Health management method and system for multi-medium conveying pipeline

Abstract

The invention provides a health management method and system for a multi-medium conveying pipeline, and the method specifically comprises the steps: obtaining pipeline monitoring data and medium recognition data; initializing a particle swarm, wherein each particle carries a weight and represents a damage geometric parameter and a noise covariance; adjusting a process noise covariance matrix based on the particle weight dispersion at the previous moment, and predicting the particle state by using a medium related state transfer function; calculating the particle likelihood and updating the weight according to the current monitoring data; if the effective particle number proportion is too low, medium type switching or damage parameter change rate exceeds a threshold value, triggering resampling; weighted summation is carried out on particle state vectors, damage geometric parameters and a noise covariance matrix are output, and the health state of the pipeline is accurately represented.

Description

Technical Field

[0001] This application belongs to the field of health management, and in particular relates to a health management method and system for multi-media delivery pipelines. Background Technology

[0002] Pipelines, as critical transportation facilities for fluid media such as oil, natural gas, and chemical products, are essential for national economic development and public safety due to their safe and stable operation. During long-term service, pipelines can suffer various forms of damage, including cracks and thinning, due to corrosion, fatigue, and third-party sabotage. Failure to detect and address these damages promptly can lead to catastrophic accidents such as leaks or even explosions. Currently, analysis of monitoring signals such as acoustic emission, vibration, and strain is a common method for pipeline health monitoring. For example, CN112251756A discloses a dynamic DC corrosion risk assessment system and method for buried metal pipelines, which mainly judges pipeline corrosion based on direct and indirect corrosion parameters. Lu Zhongqi et al.'s paper, "Corrosion and Leakage Prediction of Submarine Oil and Gas Pipelines Based on Parameter-Optimized Neural Networks," uses feedforward neural networks and random forest regression to predict pipeline corrosion. However, in practical engineering applications, many pipelines need to transport different types of media, such as crude oil, refined oil, water, or natural gas, depending on production needs. Different media differ in density, viscosity, temperature, and flow rate, and the structural characteristics of the pipeline and the features of the monitoring signals also change accordingly.

[0003] Traditional particle filtering algorithms typically rely on a pre-established and fixed system model, including state transition models and the statistical characteristics of process and measurement noise. When the medium type changes, the response characteristics of the pipeline system and the noise level of the sensor signals both change. Existing predictive models cannot accurately represent the actual state evolution, leading to model mismatch and severely impacting the accuracy of damage parameter estimation, potentially even causing filter divergence. Particle filtering methods inherently suffer from particle degradation, where a few particles have excessively large weights while the weights of others approach zero. When the medium type changes abruptly, the system state may jump. Relying solely on traditional particle number thresholds to trigger resampling may fail to respond promptly to such abrupt changes, resulting in loss of particle set diversity and tracking failure. Therefore, there is an urgent need for a pipeline health monitoring method that can adapt to changes in the medium, improving the reliability of monitoring under complex operating conditions. Summary of the Invention

[0004] This invention proposes a health management method for multi-media transport pipelines to address the problem of monitoring model mismatch in existing methods when the transported media changes within the pipeline. The method includes the following steps: Acquire monitoring data of the pipeline and current medium identification data indicating the type of medium being transported in the pipeline; initialize a set of particles, each carrying a weight and represented by a state vector, the state vector including the pipeline's damage geometry parameters and measurement noise covariance matrix parameters; Based on the weight dispersion of the particle set at the previous moment, the process noise covariance matrix is ​​adjusted; and using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data, the state vector of each particle is predicted to obtain a predicted particle set; based on the current monitoring data, the likelihood of each predicted particle is calculated using the measurement noise covariance matrix parameters in the state vector of each predicted particle; and the weights of each predicted particle are updated by combining the previous weights of each particle with the calculated likelihood to obtain a weighted predicted particle set. If the ratio of effective particles to total particles is less than a first preset threshold, the current medium identification data indicates a change in medium type, or the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, the weighted predicted particle set is resampled. Based on the particle set after resampling or without resampling, the damage geometric parameters and measurement noise covariance matrix parameters representing the health status of the pipeline are output by weighted summation of the parameter components in the state vector.

[0005] Optionally, the initialization process involves a set of particles, each carrying a weight and represented by a state vector. The state vector includes damage geometry parameters of the pipeline and measurement noise covariance matrix parameters, including: Set the total number of particles N; For each particle, random samples are taken from a pre-defined uniform distribution of damage geometry parameters to generate initial values ​​for the damage geometry parameters. Set the same initial measurement noise covariance matrix parameters for each particle; The initial weight of each particle is set to 1 / N.

[0006] Optionally, adjusting the process noise covariance matrix based on the weight dispersion of the particle set at the previous time step includes: Calculate the effective number of particles in the particle set at the previous time step. The calculation formula is: ; Where N is the total number of particles. The normalized weight of the i-th particle at the previous time step; When the effective number of particles If the particle set weights are less than the product of the total number of particles N and a preset ratio, the particle set weights are considered to be too discrete. The diagonal elements of the process noise covariance matrix are then increased by a preset amount to increase particle diversity.

[0007] Optionally, the step of predicting the state vector of each particle using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data includes: The current medium identification data includes at least one of the following: medium composition, pH value, sulfur content, sand content, flow rate, and temperature; based on the current medium identification data, the corrosivity level of the current medium is determined: When the corrosion level is lower than a preset level threshold, a first state transition model representing a stable corrosion process is adopted. When the corrosion level is higher than or equal to the preset level threshold, a second state transition model representing an accelerated corrosion process is adopted.

[0008] Optionally, the step of calculating the likelihood of each predicted particle based on the current monitoring data and using the measurement noise covariance matrix parameter in the state vector of each predicted particle includes: For each predicted particle, the damage geometry parameters in the particle state vector are input into the pre-established pipeline finite element model to calculate a set of predicted monitoring data. Based on the predicted monitoring data and the current monitoring data, the likelihood of the predicted particle is calculated using a multidimensional Gaussian probability density function. .

[0009] Optionally, the rate of change of the output damage geometry parameter within a preset time window is greater than a second preset threshold, including: Calculate the damage geometry parameters output at the current time within a preset time window M. Compared with the average value of the damage geometric parameters output at the previous M-1 monitoring times The difference is calculated, and the difference is normalized relative to the preset nominal size of the pipe to obtain the rate of change. And determine the rate of change. Whether the absolute value is greater than the second preset threshold; The rate of change The calculation formula is: ; in, , The nominal size of the pipe is the preset nominal size.

[0010] Optionally, the step of outputting damage geometric parameters and measurement noise covariance matrix parameters representing the pipeline health state by weighted summation of each parameter component in the state vector, based on the resampled or unresampled particle set, includes: Based on particle sets, by analyzing any parameter component in the state vector We use weighted summation to calculate the output estimate. The calculation formula is: ; Where N is the total number of particles. Let be the normalized weight of the i-th particle. This represents the value of the j-th parameter component of the state vector in the i-th particle.

[0011] Furthermore, the present invention also relates to a health management system for a multi-media delivery pipeline, comprising the following modules: The acquisition module is used to acquire monitoring data of the pipeline and current medium identification data representing the type of medium being transported in the pipeline; initialize a set of particles, each carrying a weight and represented by a state vector, the state vector including the pipeline's damage geometry parameters and measurement noise covariance matrix parameters; The calculation module is used to adjust the process noise covariance matrix according to the weight dispersion of the particle set at the previous time step; and use the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data to predict the state vector of each particle to obtain a predicted particle set; based on the current monitoring data, use the measurement noise covariance matrix parameters in the state vector of each predicted particle to calculate the likelihood of each predicted particle; and combine the previous weights of each particle with the calculated likelihood to update the weights of each predicted particle to obtain a weighted predicted particle set. The judgment module is used to determine whether the ratio of the effective number of particles to the total number of particles is less than a first preset threshold, the current medium identification data indicates a change in the medium type, or the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, and to resample the weighted predicted particle set. The output module is used to output the damage geometry parameters and measurement noise covariance matrix parameters representing the health status of the pipeline by weighted summation of each parameter component in the state vector, based on the particle set after resampling or without resampling.

[0012] Preferably, the initialization involves a set of particles, each carrying a weight and represented by a state vector. The state vector includes damage geometry parameters of the pipeline and measurement noise covariance matrix parameters, including: Set the total number of particles N; For each particle, random samples are taken from a pre-defined uniform distribution of damage geometry parameters to generate initial values ​​for the damage geometry parameters. Set the same initial measurement noise covariance matrix parameters for each particle; The initial weight of each particle is set to 1 / N.

[0013] Preferably, adjusting the process noise covariance matrix based on the weight dispersion of the particle set at the previous time step includes: Calculate the effective number of particles in the particle set at the previous time step. The calculation formula is: ; Where N is the total number of particles. The normalized weight of the i-th particle at the previous time step; When the effective number of particles If the particle set weights are less than the product of the total number of particles N and a preset ratio, the particle set weights are considered to be too discrete. The diagonal elements of the process noise covariance matrix are then increased by a preset amount to increase particle diversity.

[0014] Preferably, the step of predicting the state vector of each particle using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data includes: The current medium identification data includes at least one of the following: medium composition, pH value, sulfur content, sand content, flow rate, and temperature; based on the current medium identification data, the corrosivity level of the current medium is determined: When the corrosion level is lower than a preset level threshold, a first state transition model representing a stable corrosion process is adopted. When the corrosion level is higher than or equal to the preset level threshold, a second state transition model representing an accelerated corrosion process is adopted.

[0015] Preferably, the step of calculating the likelihood of each predicted particle based on the current monitoring data and using the measurement noise covariance matrix parameter in the state vector of each predicted particle includes: For each predicted particle, the damage geometry parameters in the particle state vector are input into the pre-established pipeline finite element model to calculate a set of predicted monitoring data. Based on the predicted monitoring data and the current monitoring data, the likelihood of the predicted particle is calculated using a multidimensional Gaussian probability density function. .

[0016] Preferably, the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, including: Calculate the damage geometry parameters output at the current time within a preset time window M. Compared with the average value of the damage geometric parameters output at the previous M-1 monitoring times The difference is calculated, and the difference is normalized relative to the preset nominal size of the pipe to obtain the rate of change. And determine the rate of change. Whether the absolute value is greater than the second preset threshold; The rate of change The calculation formula is: ; in, , The nominal size of the pipe is the preset nominal size.

[0017] Preferably, the step of outputting damage geometric parameters and measurement noise covariance matrix parameters representing the pipeline health state by weighted summation of each parameter component in the state vector, based on the resampled or unresampled particle set, includes: Based on particle sets, by analyzing any parameter component in the state vector We use weighted summation to calculate the output estimate. The calculation formula is: ; Where N is the total number of particles. Let be the normalized weight of the i-th particle. This represents the value of the j-th parameter component of the state vector in the i-th particle.

[0018] This invention provides a health monitoring method for multi-media alternating transport pipelines, which can solve the monitoring model mismatch problem caused by changes in the transported media within the pipeline. By utilizing media identification data and employing a state transition function corresponding to the current media type for particle state prediction, the accuracy of the system state evolution model under different media conditions is ensured, improving the precision of damage identification. Simultaneously, by incorporating the measurement noise covariance matrix into the particle state vector estimation, the method can detect and compensate for changes in the noise characteristics of the measurement signal caused by different media flowing through the pipeline, enhancing the reliability of the results. By setting a mechanism for immediate resampling upon media type switching, the particle set can be rapidly adjusted when system characteristics undergo abrupt changes, ensuring the continuity and stability of the monitoring process. Attached Figure Description

[0019] Figure 1 A flowchart of the first embodiment; Figure 2 A schematic diagram for particle initialization; Figure 3 Schematic diagrams of state transition models under different media; Figure 4 This is a schematic diagram illustrating the likelihood calculation based on monitoring data. Figure 5 This is a schematic diagram of the resampling process; Figure 6 This is a schematic diagram illustrating the resampling condition determination. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the first embodiment, the present invention proposes a health management method for multi-media delivery pipelines, such as... Figure 1 This includes the following steps: S1, acquire monitoring data of the pipeline and current medium identification data indicating the type of medium being transported in the pipeline; initialize a set of particles, each carrying a weight and represented by a state vector, the state vector including the pipeline's damage geometry parameters and measurement noise covariance matrix parameters; Specifically, monitoring equipment such as accelerometers or strain gauges are installed along the pipeline to continuously collect vibration or strain response signals as monitoring data. At the same time, the pipeline control system, such as SCADA system, or equipment such as flow meters and densitometers installed on the pipeline, can obtain the type identifier of the medium currently being transported in the pipeline, such as crude oil, water or natural gas, in real time. For example, the number 1 represents crude oil and 2 represents water. This type identifier is the current medium identification data.

[0022] Set the total number of particles to N, for example, N=5000. At the initial moment, N particles are generated. Each particle's state vector contains parameters to be estimated. For example, for pipe crack damage, the state vector can be set to x=[a,R], where a is the damage geometry parameter of the crack depth, and R is the measurement noise covariance matrix parameter. The state vectors of all N particles are initialized by random sampling within the possible range of parameters, and each particle is assigned an equal initial weight, i.e., 1 / N.

[0023] In an optional embodiment, the initialization involves a set of particles, each carrying a weight and represented by a state vector, the state vector including damage geometry parameters of the pipeline and measurement noise covariance matrix parameters, including: Set the total number of particles N; For each particle, random samples are taken from a pre-defined uniform distribution of damage geometry parameters to generate initial values ​​for the damage geometry parameters. Set the same initial measurement noise covariance matrix parameters for each particle; The initial weight of each particle is set to 1 / N.

[0024] The initialization process provides an initial set of state assumptions for the particle filter algorithm. The total number of particles, N, is set to a large value, such as N=500, to ensure sufficient exploration of the state space. An initial state is generated for each particle. It is assumed that pipe damage is represented by two geometric parameters: depth d and length l. Based on prior knowledge, the depth d ranges from 0 to 10 mm, and the length l ranges from 0 to 50 mm. For the i-th particle, a value is randomly selected from a uniform distribution of 0 to 10 as the initial depth. And randomly select a value from a uniform distribution of 0 to 50 as the initial length. .

[0025] Each particle's state vector also includes a measurement noise covariance matrix. The parameters, the matrix, reflect the uncertainty of the monitoring data. During the initialization phase, all particles are assigned the same initial measurement noise covariance matrix. For example, if the monitoring data is two-dimensional, This can be set as a diagonal matrix, where the diagonal elements represent the initial noise variance of the two measurement channels. To represent that all initial assumptions are equally likely before any measurement data is available, the initial weights for each particle are... All are set to the same value, which is the reciprocal of the total number of particles, 1 / N, or 1 / 500 in this example. After completing the above steps, an initial particle set of 500 particles is obtained, where each particle represents a possible initial state of pipe damage, such as... Figure 2 .

[0026] S2, adjust the process noise covariance matrix according to the weight dispersion of the particle set at the previous moment; and use the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data to predict the state vector of each particle to obtain a predicted particle set; based on the current monitoring data, use the measurement noise covariance matrix parameters in the state vector of each predicted particle to calculate the likelihood of each predicted particle; and combine the previous weights of each particle with the calculated likelihood to update the weights of each predicted particle to obtain a weighted predicted particle set. Specifically, the dispersion of the particle set weights at the previous time step is calculated, for example, by the number of particles; the more concentrated the weights, the smaller the number of particles. The magnitude of the process noise covariance matrix Q is adjusted based on this dispersion; for example, when the weights are concentrated, the value of Q is increased to increase the particle's exploration range and prevent premature convergence. Based on the medium identification data acquired at the current time step, a corresponding state transition function is selected. For example, when the medium is water, a model is selected... When the medium is oil, select the model. The state transition function represents the evolution of damage parameters over time. Different media, with their corrosive or erosive properties, may lead to varying damage evolution rates. For each particle, the state vector is predicted in one step using the selected state transition function and the adjusted process noise covariance matrix. For example, the predicted state of the i-th particle is... , where v is the process noise sampled from a normal distribution with zero mean and covariance Q. After all particles are predicted, a predicted particle set is formed.

[0027] For each predicted particle, the damage geometry parameters, such as crack depth α, from the particle state vector are substituted into an observation model corresponding to the current medium type. The observation model is a physical or numerical model, such as a finite element model, capable of calculating theoretical monitoring data values ​​based on damage parameters and medium type. The theoretical monitoring values ​​are then obtained. Then, using the measurement noise covariance matrix parameter R in the predicted particle state vector, the deviation between the current actual monitoring data z and the theoretical monitoring value is calculated, and a likelihood value is calculated based on the Gaussian distribution assumption. This likelihood value represents the degree of agreement between the state represented by the particle and the actual situation. This likelihood value is multiplied by the particle's weight at the previous time step to obtain the particle's unnormalized new weight at the current time step. After performing this operation on all particles, all new weights are normalized so that their sum equals 1, thus obtaining the weighted predicted particle set.

[0028] To prevent particle degradation, in an optional embodiment, adjusting the process noise covariance matrix based on the weight dispersion of the particle set at the previous time step includes: Calculate the effective number of particles in the particle set at the previous time step. The calculation formula is: ; Where N is the total number of particles. The normalized weight of the i-th particle at the previous time step; When the effective number of particles If the particle set weights are less than the product of the total number of particles N and a preset ratio, the particle set weights are considered to be too discrete. The diagonal elements of the process noise covariance matrix are then increased by a preset amount to increase particle diversity.

[0029] After updating the weights at time step (k-1), a formula is used to evaluate the dispersion of the particle weights. For example, assuming N=1000 particles, if the weights are uniformly distributed, then... Approaching 1000; if the weights are highly concentrated, then It will decrease. A preset percentage threshold, such as 0.6, is set. Therefore, the threshold value is 600.

[0030] In the calculation, if the number of effective particles in the previous time step... The calculated value is 450. Since 450 < 600, the particle set weight dispersion is too large, indicating a severe lack of diversity. Therefore, it is necessary to artificially increase the particle dispersion range in the state space by increasing the process noise. The process noise covariance matrix Q controls the magnitude of random perturbations in the state prediction step. The diagonal elements of Q, i.e., the process noise variance of each state variable, are multiplied by a preset inflation factor greater than 1, such as 1.1. This adjustment ensures that even particles with high weights will experience greater random perturbations in subsequent prediction steps, resulting in more diverse descendant particles. This helps the algorithm explore new state regions and avoids premature convergence to a local optimum.

[0031] In an optional embodiment, predicting the state vector of each particle using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data includes: The current medium identification data includes at least one of the following: medium composition, pH value, sulfur content, sand content, flow rate, and temperature; based on the current medium identification data, the corrosivity level of the current medium is determined: When the corrosion level is lower than a preset level threshold, a first state transition model representing a stable corrosion process is adopted. When the corrosion level is higher than or equal to the preset level threshold, a second state transition model representing an accelerated corrosion process is adopted.

[0032] This embodiment improves the accuracy of state prediction by inputting knowledge related to the physicochemical properties of the medium being transported in the pipeline. At each time step k, data from the medium identification sensor is queried. Assuming that at the current moment, the sensor data shows that the pipeline is transporting low-sulfur refined oil with a stable flow rate and neutral pH, and the calculated medium corrosivity level based on these parameters is below a preset threshold, the current state is determined to be in a stable corrosion stage. A first state transition model is selected to predict the changes in the damage geometry parameters of each particle. The structure of the first state transition model is a linear growth model; for example, for damage depth d, the prediction formula is... ,in It is a small constant representing the slow rate of damage in a low-corrosion environment. It is the time step. It is process noise.

[0033] If at another moment, the medium identification data changes to high-sulfur crude oil, or is accompanied by high sand content and acidic gases, such as... , High-velocity fluids can cause a significant increase in the calculated corrosion level, exceeding or equaling the preset threshold. The system automatically switches to a second state transition model. The second state transition model is preferably a nonlinear accelerated growth model to represent the rapid deterioration of damage under harsh conditions. For example, the prediction formula becomes... ,in It is a ratio A much larger corrosion coefficient, with an exponent α greater than 1, indicates that the rate of damage growth increases dramatically with the severity of the damage and the increased corrosivity of the medium. By leveraging multidimensional medium characteristics and model switching, the state prediction for each particle more closely approximates the actual damage change process, thereby improving the tracking performance and accuracy of the entire particle filtering algorithm. Figure 3 .

[0034] In an optional embodiment, calculating the likelihood of each predicted particle based on current monitoring data, using the measurement noise covariance matrix parameters in the state vector of each predicted particle, includes: For each predicted particle, the damage geometric parameters in the particle state vector are input into the pre-established pipeline finite element model to calculate a set of predicted monitoring data. The pre-established pipeline finite element model is a numerical simulation model built based on the actual geometric dimensions, material properties, characteristics of the transport medium, and operating conditions of the pipeline. Its modeling steps include geometric modeling, mesh generation, parameter assignment, and boundary condition setting. The damage geometric parameters carried by the particles can be input, and theoretical monitoring data under the corresponding operating conditions can be output through numerical simulation calculation.

[0035] Based on the predicted monitoring data and the current monitoring data, the likelihood of the predicted particle is calculated using a multidimensional Gaussian probability density function. The calculation formula is: ; in, For current monitoring data, The damage geometry parameters of the i-th predicted particle The predicted monitoring data obtained by inputting it into the pipeline finite element model Let m be the measurement noise covariance matrix in the predicted particle state vector, where m is the dimension of the monitoring data.

[0036] For the i-th predicted particle, the state vector It includes a set of assumed damage geometry parameters, such as a damage depth of 5.2 mm and a length of 30 mm. These parameters are used as input to a pre-established high-precision finite element model of the pipeline. The structure of this finite element model is based on physical equations to numerically simulate the mechanical behavior of the pipeline. It can calculate the pipeline's response at the corresponding monitoring point location under specific operating conditions, such as strain or vibration frequency, based on the input damage geometry.

[0037] For example, the predictive monitoring data calculated by the finite element model This is a vector containing two strain values, such as [150.5, 310.2] microstrain. Simultaneously, monitoring data at the current moment is acquired from real sensors on the pipeline. For example, the microstrain is [148.9, 313.5]. The likelihood of the particle is calculated using the multidimensional Gaussian probability density function formula. The above formula calculates the actual monitoring data. With predictive monitoring data The Mahalanobis distance between them was calculated, and the measurement noise covariance matrix carried by the particle was taken into account. The closer the predicted value is to the actual value, the higher the calculated likelihood value, and the more consistent it is with the actual situation. Figure 4 .

[0038] In an optional embodiment, at the current moment, assuming the pipeline is transporting crude oil, the system has completed state prediction and obtained a set of predicted particles. Each predicted particle contains a state vector such as crack depth, measurement noise covariance, and weights inherited from the previous moment. The system then acquires current actual monitoring data. Micro-strain. Taking the i-th particle as an example, its predicted crack depth... Measure noise covariance The damage parameter is input into the pipeline finite element observation model corresponding to the crude oil medium. The theoretical monitoring value was calculated. Micro-strain. According to the formula... Calculate the likelihood, and the result is: Combined with the particle's previous weights Calculate the new unnormalized weights This process is performed in parallel for all N=500 particles. Assume that after the computation, the sum of the unnormalized weights of all particles is S=0.0943. Then, normalization is performed, resulting in the final weight of the particle. Similarly, the normalized weights of all particles are calculated to obtain the weighted predicted particle set.

[0039] S3, if the ratio of the effective number of particles to the total number of particles is less than the first preset threshold, the current medium identification data indicates a change in the medium type, or the rate of change of the output damage geometry parameter within a preset time window is greater than the second preset threshold, the weighted predicted particle set is resampled. Calculate the ratio of the effective number of particles in the weighted particle set to the total number of particles N. If this ratio is less than a preset value, such as 0.5, the first condition is met. Compare the current medium identification data with that of the previous time step. If they are inconsistent, for example, from 1 representing crude oil to 2 representing water, the second condition is met. Calculate the estimated damage geometry parameters for the current time step based on the weights of all particles, and record the estimated value sequence over a recent period, such as 10 time steps. Calculate the slope of change. If the absolute value of this slope is greater than a set second threshold, the third condition is met. When any condition is triggered, a resampling step is performed. For example, using a system resampling method, N new particles are drawn with replacement from the current weighted predicted particle set, where particles with larger weights have a higher probability of being drawn. After resampling, the weights of all new particles are reset to an equal 1 / N, such as... Figure 5 .

[0040] In some embodiments, the ratio of the effective number of particles to the total number of particles is less than a first preset threshold, including: Calculate the current ratio of effective particles to total particles, and compare the current ratio with the first preset threshold.

[0041] For example, the ratio of effective particles to total particles is calculated to be 0.272. This ratio is then compared to a preset first threshold, such as 0.5. Since 0.272 < 0.5, it indicates that the particle set weights are highly concentrated, and the contribution of the vast majority of particles is negligible. The algorithm has severely degraded, thus triggering a resampling operation.

[0042] To address drastic changes in estimation results due to model mismatch or unforeseen circumstances, in an optional embodiment, the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, including: Calculate the damage geometry parameters output at the current time within a preset time window M. Compared with the average value of the damage geometric parameters output at the previous M-1 monitoring times The difference is calculated, and the difference is normalized relative to the preset nominal size of the pipe to obtain the rate of change. And determine the rate of change. Whether the absolute value is greater than the second preset threshold; The rate of change The calculation formula is: ; in, , The nominal size of the pipe is the preset nominal size.

[0043] Define a time window M, for example, M equals 12 monitoring cycles, and a reference dimension for the pipe, such as the nominal wall thickness. The depth of damage is 20 mm. At the current time k, the estimated value of the damage depth is calculated as follows: mm. Damage depth estimates for 11 historical moments from time k-11 to k-1 are stored.

[0044] Calculate the average of the above 11 historical estimates. For example, the calculated result is 6.2 mm. The normalized rate of change is calculated using the formula. That is, 0.03. A second threshold, such as 0.025, is preset to define the acceptable estimated range of change. This is based on the calculated rate of change. A value of 0.03 greater than the threshold of 0.025 indicates a significant shift in the current estimated value compared to recent trends. This shift suggests potential failure of particle filter tracking, triggering a resampling process to readjust particle distribution and detect possible state abrupt changes. Figure 6 .

[0045] S4, based on the particle set after resampling or without resampling, outputs the damage geometric parameters and measurement noise covariance matrix parameters representing the health status of the pipeline by weighted summation of each parameter component in the state vector.

[0046] Specifically, the weights of all N particles in the current particle set are multiplied by the corresponding parameter components in the state vector, and then summed to obtain the estimated values ​​of each parameter. For example, the estimated value of the pipe crack depth is... Where i ranges from 1 to N, It is the weight of the i-th particle. This is the crack depth value in the state vector. Similarly, the estimated value of the parameters of the measurement noise covariance matrix is... The measurement noise covariance matrix parameter is a statistical parameter within the particle filter algorithm, representing the reliability or noise level of the sensor data. The output estimate is the assessment result of the pipeline health status at the current moment.

[0047] In an optional embodiment, the step of outputting damage geometry parameters and measurement noise covariance matrix parameters representing the pipeline health state by weighted summation of each parameter component in the state vector, based on the resampled or unresampled particle set, includes: Based on particle sets, by analyzing any parameter component in the state vector We use weighted summation to calculate the output estimate. The calculation formula is: ; Where N is the total number of particles. Let be the normalized weight of the i-th particle. This represents the value of the j-th parameter component of the state vector in the i-th particle.

[0048] At the end of each iteration of the particle filter, a definite state estimate is extracted from the particle swarm. At time k, after prediction and update steps, a state estimate containing N particles and their corresponding normalized weights is obtained. The posterior particle set. Each particle's state vector contains multiple parametric components, such as the j-th component being the damage depth. N discrete hypotheses regarding the damage depth are then used. Combine them into a single output result.

[0049] The synthesis process is achieved through a weighted average. It iterates through all N particles, calculating the damage depth value for each particle. With the corresponding weights Multiply the products and add them together. For example, suppose there are three particles with damage depths of 4.1 mm, 4.5 mm, and 3.9 mm, with corresponding weights of 0.2, 0.7, and 0.1. Then, the output damage depth estimate is... The calculated result is 4.36 mm. This result is the expected value of the damage depth based on all currently available information; particles with higher weights contribute more to the result. Other parameters in the state vector, such as the damage length, are estimated using the same method.

[0050] In a second embodiment, the present invention also provides a health management system for a multi-media delivery pipeline, comprising the following modules: The acquisition module is used to acquire monitoring data of the pipeline and current medium identification data representing the type of medium being transported in the pipeline; initialize a set of particles, each carrying a weight and represented by a state vector, the state vector including the pipeline's damage geometry parameters and measurement noise covariance matrix parameters; The calculation module is used to adjust the process noise covariance matrix according to the weight dispersion of the particle set at the previous time step; and use the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data to predict the state vector of each particle to obtain a predicted particle set; based on the current monitoring data, use the measurement noise covariance matrix parameters in the state vector of each predicted particle to calculate the likelihood of each predicted particle; and combine the previous weights of each particle with the calculated likelihood to update the weights of each predicted particle to obtain a weighted predicted particle set. The judgment module is used to determine whether the ratio of the effective number of particles to the total number of particles is less than a first preset threshold, the current medium identification data indicates a change in the medium type, or the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, and to resample the weighted predicted particle set. The output module is used to output the damage geometry parameters and measurement noise covariance matrix parameters representing the health status of the pipeline by weighted summation of each parameter component in the state vector, based on the particle set after resampling or without resampling.

[0051] In this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0052] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0053] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A health management method for multi-media transport pipelines, characterized in that, Includes the following steps: Acquire monitoring data of the pipeline and current medium identification data indicating the type of medium being transported in the pipeline; initialize a set of particles, each carrying a weight and represented by a state vector, the state vector including the pipeline's damage geometry parameters and measurement noise covariance matrix parameters; Adjust the process noise covariance matrix based on the weight dispersion of the particle set at the previous moment; The state vectors of each particle are predicted using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data to obtain a predicted particle set. Based on the current monitoring data, the likelihood of each predicted particle is calculated using the measurement noise covariance matrix parameters in the state vectors of each predicted particle. The weights of each predicted particle are updated by combining the previous weights of each particle with the calculated likelihoods to obtain a weighted predicted particle set. If the ratio of effective particles to total particles is less than a first preset threshold, the current medium identification data indicates a change in medium type, or the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, the weighted predicted particle set is resampled. Based on the particle set after resampling or without resampling, the damage geometric parameters and measurement noise covariance matrix parameters representing the health status of the pipeline are output by weighted summation of the parameter components in the state vector.

2. The method according to claim 1, characterized in that, The initialization process involves a set of particles, each carrying a weight and represented by a state vector. The state vector includes the pipe's damage geometry parameters and the measurement noise covariance matrix parameters, including: Set the total number of particles N; For each particle, random samples are taken from a pre-defined uniform distribution of damage geometry parameters to generate initial values ​​for the damage geometry parameters. Set the same initial measurement noise covariance matrix parameters for each particle; The initial weight of each particle is set to 1 / N.

3. The method according to any one of claims 1 and 2, characterized in that, The adjustment of the process noise covariance matrix based on the weight dispersion of the particle set at the previous time step includes: Calculate the effective number of particles in the particle set at the previous time step. The calculation formula is: ； Where N is the total number of particles. The normalized weight of the i-th particle at the previous time step; When the effective number of particles If the particle set weights are less than the product of the total number of particles N and a preset ratio, the particle set weights are considered to be too discrete. The diagonal elements of the process noise covariance matrix are then increased by a preset amount to increase particle diversity.

4. The method according to any one of claims 1 and 2, characterized in that, The step of predicting the state vector of each particle using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data includes: The current medium identification data includes at least one of the following: medium composition, pH value, sulfur content, sand content, flow rate, and temperature; based on the current medium identification data, the corrosivity level of the current medium is determined: When the corrosion level is lower than a preset level threshold, a first state transition model representing a stable corrosion process is adopted. When the corrosion level is higher than or equal to the preset level threshold, a second state transition model representing an accelerated corrosion process is adopted.

5. The method according to any one of claims 1 and 2, characterized in that, The step of calculating the likelihood of each predicted particle based on the current monitoring data and using the measurement noise covariance matrix parameters in the state vector of each predicted particle includes: For each predicted particle, the damage geometry parameters in the particle state vector are input into the pre-established pipeline finite element model to calculate a set of predicted monitoring data. Based on the predicted monitoring data and the current monitoring data, the likelihood of the predicted particle is calculated using a multidimensional Gaussian probability density function. .

6. The method according to claim 1, characterized in that, The rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, including: Calculate the damage geometry parameters output at the current time within a preset time window M. Compared with the average value of the damage geometric parameters output at the previous M-1 monitoring times The difference is calculated, and the difference is normalized relative to the preset nominal size of the pipe to obtain the rate of change. And determine the rate of change. Whether the absolute value is greater than the second preset threshold.

7. The method according to claim 6, characterized in that, The particle set, whether resampled or not, outputs damage geometric parameters and measurement noise covariance matrix parameters representing the pipeline health state by weighted summation of each parameter component in the state vector, including: Based on particle sets, by analyzing any parameter component in the state vector We use weighted summation to calculate the output estimate. The calculation formula is: ； Where N is the total number of particles. Let be the normalized weight of the i-th particle. This represents the value of the j-th parameter component of the state vector in the i-th particle.

8. A health management system for a multi-media delivery pipeline, characterized in that, Includes the following modules: The acquisition module is used to acquire monitoring data of the pipeline and current medium identification data representing the type of medium being transported in the pipeline; initialize a set of particles, each carrying a weight and represented by a state vector, the state vector including the pipeline's damage geometry parameters and measurement noise covariance matrix parameters; The calculation module is used to adjust the process noise covariance matrix based on the weight dispersion of the particle set at the previous time step. The state vectors of each particle are predicted using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data to obtain a predicted particle set. Based on the current monitoring data, the likelihood of each predicted particle is calculated using the measurement noise covariance matrix parameters in the state vectors of each predicted particle. The weights of each predicted particle are updated by combining the previous weights of each particle with the calculated likelihoods to obtain a weighted predicted particle set. The judgment module is used to determine whether the ratio of the effective number of particles to the total number of particles is less than a first preset threshold, the current medium identification data indicates a change in the medium type, or the rate of change of the output damage geometry parameters within a preset time window is greater than a second preset threshold, and to resample the weighted predicted particle set. The output module is used to output the damage geometry parameters and measurement noise covariance matrix parameters representing the health status of the pipeline by weighted summation of each parameter component in the state vector, based on the particle set after resampling or without resampling.

9. The system according to claim 8, characterized in that, The initialization process involves a set of particles, each carrying a weight and represented by a state vector. The state vector includes the pipe's damage geometry parameters and the measurement noise covariance matrix parameters, including: Set the total number of particles N; For each particle, random samples are taken from a pre-defined uniform distribution of damage geometry parameters to generate initial values ​​for the damage geometry parameters. Set the same initial measurement noise covariance matrix parameters for each particle; The initial weight of each particle is set to 1 / N.

10. The system according to claim 8, characterized in that, The step of predicting the state vector of each particle using the adjusted process noise covariance matrix and the state transition function corresponding to the current medium identification data includes: The current medium identification data includes at least one of the following: medium composition, pH value, sulfur content, sand content, flow rate, and temperature; based on the current medium identification data, the corrosivity level of the current medium is determined: When the corrosion level is lower than a preset level threshold, a first state transition model representing a stable corrosion process is adopted. When the corrosion level is higher than or equal to the preset level threshold, a second state transition model representing an accelerated corrosion process is adopted.

Images