A Filter Flow Anomaly Analysis System and Method Based on Digital Twin

Abstract

This invention discloses a filter flow anomaly analysis system and method based on digital twins, relating to the field of anomaly monitoring technology. The acquisition module periodically collects inlet parameter sets and outlet pressures and calculates cumulative usage time. The twin module substitutes the inlet parameter sets and cumulative usage time into the filter element resistance model and calculates the theoretical outlet pressure using Darcy's law. The anomaly judgment module makes decisions based on aging factors and anomaly scores calculated based on theoretical outlet pressure and outlet pressure. The root cause analysis module processes historical data, determines the probability distribution of causes when anomalies occur based on a root cause classification model, and generates maintenance notifications to facilitate efficient operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of anomaly monitoring technology, specifically to a filter flow anomaly analysis system and method based on digital twins. Background Technology

[0002] Return oil filters are critical components in hydraulic and other fluid transport systems. They are primarily used to filter impurities and particulate contaminants from the return oil lines, maintaining oil cleanliness and ensuring the normal operation of fluid components such as pumps and valves. Abnormal flow rates in the return oil filter directly lead to increased system pressure loss and reduced transport efficiency. Long-term operation may also cause component wear and jamming. Therefore, timely and accurate monitoring and analysis of return oil filter anomalies are crucial for ensuring stable fluid system operation and extending equipment lifespan.

[0003] Existing return oil filter anomaly monitoring technologies have shortcomings: In the filter element resistance calculation stage, the gradient distribution characteristics of aging factors along the filter element axis are not fully considered. The filter element resistance is extrapolated only through a single parameter or simplified model, which cannot reflect the non-uniform state of filter element aging under actual working conditions. This results in a large deviation between the theoretical outlet pressure and the actual value, and the anomaly judgment lacks a precise benchmark. In the anomaly root cause analysis stage, it mostly relies on the static threshold judgment of a single parameter, without combining the time-frequency characteristics of pressure deviation and the dynamic correlation between multiple parameters such as viscosity, flow rate, and temperature for integrated analysis. This makes it difficult to accurately locate the cause of the anomaly, resulting in a lack of clear direction for operation and maintenance work and an inability to efficiently solve problems. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by proposing a filter flow anomaly analysis system and method based on digital twins. It extrapolates aging factors by combining filter element resistance models with accumulated usage time and liquid temperature, obtains permeability and tortuosity distributions using porous media theory, and accurately extrapolates theoretical outlet pressure based on Darcy's law to provide a reliable benchmark for anomaly detection. Furthermore, it extracts pattern coding and time-frequency features of pressure deviation vectors using a root cause classification model, and constructs fusion features using multi-parameter correlation weight vectors to accurately pinpoint the causes of anomalies. This provides clear direction for operation and maintenance work, effectively ensuring the stable operation of the filter and its associated fluid system.

[0005] A filter flow anomaly analysis system based on digital twins includes a data acquisition module, a twin module, an anomaly detection module, and a root cause analysis module.

[0006] The data acquisition module periodically collects and acquires the input parameter set and the output pressure. And simultaneously calculate the cumulative usage time. The inlet parameter set includes inlet pressure. Liquid flow rate Liquid temperature and liquid viscosity ;

[0007] The twin module utilizes a filter resistance model based on cumulative usage time. and liquid temperature Deducing aging factors According to the liquid temperature Adjusting liquid viscosity Obtain the corrected viscosity of the liquid The permeability distribution was obtained based on the porous media theory. and tortuosity distribution Based on Darcy's law, and taking into account filter size and liquid corrected viscosity... Liquid flow rate Permeability distribution and tortuosity distribution Theoretical Filter Resistance , will increase inlet pressure Subtract theoretical filter resistance Theoretical export pressure , For the filter element axial coordinate;

[0008] The anomaly detection module compares aging factors. With aging upper limit Sending replacement notices based on decision-making or based on theoretical export pressure Export pressure Calculate anomaly scores Compare abnormal scores With the upper limit score Choose to wait or send a start command;

[0009] The root cause analysis module receives the start command and calls up the input parameter set and output pressure from the historical cycle. Liquid corrected viscosity and theoretical export pressure And the pressure deviation vector is obtained through processing. Viscosity deviation vector Liquid flow vector and liquid temperature vector Using a root cause classification model, the pressure deviation vector is analyzed. Perform wavelet transform and generate pattern code And time-frequency characteristics, based on viscosity deviation Liquid flow rate and liquid temperature Constructing a correlation weight vector based on correlation and respectively with the viscosity deviation vector Liquid flow vector and liquid temperature vector Perform a dot product, concatenate the three dot product results, and encode the pattern. The fusion features are obtained from the time and frequency features. The modulation mapping is used to obtain the cause probability distribution, and the abnormal cause corresponding to the maximum cause probability is selected to generate an operation and maintenance notification and send it.

[0010] Furthermore, the acquisition module constructs a global time reference based on the NTP protocol. When it receives the oil pump operation signal, it records the start time. After each acquisition cycle, it sends TTL trigger pulses to different sensors to acquire inlet pressure, liquid flow rate, liquid temperature, liquid viscosity, and outlet pressure. It synchronously retrieves historical usage time and adds the current usage time from the start time to the acquisition time to the historical usage time to obtain the cumulative usage time. When the oil pump operation signal is disconnected, it records the shutdown time and replaces the original historical usage time with the sum of the current usage time from the start time to the shutdown time and the historical usage time.

[0011] Furthermore, constructing the filter element resistance model includes the following steps:

[0012] Modeling aging factors using S-curves aging factors The molecule is 1 plus the liquid temperature Compared with standard temperature The difference multiplied by the first temperature coefficient aging factors The denominator is 1 plus the natural constant as the base, and the inflection point of duration is... With cumulative usage time The difference multiplied by the duration factor The power value of the exponent;

[0013] Modeling liquid corrected viscosity based on Einstein's viscosity formula for dilute suspensions Liquid corrected viscosity Equal to the viscosity of the liquid The product of the temperature correction term and the aging correction term, where the temperature correction term equals the standard temperature. With liquid temperature The difference multiplied by the second temperature coefficient Add 1, and the aging correction term equals the aging coefficient. Multiplied by aging factors And add 1;

[0014] Based on porous media theory, utilizing tortuosity Related aging factors and penetration rate tortuosity Equals 1 minus aging factor As the base, with the tortuosity coefficient Multiply the power of the exponent by the initial tortuosity. Combining the Kölzner-Kalman equation, permeability Equal to 1 and aging factor The difference multiplied by the initial permeability ;

[0015] Considering aging factors The influence of gradient distribution along the axial direction, aging distribution Equal to aging factors Multiply by filter length Filter element axial coordinate The difference divided by the filter element length , aging distribution Replacement penetration rate and tortuosity aging factors Permeability distribution was obtained. and tortuosity distribution ;

[0016] Based on Darcy's law, and taking the filter element length as a reference... Filter element cross section Liquid corrected viscosity Liquid flow rate Permeability distribution and tortuosity distribution Construct a filter element resistance model and calculate the theoretical filter element resistance. .

[0017] Furthermore, the anomaly detection module compares aging factors. With aging upper limit When aging factors Greater than the upper limit of aging When the aging factor is activated, a replacement notification is sent. Less than or equal to the upper limit of aging At that time, the theoretical export pressure Export pressure The absolute value of the deviation divided by the theoretical outlet pressure Resulting in abnormal scores and with the upper limit score Comparison, when abnormal scores Less than or equal to the maximum score When the abnormal score occurs, wait for the next collection cycle to arrive. Score greater than the upper limit At that time, a start command is sent.

[0018] Furthermore, the root cause analysis module receives the start command and calls the pre-launch command. The inlet parameter set and outlet pressure for each cycle Liquid corrected viscosity Theoretical filter element resistance ,Will Theoretical export pressure per cycle Export pressure Deviation, liquid corrected viscosity With liquid viscosity Deviation, liquid flow rate and liquid temperature The pressure deviation vector is obtained by sequentially splicing the components. Viscosity deviation vector Liquid flow vector and liquid temperature vector .

[0019] Furthermore, the root cause classification model includes a time-frequency extraction layer, a correlation fusion layer, and a logistic regression layer;

[0020] The time-frequency extraction layer utilizes Morlet wavelets to analyze the pressure deviation vector. Perform wavelet transform to obtain the wavelet coefficient vector. And calculate the mutation index Comparison of mutation indices With upper limit step threshold And determine the step mode label. For wavelet coefficient vectors The power spectrum is obtained by performing a Fourier transform. Obtain peak frequency Peak signal power at and with upper power threshold Comparison to determine the periodic pattern label Step mode label and periodic pattern tags The pattern code is obtained by concatenating the one-hot encoding. Mutation index and peak signal power Integrate into time-frequency features;

[0021] The viscosity deviation vectors of the relevant fusion layers are calculated respectively. Liquid flow vector Liquid temperature vector Construct a correlation matrix by considering the cross-correlation coefficients between each pair of elements and the autocorrelation coefficients among the three elements themselves. , the correlation matrix The associated weight vector is obtained through linear modulation and softmax function mapping. and respectively with the viscosity deviation vector Liquid flow vector Liquid temperature vector By performing a dot product, the weighted viscosity deviation is obtained. Weighted liquid flow rate Weighted liquid temperature , and pattern encoding Mutation index and peak signal power The splicing results in fusion features ;

[0022] The logistic regression layer will fuse features The cause probability distribution is obtained by mapping linear modulation with the Softmax function. The abnormal cause corresponding to the highest cause probability is selected to generate an operation and maintenance notification and send it.

[0023] Furthermore, the time-frequency extraction layer selects wavelet coefficient vectors. The maximum absolute deviation of adjacent wavelet coefficients is used as the mutation index. Comparison of mutation indices With upper limit step threshold When the mutation index Greater than the upper limit step threshold At that time, the step mode label will be displayed. The value is assigned to the mutation index. When the mutation index Less than or equal to the upper limit step threshold At that time, the step mode label will be displayed. Set the value to 0 and compare the peak signal power. With upper power threshold When the peak signal power Greater than the upper power threshold At that time, the periodic pattern label will be displayed. Assigned value: peak frequency The reciprocal of the peak signal power Less than or equal to the upper power threshold At that time, the periodic pattern label will be displayed. The value is assigned to 0.

[0024] The filter flow anomaly analysis method based on digital twins includes the following steps:

[0025] In the current cycle, the inlet parameter set and outlet pressure are collected. And calculate the cumulative usage time. The inlet parameter set includes inlet pressure. Liquid flow rate Liquid temperature and liquid viscosity ;

[0026] Based on the filter element resistance model, combined with cumulative usage time and liquid temperature Deducing aging factors And based on the liquid temperature Adjusting liquid viscosity Obtain the corrected viscosity of the liquid Combining porous media theory with aging factors The influence distribution yields the permeability distribution. and tortuosity distribution Based on Darcy's law, and taking into account filter size and liquid corrected viscosity... Liquid flow rate Permeability distribution and tortuosity distribution Theoretical Filter Resistance And combined with inlet pressure Calculate theoretical outlet pressure , For the filter element axial coordinate;

[0027] Comparison of aging factors With aging upper limit Sending replacement notices based on decision-making or based on theoretical export pressure Export pressure Calculate anomaly scores ;

[0028] Comparison of abnormal scores With the upper limit score To make a decision and wait for the next cycle to arrive or before calling. The inlet parameter set and outlet pressure for each cycle Liquid corrected viscosity and theoretical export pressure And the pressure deviation vector is obtained through processing. Viscosity deviation vector Liquid flow vector and liquid temperature vector Using a root cause classification model from pressure deviation vector Extracting the pattern encoding And time-frequency characteristics, based on viscosity deviation Liquid flow rate and liquid temperature The dynamic cross-correlation yields the correlation weight vector. and respectively with the viscosity deviation vector Liquid flow vector and liquid temperature vector Perform dot products separately, and then combine the three dot product results with the pattern encoding. The fused features are obtained by splicing time and frequency features. The modulation mapping is then transformed into a cause probability distribution. The anomaly cause corresponding to the highest cause probability is selected to generate and send an operation and maintenance notification. This refers to the total number of reference periods.

[0029] The beneficial effects of this invention are as follows:

[0030] A multi-factor coupled filter element resistance model based on digital twins was constructed. By combining the cumulative usage time and liquid temperature, the aging factor was deduced. Combined with porous media theory, the gradient distribution of the aging factor along the filter element axis was considered to obtain the permeability distribution and tortuosity distribution. Based on Darcy's law, the theoretical filter element resistance was deduced and the theoretical outlet pressure was calculated by taking into account parameters such as filter element size and liquid corrected viscosity. This provides an accurate theoretical benchmark that fits the actual working conditions for the judgment of abnormal filter flow, avoiding the one-sidedness of judgment based on a single parameter.

[0031] A multi-dimensional fusion root cause classification model was designed. The pressure deviation vector was subjected to wavelet transform through the time-frequency extraction layer to obtain pattern coding and time-frequency features. Combined with the relevant fusion layer, an associated weight vector was constructed based on the dynamic correlation of viscosity deviation, liquid flow rate and liquid temperature. The vectors of each parameter were weighted and spliced ​​to form a fusion feature, which was then mapped to the cause probability distribution. This enabled accurate identification of the causes of abnormal filter flow, provided clear guidance for operation and maintenance work, and improved the efficiency of fault handling. Attached Figure Description

[0032] Figure 1 This is a diagram of the system architecture for filter flow anomaly analysis based on digital twins in this invention.

[0033] Figure 2 This is a flowchart of the dual progressive determination process in this invention;

[0034] Figure 3 This is a flowchart of the root cause classification model processing in this invention;

[0035] Figure 4 This is a flowchart of the filter flow anomaly analysis method based on digital twins in this invention. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0037] Example 1:

[0038] like Figure 1 As shown, the present invention discloses a filter flow anomaly analysis system based on digital twin, including a data acquisition module, a twin module, an anomaly determination module, and a root cause analysis module;

[0039] The data acquisition module periodically collects and acquires the input parameter set and the output pressure. And simultaneously calculate the cumulative usage time. The inlet parameter set includes inlet pressure. Liquid flow rate Liquid temperature and liquid viscosity ;

[0040] The twin module utilizes a pre-built filter resistance model based on cumulative usage time. and liquid temperature Deducing aging factors According to the liquid temperature and liquid viscosity The coupling correlation adjustment yields the corrected viscosity of the liquid. Combining porous media theory and aging factors The influence gradient distribution mapping yields the permeability distribution and tortuosity distribution Based on Darcy's law, and considering the filter element size and liquid corrected viscosity... Liquid flow rate Permeability distribution and tortuosity distribution Theoretical Filter Resistance , will increase inlet pressure Subtract theoretical filter resistance Theoretical export pressure , This represents the axial coordinate of the filter element, with a maximum value of the filter element length. ;

[0041] The anomaly detection module performs a dual progressive detection, when the aging factor... Greater than the upper limit of aging Immediately send a replacement notification when the aging factor... Less than or equal to the upper limit of aging At that time, based on theoretical export pressure Export pressure Calculate anomaly scores Based on abnormal scores With the upper limit score The relationship decision-making process either silently waits or sends a start command;

[0042] The root cause analysis module receives the start command and calls the pre-launch command. The inlet parameter set and outlet pressure for each cycle Liquid corrected viscosity and theoretical export pressure And the pressure deviation vector is obtained through processing. Viscosity deviation vector Liquid flow vector and liquid temperature vector Using a pre-built root cause classification model, the pressure deviation vector is analyzed. Perform continuous wavelet transform and generate pattern codes Calculate viscosity deviation based on time-frequency characteristics. Liquid flow rate and liquid temperature Correlation matrix And map to obtain the associated weight vector , viscosity deviation vector Liquid flow vector and liquid temperature vector Respectively related weight vectors Perform a dot product, and combine the results of the three dot products with the pattern encoding. The fused features are obtained by concatenating time-frequency features with time-frequency features. , will integrate features Adjust the data to the same dimension as the anomaly cause type and map it to obtain the cause probability distribution. Select the anomaly cause corresponding to the highest cause probability, generate an operation and maintenance notification, and send it. This refers to the total number of reference periods.

[0043] Furthermore, the acquisition module uses a PLC as the main controller and constructs a global time reference based on the NTP protocol. When the oil pump operation signal is received, the start time is recorded. After each preset acquisition cycle, TTL trigger pulses are sent to the external trigger interfaces of the pressure sensor, flow sensor, temperature sensor, and viscometer through the internal signal bus to acquire inlet pressure, liquid flow rate, liquid temperature, liquid viscosity, and outlet pressure. The historical usage time is synchronously retrieved from the PLC's internal register. The current usage time from the start time to the acquisition time is added to the historical usage time to obtain the cumulative usage time. When the oil pump operation signal is disconnected, the stop time is recorded. The sum of the current usage time from the start time to the stop time and the historical usage time is used to replace the original historical usage time in the internal register. The pressure sensor is deployed at the filter element inlet and filter element outlet, respectively, while the flow sensor, temperature sensor, and viscometer are deployed at the filter element inlet.

[0044] Furthermore, the filter element resistance model is improved and established based on the framework of fluid dynamics theory, including the following steps:

[0045] In actual use, filter element aging is a continuous process that gradually accelerates over time, and high temperatures further accelerate the aging of filter materials. Therefore, aging factors are utilized. Describe the filter element aging process, including aging factors. A continuously differentiable S-shaped curve is used to reflect the continuous changing trend of the aging process, as detailed below:

[0046] ,

[0047] in, and These are the first temperature coefficient and the duration coefficient, respectively. This is the inflection point in usage time, when the cumulative usage time... Greater than the inflection point of duration At this time, the filter element enters an accelerated aging state. This is the standard temperature, specifically equal to 25 degrees Celsius. Represents an exponential function with the natural constant as its base;

[0048] Based on Einstein's viscosity formula for dilute suspensions, particles disturb the liquid flow, causing the effective viscosity of the liquid to be higher than the measured viscosity. aging factors Directly affects the filter element's filtration efficiency for particles in liquids, as aging factors... The increase in viscosity reduces the particle filtration effect, and the superposition and disturbance of particles increases the viscosity of the liquid. Increase, and at the same time, the viscosity of the liquid For liquid temperature It is dependent on the liquid temperature. As the viscosity increases, the intermolecular forces in the liquid weaken, and the liquid viscosity increases. Reduce, decrease the viscosity of the liquid Regarding liquid temperature The correction term, combined with Einstein's viscosity formula for dilute suspensions, utilizes the aging factor. and liquid temperature Adjusting liquid viscosity The corrected viscosity of the liquid is obtained. The details are as follows:

[0049] ,

[0050] in, and These are the second temperature coefficient and the aging coefficient, respectively. The second temperature coefficient and the aging coefficient are unknown parameters determined through experimental fitting in the filter element resistance model.

[0051] Based on the porous media theory, the permeability of the filter element and tortuosity The correlation can be established through the Kozeny-Kalman equation, while the tortuosity Used to describe the tortuosity of the fluid flow path, when the aging factor As the filter element is enlarged, the path of the liquid through the filter element becomes more tortuous, thus increasing the tortuosity. Regarding aging factors The expression is to make the tortuosity Substituting into the Körzeny-Kalman equation, based on permeability The penetration rate is obtained by simplifying the merging logic. Regarding aging factors The expression is ,in, The tortuosity index. and The initial permeability and initial tortuosity were determined by Darcy flow experiments and tracer diffusion experiments on unused filter cartridges.

[0052] Because the effects of filter element aging are not uniformly distributed along all axial positions of the filter element, the inlet bears more pressure and experiences the most direct aging effects. The overall impact of filter element aging is distributed in a gradient along the axial direction, with the aging factors increasing closer to the filter outlet. The impact is even less severe, by introducing the filter element axial coordinate. Quantifying aging factors The distribution of aging is affected by the distribution of aging. Specifically as follows:

[0053] ,

[0054] in, and The aging distribution is calculated using the filter element's axial coordinates and length, respectively. Replacement penetration rate and tortuosity aging factors The permeability distribution can then be obtained. and tortuosity distribution ;

[0055] Collaborative reference filter length Filter element cross section Liquid corrected viscosity Liquid flow rate Permeability distribution and tortuosity distribution Construct a filter element resistance model, which indicates the theoretical filter element resistance. Theoretical filter element resistance component Regarding the axial coordinates of the filter element The integrals are as follows:

[0056] ,

[0057] Among them, the theoretical filter element resistance component Based on Darcy's law, the filter element resistance model is used to calculate the theoretical resistance of a filter element under normal conditions and specified operating conditions, providing a standard for judging filter element abnormalities. This is due to aging factors... Permeability distribution and tortuosity distribution Since it cannot be directly observed, it is mapped to observable filter element resistance using Darcy's law. The actual filter element resistance... equal to inlet pressure Reduce export pressure By collecting data on different cumulative usage times The inlet parameter set and outlet pressure of the filter element during actual use The unknown parameters in the filter element resistance model can be determined by fitting the model, including the first temperature coefficient. Duration coefficient Duration inflection point Second temperature coefficient aging coefficient and tortuosity index .

[0058] like Figure 2 As shown, the anomaly detection module further performs a dual progressive detection, including the following steps:

[0059] Obtaining aging factors and aging limit value Compare;

[0060] When aging factors Greater than the upper limit of aging If the filter element is determined to be too worn, a replacement notice should be sent to the staff immediately, regardless of whether any abnormalities have occurred.

[0061] When aging factors Less than or equal to the upper limit of aging At that time, obtain the theoretical export pressure Export pressure Theoretical export pressure Export pressure The absolute value of the deviation divided by the theoretical outlet pressure Resulting in abnormal scores Due to theoretical export pressure The abnormal score is obtained by simulating a filter element in normal condition for a twin module under given operating conditions. It can effectively reflect the degree of deviation of the filter element from its normal state;

[0062] Comparison of abnormal scores Compared to the preset upper limit score When abnormal scores Less than or equal to the maximum score If the filter element is deemed to have no abnormal risk, a silent wait is initiated, awaiting the next data collection cycle.

[0063] When abnormal scores Score greater than the upper limit If the filter element is found to be abnormal, a start command is immediately sent to the root cause analysis module.

[0064] Furthermore, the root cause analysis module receives the start command and calls the pre-launch command. The inlet parameter set and outlet pressure for each cycle Liquid corrected viscosity Theoretical filter element resistance ,Will Theoretical export pressure per cycle Export pressure The pressure deviation vector is obtained by concatenating the deviations according to the time sequence. ,Will Liquid corrected viscosity per cycle With liquid viscosity The viscosity deviation vector is obtained by concatenating the deviations according to the time sequence. ,Will Liquid flow rate per cycle and liquid temperature Liquid flow vectors are obtained by concatenating columns according to time sequence. and liquid temperature vector .

[0065] like Figure 3 As shown, the root cause classification model further includes a time-frequency extraction layer, a correlation fusion layer, and a logistic regression layer;

[0066] The time-frequency extraction layer extracts the pressure deviation vector. As the primary object of analysis, Morlet wavelets were used to analyze the pressure deviation vector. Perform a continuous wavelet transform to obtain the wavelet coefficient vector. And calculate the mutation index Comparison of mutation indices With upper limit step threshold And determine the step mode label. For wavelet coefficient vectors The power spectrum is obtained by performing a Fourier transform. Comparing power spectra At peak frequency Peak signal power at With upper power threshold And determine the periodic pattern label Step mode label and periodic pattern tags The pattern code is obtained by converting the unique hot code and concatenating the columns. Simultaneously, the mutation index and peak signal power Integrate into time-frequency features;

[0067] The relevant fusion layer will affect the viscosity deviation. Liquid flow rate and liquid temperature As an auxiliary analysis object, the viscosity deviation vector is calculated respectively. With liquid flow vector viscous flow cross-correlation coefficients between Viscosity deviation vector With liquid temperature vector Viscosity-temperature cross-correlation coefficient Liquid flow vector With liquid temperature vector The cross-correlation coefficient between flow and temperature Combining and constructing dimensions Correlation matrix Correlation matrix It is a symmetric matrix, with the diagonal elements being the viscosity deviation vectors. Liquid flow vector Liquid temperature vector The autocorrelation coefficient of the product itself, and the off-diagonal lines represent the cross-correlation coefficients between each pair of products, and the correlation matrix. Effectively reflects the dynamic relationships between auxiliary analysis objects, and the correlation matrix The dimension is obtained through linear modulation and Softmax function mapping transformation. The associated weight vector , associate weight vector Respectively with viscosity deviation vector Liquid flow vector Liquid temperature vector The weighted viscosity deviation is obtained by taking the dot product and summing the weighted values. Weighted liquid flow rate Weighted liquid temperature , and pattern encoding Mutation index and peak signal power The fusion feature is obtained by splicing columns together. fusion features Not only reflects pressure deviation The time-frequency variation trend also effectively reflects the viscosity deviation. Liquid flow rate and liquid temperature The temporal relationship between them;

[0068] The logistic regression layer will fuse features The dimension is adjusted to be the same as the anomaly cause type by linear modulation, and the cause probability distribution is obtained by mapping through the Softmax function. The anomaly cause corresponding to the highest cause probability is selected to generate an operation and maintenance notification and send it. The anomaly cause types include measurement error, system fluctuation, fluid impurity mutation and filter blockage.

[0069] Furthermore, due to pressure deviation The pressure deviation vector directly reflects fluid abnormalities in the filter. As the primary analysis object, the time-frequency extraction layer selects wavelet coefficient vectors. The maximum absolute deviation of adjacent wavelet coefficients is used as the mutation index. Comparison of mutation indices With upper limit step threshold When the mutation index Greater than the upper limit step threshold At that time, the step mode label will be displayed. The value is assigned to the mutation index. When the mutation index Less than or equal to the upper limit step threshold At that time, the step mode label will be displayed. Similarly, assigning a value of 0 allows for comparison of peak signal power. With upper power threshold When the peak signal power Greater than the upper power threshold At that time, the periodic pattern label will be displayed. Assigned value: peak frequency The reciprocal of the peak signal power Less than or equal to the upper power threshold At that time, the periodic pattern label will be displayed. The value is assigned to 0.

[0070] Example 2:

[0071] like Figure 4 As shown, this invention discloses a filter flow anomaly analysis method based on digital twins, implemented based on the aforementioned filter flow anomaly analysis system based on digital twins, and includes the following steps:

[0072] In the current cycle, the inlet parameter set and outlet pressure are collected. And calculate the cumulative usage time. The inlet parameter set includes inlet pressure. Liquid flow rate Liquid temperature and liquid viscosity ;

[0073] Based on the filter element resistance model, combined with cumulative usage time and liquid temperature Deducing aging factors And based on the liquid temperature Adjusting liquid viscosity Obtain the corrected viscosity of the liquid Combining porous media theory and aging factors The influence gradient distribution mapping yields the permeability distribution and tortuosity distribution Based on Darcy's law, and taking into account filter size and liquid corrected viscosity... Liquid flow rate Permeability distribution and tortuosity distribution Theoretical Filter Resistance And combined with inlet pressure Calculate theoretical outlet pressure , For the filter element axial coordinate;

[0074] When aging factors Greater than the upper limit of aging When the aging factor is activated, a replacement notification is sent. Less than or equal to the upper limit of aging At that time, based on theoretical export pressure Export pressure Calculate anomaly scores ;

[0075] When abnormal scores Less than or equal to the maximum score Then, wait for the next cycle to arrive;

[0076] When abnormal scores Score greater than the upper limit When, before calling The inlet parameter set and outlet pressure for each cycle Liquid corrected viscosity and theoretical export pressure And the pressure deviation vector is obtained through processing. Viscosity deviation vector Liquid flow vector and liquid temperature vector Using a root cause classification model from pressure deviation vector Extracting the pattern encoding And time-frequency characteristics, based on viscosity deviation Liquid flow rate and liquid temperature The dynamic cross-correlation yields the correlation weight vector. , viscosity deviation vector Liquid flow vector and liquid temperature vector Based on the correlation weight vector respectively Weighted summation and pattern encoding The fused features are obtained by concatenating time-frequency features with time-frequency features. , will integrate features The modulation mapping is based on the causal probability distribution. The anomaly cause corresponding to the highest causal probability is selected to generate an operation and maintenance notification and send it. This refers to the total number of reference periods.

[0077] This invention discloses a filter flow anomaly analysis system and method based on digital twins. The acquisition module constructs a global time reference based on the NTP protocol and periodically acquires the inlet parameter set and outlet pressure through TTL trigger pulses, while simultaneously calculating the cumulative usage time, ensuring the synchronization and accuracy of data acquisition and providing a reliable data foundation for subsequent analysis. The twin module uses the filter element resistance model, combined with the cumulative usage time and liquid temperature, to deduce the aging factor. Based on the liquid temperature, the liquid viscosity is adjusted to obtain the liquid corrected viscosity. Combining porous media theory and the axial gradient distribution of the aging factor, the permeability distribution and tortuosity distribution are obtained. Then, based on Darcy's law, the theoretical filter element resistance is deduced and the theoretical outlet pressure is calculated, which closely matches the non-uniform state of filter element aging under actual working conditions, making the theoretical value more consistent with the actual working conditions and providing an accurate benchmark for anomaly judgment. The anomaly detection module first compares the aging factor with the aging limit. If it exceeds the aging limit, a replacement notification is sent. If it is less than or equal to the aging limit, an anomaly score is calculated and compared with the limit score. A decision is then made to wait or send a start command, achieving a dual progressive judgment. This avoids the continued use of over-aged filters and reduces false judgments, improving the rationality of anomaly detection. After receiving the start command, the root cause analysis module calls historical periodic data to process and obtain pressure deviation vectors, viscosity deviation vectors, etc. It uses a root cause classification model to extract the pattern encoding and time-frequency features of the pressure deviation vector. Combined with a multi-parameter correlation matrix, it obtains the correlation weight vector and weights the multi-parameter vector. After splicing and fusing the features, it maps the cause probability distribution and generates an operation and maintenance notification. This achieves multi-dimensional fusion analysis, accurately locates the cause of the anomaly, provides clear guidance for operation and maintenance work, and improves fault handling efficiency.

[0078] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A filter flow anomaly analysis system based on digital twins, characterized in that, It includes a twin module, an anomaly detection module, and a root cause analysis module; The twin module utilizes a filter element resistance model to deduce aging factors based on cumulative usage time and liquid temperature, and combines porous media theory to obtain permeability and tortuosity distribution. It adjusts the liquid viscosity according to the liquid temperature to obtain the corrected viscosity. Based on Darcy's law, it combines filter element size, corrected liquid viscosity, liquid flow rate, permeability distribution, and tortuosity distribution to deduce theoretical filter element resistance. The theoretical outlet pressure is obtained by subtracting the theoretical filter element resistance from the inlet pressure. The anomaly detection module compares the aging factor with the aging upper limit value to decide whether to send a replacement notification or calculate an anomaly score based on the theoretical outlet pressure and the outlet pressure. It then compares the anomaly score with the upper limit score and chooses to wait or send a start command. The root cause analysis module receives the start command and organizes the pressure deviation vector, viscosity deviation vector, liquid flow vector, and liquid temperature vector based on historical periodic data. Using the root cause classification model, it performs wavelet transform on the pressure deviation vector and generates pattern coding and time-frequency features. Based on the correlation mapping of viscosity deviation, liquid flow, and liquid temperature, it obtains the associated weight vector and performs dot product with the corresponding vectors respectively. It concatenates the dot product results, pattern coding, and time-frequency features to obtain fusion features and modulates the mapping to generate the cause probability distribution. It selects the abnormal cause corresponding to the highest cause probability, generates the operation and maintenance notification, and sends it. In the filter element resistance model, the aging factor is modeled using an S-curve, the liquid corrected viscosity is modeled based on Einstein's formula for the viscosity of dilute suspensions, and the liquid corrected viscosity is equal to the product of the liquid viscosity and the temperature correction term and the aging correction term. The permeability and tortuosity are obtained by relating the aging factor to the porous media theory. The permeability distribution and tortuosity distribution are obtained by replacing the aging factor with an aging factor distribution that is gradient-distributed along the filter element axis.

2. The filter flow anomaly analysis system based on digital twin as described in claim 1, characterized in that, In the aging factor modeled by the S-curve, the numerator is 1 plus the difference between the liquid temperature and the standard temperature multiplied by the first temperature coefficient, and the denominator is 1 plus the power value with the natural constant as the base and the difference between the inflection point of time and the cumulative usage time multiplied by the time coefficient as the exponent.

3. The filter flow anomaly analysis system based on digital twin as described in claim 1, characterized in that, The temperature correction term for liquid viscosity correction is the difference between the standard temperature and the liquid temperature multiplied by the second temperature coefficient plus 1. The aging correction term is the aging coefficient multiplied by the aging factor plus 1. The second temperature coefficient and the aging coefficient are unknown parameters determined by experimental fitting in the filter element resistance model.

4. The filter flow anomaly analysis system based on digital twin as described in claim 1, characterized in that, Based on the porous media theory, tortuosity is equal to the power of the tortuosity coefficient with 1 minus the aging factor as the base multiplied by the initial tortuosity, and permeability is equal to the difference between 1 and the aging factor multiplied by the initial permeability. The axial gradient distribution of the aging factor is the aging factor multiplied by the difference between the filter element length and the filter element axial coordinate, and then divided by the filter element length. The maximum value of the filter element axial coordinate is the filter element length.

5. The filter flow anomaly analysis system based on digital twin as described in claim 1, characterized in that, The anomaly score calculated by the anomaly determination module is the absolute value of the deviation between the theoretical outlet pressure and the outlet pressure divided by the theoretical outlet pressure. The corresponding calculation operation is performed when the aging factor is less than or equal to the upper limit of the aging value. When the abnormal score is less than or equal to the upper limit score, the system waits for the next collection cycle; when the abnormal score is greater than the upper limit score, the abnormal judgment module sends a start command.

6. The filter flow anomaly analysis system based on digital twin as described in claim 1, characterized in that, The root cause classification model includes a time-frequency extraction layer, a correlation fusion layer, and a logistic regression layer. The time-frequency extraction layer uses Morlet wavelet to perform wavelet transform on the pressure deviation vector and generates pattern codes and time-frequency features. The correlation fusion layer constructs a correlation matrix and maps it to obtain the correlation weight vector. The logistic regression layer modulates and maps the fused features to generate the cause probability distribution.

7. The filter flow anomaly analysis system based on digital twin as described in claim 6, characterized in that, The correlation matrix of the correlation fusion layer consists of the cross-correlation coefficients between each pair of viscosity deviation vector, liquid flow rate vector, and liquid temperature vector, as well as the autocorrelation coefficients of the three. The correlation weight vector is obtained by linear modulation of the correlation matrix and mapping using the Softmax function. The correlation weight vector is then dot-producted with the viscosity deviation vector, liquid flow rate vector, and liquid temperature vector, respectively. The dot-product results are concatenated with the pattern coding and time-frequency features to form the fusion features.

8. The filter flow anomaly analysis system based on digital twin as described in claim 1, characterized in that, It also includes a data acquisition module; The acquisition module synchronizes time based on the NTP protocol. When it receives the oil pump operation signal, it records the start time. In each acquisition cycle, it collects inlet pressure, liquid flow rate, liquid temperature, liquid viscosity and outlet pressure through pulse-triggered sensors. It calls up the historical usage time and adds the current usage time from the start time to the acquisition time to the historical usage time to obtain the cumulative usage time. When the oil pump operation signal is disconnected, it records the shutdown time and replaces the original historical usage time with the sum of the current usage time from the start time to the shutdown time and the historical usage time.

9. A filter flow anomaly analysis method based on digital twins, implemented based on the filter flow anomaly analysis system based on digital twins as described in any one of claims 1-8, characterized in that, Includes the following steps: The current cycle acquires inlet pressure, liquid flow rate, liquid temperature, liquid viscosity, outlet pressure, and cumulative usage time. Using the filter element resistance model, the aging factor is derived based on the cumulative usage time and liquid temperature, and the permeability distribution and tortuosity distribution are obtained by combining the porous media theory. The liquid viscosity is adjusted according to the liquid temperature to obtain the liquid corrected viscosity. Based on Darcy's law, the theoretical filter element resistance is derived by combining the filter element size, liquid corrected viscosity, liquid flow rate, permeability distribution and tortuosity distribution. The theoretical outlet pressure is obtained by subtracting the theoretical filter element resistance from the inlet pressure. Compare the aging factor with the upper limit of aging to decide whether to send a replacement notice or calculate anomaly scores based on theoretical export pressure and export pressure. The abnormal score is compared with the upper limit score to decide whether to wait for the next cycle or to organize the pressure deviation vector, viscosity deviation vector, liquid flow vector, and liquid temperature vector based on historical cycle data. Using the root cause classification model, wavelet transform is performed on the pressure deviation vector to generate pattern coding and time-frequency features. Based on the correlation mapping of viscosity deviation, liquid flow, and liquid temperature, the associated weight vector is obtained and dot product is performed with the corresponding vectors respectively. The dot product result, pattern coding, and time-frequency features are concatenated to obtain fusion features and modulated mapping to generate cause probability distribution. The abnormal cause corresponding to the highest cause probability is selected to generate maintenance notification and send it.

Images