component score sequences but also the sensitivity of the principal component score sequences to anomaly features, thus more effectively retaining principal components related to anomalies and improving the robustness and accuracy of the monitoring system.

[0049] For example, in industrial environments, the score sequences of certain low-variance principal components may contain important anomalous information (such as weak shock signals or localized leak signals), which traditional PCA methods might overlook. By introducing anomaly sensitivity, it can be ensured that these important low-variance principal components are preserved when reconstructing the data.

[0050] By quantifying the anomaly sensitivity of principal components, principal components related to anomalies can be identified more accurately, thereby retaining more anomaly information and reducing misjudgments and omissions during data reconstruction and denoising.

[0051] The methods for calculating the importance of each principal component include:

[0052] The proportion of each principal component's eigenvalue to the sum of all eigenvalues ​​is used as the variance contribution rate, and the product of the variance contribution rate and the anomaly sensitivity is used as the importance of each principal component.

[0053] Specifically, the importance level satisfies the following relationship:

[0054] ;

[0055] In the formula, Indicates the first The importance of each principal component Indicates the first The anomaly sensitivity of each principal component is considered. The greater the anomaly sensitivity, the more important the principal component is, and the higher its ranking. It should not be discarded when reconstructing and denoising data. Indicates the first The standard PCA algorithm sorts the principal components based on their eigenvalues. The eigenvalues ​​represent the variance in each eigenvector direction. The larger the eigenvalue, the more information is captured in that direction. Represents the sum and ratio of all eigenvalues. The larger the value, the more likely it is to indicate the first... The larger the proportion of the eigenvalues ​​of the principal components, the better the reflection of the first principal component. The greater the variance captured by each principal component, the higher its importance.

[0056] By combining anomaly sensitivity with variance, the priority of principal components can be dynamically adjusted, thereby prioritizing the retention of information closely related to the abnormal state of the PDS ball valve. This significantly improves the system's adaptability to non-stationary signals and effectively reduces the adverse effects of noise interference on the state identification process. This method can significantly improve the accuracy of the entire monitoring system, enhance the reliability of fault diagnosis, and ensure accurate monitoring and timely early warning of the PDS ball valve's state.

[0057] S3: Sort the principal components according to their importance, traverse different numbers of retained principal components, calculate the denoising effect of different numbers of principal components, select the number of principal components corresponding to the largest denoising effect, reconstruct the running data, and complete the denoising.

[0058] Extract the score sequence of retained principal components from the initial matrix, construct the reconstruction matrix, calculate the reconstruction error of each sampling point in the initial matrix and the reconstruction matrix, calculate the ratio between the sum of anomaly sensitivities corresponding to the number of retained principal components and the sum of anomaly sensitivities corresponding to the number of all principal components, use the reconstruction error as the influence weight, and process the reconstruction error using a normalization function to obtain the denoising effect corresponding to different numbers of retained principal components.

[0059] Specifically, the noise reduction effect satisfies the following relationship:

[0060] ;

[0061] In the formula, Indicates to retain the previous text Denoising effect of principal components Indicates to retain the previous text The reconstruction error between the denoised running data and the running data is a quantitative indicator of the amount of information lost, reflecting the denoising effect. The larger the reconstruction error, the more information is discarded, the more noise is removed, and the better the denoising effect. Represents the normalization function. Indicates to retain the previous text Number of principal components Indicates the first Abnormal sensitivity of principal components Indicates to retain the previous text The sum of the abnormal sensitivities of each principal component, This represents the total number of principal components. This represents the sum of the abnormal sensitivities of all principal components, before... The larger the ratio of the principal component to the anomaly sensitivity, the more details related to the anomaly are retained in the energy, resulting in better noise reduction.

[0062] For example, suppose we have a 4D dataset containing 100 samples. After PCA processing, we obtain 4 principal components with anomaly sensitivities of [missing information]. The corresponding principal components are respectively .

[0063] Principal components sorted by eigenvalue size: The corresponding eigenvalues ​​are as follows: ;

[0064] Retaining 1 principal component: The reconstructed matrix is... Build, calculate reconstruction error and denoising effect .

[0065] Retaining 2 principal components: the reconstructed matrix is... and Build, calculate reconstruction error and denoising effect .

[0066] Retaining 3 principal components: The reconstructed matrix is... , and Build, calculate reconstruction error and denoising effect .

[0067] Retaining 4 principal components: The reconstructed matrix is... , , and Build, calculate reconstruction error and denoising effect .

[0068] Reconstruction error Reflects the retention The ratio of data reconstruction quality after principal components to anomaly sensitivity This reflects the proportion of retained principal components in the total anomaly sensitivity, by comparing different numbers of... This allows us to find the optimal number of principal components to retain in order to achieve the best noise reduction effect.

[0069] S4: For the denoised running data, calculate the reconstruction error of each sampling point, combine the reconstruction error to improve the distance between sampling points in the LOF algorithm, obtain the weighted distance, calculate the local anomaly factor of each sampling point based on the weighted distance, and determine the operating status of the PDS ball valve based on the local anomaly factor.

[0070] Calculate the Euclidean distance between any two sampling points in the denoised running data, calculate the reconstruction error for each sampling point, select the maximum value of the reconstruction error, and weight the Euclidean distance by the ratio between the maximum value of the reconstruction error and the mean of the reconstruction errors of all sampling points to obtain the improved weighted distance.

[0071] Specifically, the weighted distance satisfies the following relationship:

[0072] ;

[0073] In the formula, This represents the weighted distance between two sampling points. This represents the Euclidean distance between two sampling points. Indicates the first Reconstruction error of data collected from each sampling point Indicates the first Reconstruction error of data collected from each sampling point This represents the mean of the reconstruction error across all sampling points.

[0074] In other words, reconstruction error weighting amplifies the distance between sampling points containing uncaptured anomalous features (such as weak fault signals) and other normal points, thereby improving the sensitivity and isolation of the LOF (Local Outlier Factor) algorithm for outliers. By introducing reconstruction error weighting, the impact of the reconstruction process on the calculation of anomaly severity is reduced. This enables more accurate identification and differentiation between weak fault signals and normal signals, avoiding misjudgments caused by information loss during reconstruction, and improving the accuracy and robustness of anomaly detection.

[0075] Based on the improved weighted distance, the local anomaly factor of each sampling point is calculated using the LOF algorithm. When the local anomaly factor is greater than or equal to 1, it is determined that the PDS ball valve has an abnormal condition at the time point corresponding to the sampling point. Conversely, when it is less than 1, it is determined that the PDS ball valve is in normal operation. The threshold in this embodiment is set to 1, which can be adjusted according to specific circumstances.

[0076] This invention also provides a PDS ball valve condition monitoring system based on multi-sensor fusion. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the PDS ball valve status monitoring method based on multi-sensor fusion according to the first aspect of the present invention. The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, the settings and functions of which are known in the art and will not be described further here.

[0077] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A PDS ball valve condition monitoring method based on multi-sensor fusion, characterized in that, The application relates to a method for detecting the running state of a PDS ball valve. The method comprises the following steps: Based on multi-sensor acquisition of the running data of the ball valve, the running data is preprocessed; The initial matrix is fused with the preprocessed running data and subjected to PCA decomposition to obtain principal components and score sequences, the correlation between the abnormal feature sequence of the running data sequence of each sensor and the score sequence of the principal components is calculated, the abnormal sensitivity of each principal component is quantified in combination with the average abnormal level of the abnormal feature sequence, the importance of each principal component is calculated based on the abnormal sensitivity and the variance contribution rate of each principal component; The principal components are sorted according to the importance, different numbers of reserved principal components are traversed, the denoising effect of different numbers of principal components is calculated, the number of principal components corresponding to the maximum denoising effect is selected, the running data is reconstructed, and denoising is completed; The reconstruction error of each sampling point is calculated based on the denoised running data, the distance between sampling points in the improved LOF algorithm is calculated in combination with the reconstruction error, the weighted distance is obtained, the local anomaly factor of each sampling point is calculated based on the weighted distance, and the running state of the PDS ball valve is judged according to the local anomaly factor. The weighted distance is obtained in the following manner: The Euclidean distance between any two sampling points in the denoised running data is calculated, the reconstruction error of each sampling point is calculated, the maximum value of the reconstruction error is selected, the maximum value of the reconstruction error is weighted with the average value of the reconstruction error of all sampling points, and the improved weighted distance is obtained.

2. The multi-sensor fusion based PDS ball valve condition monitoring method as claimed in claim 1, wherein, Based on the improved weighted distance, the local anomaly factor of each sampling point is calculated by using the LOF algorithm, and when the local anomaly factor is greater than or equal to 1, it is determined that the PDS ball valve has an abnormal condition at the time point corresponding to the sampling point, otherwise, when the local anomaly factor is less than 1, it is determined that the PDS ball valve is in a normal running state. The calculation method of the abnormal feature sequence comprises the following steps: In the initial matrix, each row represents a sample, and each column represents the data dimension of a sensor. For the running data sequence of each sensor in the initial matrix, the average value of the running data sequence of each sensor is calculated, the ratio of the data of each sampling point to the average value is calculated, the absolute value of 1 minus the ratio is taken as the relative deviation degree of each sampling point, the information entropy value of the running data sequence of each sensor is calculated, and the information entropy value is multiplied by the relative deviation degree to obtain the abnormal feature of the data of each sampling point. The abnormal feature sequence of the running data sequence is constructed.

3. The multi-sensor fusion based PDS ball valve condition monitoring method as claimed in claim 1, wherein, The score sequence is obtained in the following manner: The average value of each column of the initial matrix is calculated, the average value is subtracted from each feature to obtain a decentralized initial matrix, a covariance initial matrix of the decentralized initial matrix is calculated, the covariance initial matrix is subjected to eigenvalue decomposition to obtain eigenvalues and corresponding eigenvectors, all the eigenvectors are arranged into an initial matrix, and the decentralized initial matrix is projected onto the eigenvector initial matrix to obtain an initial matrix of principal components, wherein each row in the initial matrix of principal components represents a score sequence of each principal component.

4. The multi-sensor fusion based PDS ball valve condition monitoring method as claimed in claim 1, wherein, The calculation method of the abnormal sensitivity of the principal components comprises: Taking the score sequence of any principal component as a target sequence, a Pearson correlation coefficient between the target sequence and each abnormal feature sequence is calculated, the Pearson correlation coefficient is taken as an absolute value, and the Pearson correlation coefficient taken as the absolute value is weighted and summed in combination with a mean value of each abnormal feature, a sum of the mean values of all abnormal features is calculated for normalization processing, and an abnormal sensitivity of the principal component is obtained.

5. The multi-sensor fusion based PDS ball valve condition monitoring method as claimed in claim 1, wherein, The calculation manner of the importance of each principal component includes: Taking a proportion of the eigenvalue of each principal component in the sum of all eigenvalues as a variance contribution rate, and taking a product of the variance contribution rate and the abnormal sensitivity as the importance of each principal component.

6. The multi-sensor fusion based PDS ball valve condition monitoring method as claimed in claim 1, wherein, The calculation manner of the denoising effect of different principal component numbers includes: The score sequence of the retained principal component is extracted from the initial matrix to construct a reconstruction matrix, the reconstruction error of each sampling point in the initial matrix and the reconstruction matrix is calculated, the ratio between the sum of the abnormal sensitivities corresponding to the number of retained principal components and the sum of the abnormal sensitivities corresponding to the number of all principal components is calculated, the reconstruction error is taken as an influence weight, the reconstruction error is processed by using a normalization function, and the denoising effects corresponding to the numbers of retained different principal components are obtained.

7. A PDS ball valve condition monitoring system based on multi-sensor fusion, characterized by, It includes: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, the method for monitoring the state of a PDS ball valve based on multi-sensor fusion according to any one of claims 1-6 is realized.

Images