Method and system for predicting the wear life of a sealing packing

Abstract

The present application relates to the technical field of sealing state monitoring and life prediction, and particularly relates to a sealing packing wear life prediction method and system. The method obtains historical operation data, historical maintenance event data and historical life result data in multiple maintenance periods, identifies bidirectional paired samples and extracts a direction separation feature group, respectively establishes reversible sealing loss labels and irreversible wear labels, trains a sealing packing fault prediction and health management model, and then constructs an equivalent wear age according to a current reversible sealing loss state, a current irreversible wear state and an operation cumulative amount, performs an unclosed condition projection, and outputs a target residual life value and a maintenance decision window. The method is used for improving the pertinence of sealing packing life prediction and maintenance decision.

Description

Technical Field

[0001] This invention relates to the field of sealing condition monitoring and life prediction technology, and in particular to a method and system for predicting the wear life of sealing packings. Background Technology

[0002] Sealing packings are widely used in valves, reciprocating compressors, piston pumps, and other equipment with reciprocating or rotating moving parts to form a seal between the moving parts and stationary components. During long-term operation, these seals are typically affected by various factors such as medium pressure, temperature, displacement, frequency of movement, gland adjustment, and maintenance operations, leading to decreased sealing capacity, increased leakage, and accumulated wear. Therefore, assessing the operating status of sealing packings and predicting their remaining lifespan has become an important technical aspect in equipment condition monitoring, predictive maintenance, and health management.

[0003] In the prior art, there are already solutions for predicting the remaining life of seals. For example, CN108398249B discloses a "method and apparatus, engineering machinery and server for predicting the remaining life of a sealing ring", which collects operating data related to the seal, including information such as pressure, oil temperature and piston displacement, further extracts feature values, and determines the remaining life of the sealing ring based on the feature values.

[0004] However, in practical applications of sealing packings, in addition to the cumulative wear caused by continuous operation, maintenance events such as retightening and replacement are frequently encountered. For sealing packings, some of the reduced sealing capacity can be restored through maintenance operations such as retightening, while another part stems from the continuous wear or degradation of the packing body itself, which is difficult to recover through maintenance operations. Existing technologies typically focus on establishing a predictive relationship directly between operating data and lifespan results, lacking an effective distinction between the two types of state changes mentioned above, and easily mixing the maintenance-recoverable sealing losses with the unrecoverable actual wear. As a result, in the presence of maintenance intervention, the lifespan prediction results may be affected in reflecting the actual degradation state of the sealing packing, which is not conducive to the reasonable judgment of subsequent maintenance timing.

[0005] Therefore, the existing technology has at least the following technical problems: how to distinguish between the maintenance-recoverable sealing loss and the unrecoverable real wear in sealing packing application scenarios where maintenance events such as re-tightening and replacement occur, so as to avoid the two from being mixed and interfering with the remaining life characterization. Summary of the Invention

[0006] To overcome the aforementioned technical deficiencies, the present invention aims to provide a method and system for predicting the wear life of sealing packings. This invention extracts directional separation feature groups from bidirectional paired samples within the same axial segment, and establishes reversible seal loss labels and irreversible wear labels by combining re-tightening events and replacement events. It then constructs an equivalent wear age using sealing packing failure prediction and health management models that output the two types of states respectively, and performs projection of the un-re-tightening condition. This achieves a distinguishable characterization of maintainable recoverable seal loss and irreversible true wear, avoiding interference from the mixing of the two types in the remaining life characterization.

[0007] This invention discloses a method for predicting the wear life of sealing packing. The method is applied to equipment with moving rods passing through the sealing packing assembly. The method includes:

[0008] S1. Obtain historical operating data, historical maintenance event data, and historical life result data of the sealing packing assembly within multiple maintenance cycles, and mark the re-tightening events and replacement events in the historical maintenance event data within the corresponding maintenance cycles;

[0009] S2. From the historical operation data of each maintenance cycle, identify bidirectional paired samples in which the moving rods pass through the same axial section and move in opposite directions, and extract the direction separation feature group from each bidirectional paired sample.

[0010] S3. Based on historical maintenance event data and historical lifespan result data, establish reversible sealing loss labels and irreversible wear labels for the separation feature groups in each direction, and construct a training sample set;

[0011] S4. Train a machine learning-based sealing packing failure prediction and health management model using the training sample set. The sealing packing failure prediction and health management model includes at least a first prediction branch for outputting the reversible sealing loss state and a second prediction branch for outputting the irreversible wear state.

[0012] S5. Obtain the current operating data, identify the current bidirectional pairing samples in the current operating data, extract the current direction separation feature group, input the current direction separation feature group into the sealing packing failure prediction and health management model, and obtain the current reversible seal loss state and the current irreversible wear state.

[0013] S6. Based on the current reversible seal loss state, the current irreversible wear state, and the cumulative amount of operation since the most recent re-tightening event, construct the equivalent wear age corresponding to the current maintenance cycle;

[0014] S7. Based on the equivalent wear age, perform untightened condition projection to obtain the target remaining life value, and output the maintenance decision window according to the current reversible seal loss state and the current irreversible wear state.

[0015] Preferably, the bidirectional paired samples are two operating samples that meet the following conditions: they are located within the same maintenance cycle; they pass through the same axial section; their corresponding movement directions are opposite; the difference in their corresponding average medium pressure difference is not greater than a preset pressure difference threshold; and the difference in their corresponding average stuffing box temperature is not greater than a preset temperature difference threshold.

[0016] Preferably, identifying bidirectional paired samples includes: dividing the entire stroke of the moving rod into multiple axial segments; determining the axial segment number to which each sampling time period belongs based on the displacement signal; pairing sampling time periods with the same segment number and opposite motion directions; eliminating sampling time periods where the medium pressure difference fluctuation rate and stuffing box temperature fluctuation rate exceed their respective preset thresholds; and determining the screened pairing results as bidirectional paired samples.

[0017] Preferably, the directional separation feature group includes at least: the difference in off-seat starting force corresponding to the bidirectional paired sample, the difference in stable frictional work density corresponding to the bidirectional paired sample, the leakage recovery hysteresis corresponding to the bidirectional paired sample, and the difference in pressure drop rate corresponding to the stop pressure holding of the bidirectional paired sample.

[0018] Preferably, establishing reversible seal loss labels includes: for maintenance cycles in which a re-tightening event occurs, extracting bidirectional paired samples located in the same axial segment before and after the re-tightening event; and determining the reversible seal loss labels of the corresponding training samples based on the leakage change before and after re-tightening, the recovery of the off-seat starting force, and the decrease in stable friction work density.

[0019] Preferably, establishing irreversible wear labels includes: for maintenance cycles in which replacement events occur, sorting bidirectional paired samples according to the cumulative amount of operation within the maintenance cycle; determining the wear state corresponding to the end sample before the replacement event as the terminal irreversible wear state; determining the wear state corresponding to the new filler reference sample after the replacement event as the reset reference state; and determining the irreversible wear label of each training sample based on the terminal irreversible wear state and the reset reference state.

[0020] Preferably, the sealing packing failure prediction and health management model includes: a shared feature encoding layer for feature mapping of directional separation feature groups; a first prediction branch for outputting reversible sealing loss state; and a second prediction branch for outputting irreversible wear state; wherein the first prediction branch and the second prediction branch share the output of the shared feature encoding layer.

[0021] Preferably, when training the sealing packing failure prediction and health management model, the model also includes applying a periodic trajectory consistency constraint. The periodic trajectory consistency constraint is used to ensure that samples before and after the re-tightening event within the same maintenance cycle correspond to the same irreversible wear trajectory after deducting the influence of reversible seal loss.

[0022] Preferably, before inputting the current direction separation feature group into the sealing packing failure prediction and health management model, the method further includes performing local sample adaptive correction; the local sample adaptive correction is performed based on historical training samples that have the same packing material category, the same gland structure category, the same moving rod surface roughness level, and the same medium category as the equipment.

[0023] Preferably, after obtaining new actual maintenance results, an incremental update is performed on the sealing packing failure prediction and health management model; wherein, only samples that meet any of the following conditions are written into the incremental training set: the samples corresponding to the re-tightening event have a change in gland displacement and a decrease in leakage after re-tightening; or, the samples corresponding to the replacement event have a baseline state reset characteristic after replacement; wherein, the samples corresponding to the re-tightening event are used to update the first prediction branch, and the samples corresponding to the replacement event are used to update the second prediction branch.

[0024] Preferably, constructing the equivalent wear age includes: taking the initial state after the most recent replacement event as the starting point of the wear age; accumulating the basic wear age based on the current irreversible wear state; performing re-tightening compensation on the basic wear age based on the current reversible seal loss state after the most recent re-tightening event; and determining the compensated result as the equivalent wear age.

[0025] Preferably, performing the projection of the untightened operating condition includes: without introducing subsequent tightening events, extrapolating forward along the irreversible wear trajectory determined based on the current irreversible wear state, based on the estimated future cumulative amount of the operating condition corresponding to the current bidirectional paired sample; determining the earlier of the leakage threshold arrival time and the wear threshold arrival time based on the extrapolation results; and obtaining the target remaining lifetime value based on the earlier time.

[0026] Preferably, the maintenance decision window includes a re-tightening decision window and a replacement decision window; wherein, when the reversible seal loss state reaches a first preset threshold and the irreversible wear state does not reach a second preset threshold, the re-tightening decision window is output; when the irreversible wear state reaches the second preset threshold, the replacement decision window is output.

[0027] Preferably, before outputting the maintenance decision window, the method further includes calculating the prediction confidence level, which is determined by at least two of the following: the coupling deviation between the outputs of the first prediction branch and the second prediction branch, the sample density between the current bidirectional paired sample and the corresponding sample in the training sample set, and the coverage integrity of the current bidirectional paired sample in each axial segment; when the prediction confidence level is lower than the preset confidence level threshold, only a restrictive maintenance prompt is output.

[0028] In view of this, the present invention also provides a sealing packing wear life prediction system, the system being applied to a device having a moving rod passing through a sealing packing assembly, the system comprising:

[0029] The data acquisition module is used to acquire historical operation data, historical maintenance event data, historical lifespan result data, and current operation data;

[0030] The pairing identification module is used to identify bidirectional paired samples that are located within the same maintenance cycle, pass through the same axial section, and move in opposite directions based on historical operation data, and to identify the current bidirectional paired sample based on current operation data.

[0031] The feature extraction module is used to extract directional separation feature groups from bidirectional paired samples and extract current directional separation feature groups from the current bidirectional paired samples.

[0032] The tag building module is used to build reversible seal loss tags and irreversible wear tags based on historical maintenance event data and historical lifespan result data, and to build a training sample set by combining directional separation feature groups.

[0033] The model building module is used to train a machine learning-based sealing packing failure prediction and health management model using a training sample set. The sealing packing failure prediction and health management model includes at least a first prediction branch and a second prediction branch.

[0034] The state prediction module is used to input the current direction separation feature group into the sealing packing failure prediction and health management model to output the current reversible seal loss state and the current irreversible wear state.

[0035] The equivalent wear age construction module is used to construct the equivalent wear age based on the current reversible seal loss state, the current irreversible wear state, and the cumulative amount of operation since the most recent re-tightening event.

[0036] The untightened condition projection and maintenance decision module is used to perform untightened condition projection based on the equivalent wear age, obtain the target remaining life value, and output the maintenance decision window according to the current reversible seal loss state and the current irreversible wear state.

[0037] Compared with existing technologies, the above technical solution has the following advantages:

[0038] 1. This invention constructs bidirectional paired samples within the same axial segment, extracts directional separation feature groups, and establishes reversible seal loss labels and irreversible wear labels by combining re-tightening events and replacement events, respectively. This allows for the distinguishing characterization of two different sources of degradation in the sealing packing during operation. Compared to methods that mix and model various degradation factors, this invention can more realistically reflect the actual degradation state of the sealing packing, thereby solving the problem of the coupling and difficulty in distinguishing between maintenance-recoverable factors and actual wear accumulation.

[0039] 2. This invention does not directly establish a single lifespan mapping relationship based on general operating data. Instead, it first characterizes directional friction behavior, leakage recovery behavior, and pressure holding behavior, and then combines a dual-branch fault prediction and health management model to output the current reversible seal loss state and the current irreversible wear state, respectively, and constructs an equivalent wear age based on these. Therefore, the lifespan prediction results more closely match the actual wear evolution process of the sealing packing, reducing the impact of maintenance interventions such as re-tightening events on lifespan estimation fluctuations.

[0040] 3. This invention not only outputs the target remaining life value, but also outputs a re-tightening decision window and a replacement decision window based on the current reversible seal loss state and the current irreversible wear state, respectively. This allows for differentiation between a decrease in sealing capacity that can be recovered through re-tightening and an accumulation of irreversible wear requiring replacement, thus avoiding a general maintenance judgment based solely on a single remaining life value and improving the accuracy and specificity of maintenance action selection.

[0041] 4. Based on the construction of the equivalent wear age, this invention further performs a projection of the untightened operating condition, using the earlier of the leakage threshold arrival time and the wear threshold arrival time as the target remaining life value. Therefore, it can provide a more conservative upper limit maintenance life result without introducing subsequent tightening events, which is more conducive to on-site preventative maintenance arrangements and spare parts replacement planning.

[0042] 5. This invention introduces local sample adaptive correction, consistent trajectory constraints within the same period, and a prediction reliability evaluation mechanism, enabling the model to maintain good adaptability under different filler material types, different gland structure types, different surface roughness levels of moving rods, and different media types. Simultaneously, when prediction reliability is insufficient, only restrictive maintenance prompts are output, thereby reducing the risk of false alarms and misjudgments and improving the stability and reliability of the system in engineering applications.

[0043] 6. The data types used in this invention are mainly derived from information obtainable during equipment operation, such as displacement, temperature, differential pressure, leakage, gland displacement, and pressure holding at the stop position. This facilitates implementation in valves, reciprocating compressors, plunger pumps, and other equipment with moving rod sealing structures. Furthermore, the method flow of this invention is clear and can be implemented by industrial controllers or edge computing devices, as well as by servers or cloud platforms, thus possessing strong engineering feasibility and promising prospects for widespread application. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the overall structure of a method for predicting the wear life of sealing packings according to the present invention.

[0045] Figure 2 This is a schematic diagram illustrating bidirectional paired sample identification and direction separation feature group extraction.

[0046] Figure 3 This diagram illustrates the construction of a labeling and training model for sealing packing failure prediction and health management based on re-tightening and replacement events.

[0047] Figure 4 This is a schematic diagram of online prediction, equivalent wear age construction, projection of untightened operating conditions, and maintenance decision output.

[0048] Figure 5 This is a schematic diagram showing the separation and change curves of the current reversible seal loss state and the current irreversible wear state within the same maintenance cycle.

[0049] Figure 6 This is a schematic diagram comparing the remaining lifetime prediction of the scheme in this embodiment and the comparative scheme.

[0050] Figure 7 The curves represent the relationship between leakage changes, wear threshold, and target remaining life value under the untightened operating condition projection. Detailed Implementation

[0051] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0053] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0054] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0055] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0056] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0057] In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the convenience of the description of the invention and have no specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.

[0058] The following detailed description, in conjunction with the accompanying drawings, illustrates a specific implementation of a method for predicting the wear life of sealing packing. In this embodiment, the method is applied to a device having a moving rod passing through the sealing packing assembly. Exemplarily, this device is a high-pressure reciprocating valve stem sealing device, the moving rod being a valve stem. The sealing packing assembly is disposed within a stuffing box, and a pressure cap is disposed outside the stuffing box. The pressure cap is capable of axially pressing the sealing packing assembly in the event of a re-tightening event. It should be noted that the experimental data, curves, and tables in the following embodiments are used to illustrate the working mechanism, implementation process, and technical effects of the present invention, with the aim of enabling those skilled in the art to clearly and completely implement the present invention.

[0059] like Figure 1 As shown, the method in this embodiment is executed according to the following process: historical data preparation, bidirectional paired sample identification, directional separation feature group extraction, reversible seal loss label construction, irreversible wear label construction, seal packing failure prediction and health management model training, online condition prediction, equivalent wear age construction, untightened condition projection, and maintenance decision output. Figure 1In the overall structural diagram shown, the upper left side represents the historical data preparation and sample construction path, the upper right side represents the label construction and model training path, and the lower right side represents the online inference, equivalent wear age construction, and maintenance decision output path. These three constitute a complete closed-loop technical link. To ensure consistency in the technical names used in subsequent descriptions, this embodiment uniformly uses the following terms: "historical operating data," "historical maintenance event data," "historical lifespan result data," "current operating data," "bidirectional paired samples," "directional separation feature group," "reversible seal loss label," "irreversible wear label," "training sample set," "sealing packing failure prediction and health management model," "current reversible seal loss state," "current irreversible wear state," "equivalent wear age," "target remaining lifespan value," and "maintenance decision window."

[0060] In this embodiment, historical operating data, historical maintenance event data, and historical lifespan result data are first acquired across multiple maintenance cycles. Retightening events and replacement events from the historical maintenance event data are then marked within their respective maintenance cycles. Here, a maintenance cycle refers to the continuous operating phase from the moment the new packing is put into operation after a replacement event until the next replacement event occurs. A retightening event occurs within a maintenance cycle but does not terminate the cycle; after a replacement event, the previous maintenance cycle ends, and the next maintenance cycle begins. Historical operating data includes at least valve stem displacement, valve stem speed, stuffing box temperature, media differential pressure, leakage, gland displacement, and pressure timing data during the stop-and-hold phase. Current operating data is acquired using the same sampling method and data type as historical operating data. For example, the sampling period is 0.1 seconds, the valve stem displacement is collected by a displacement sensor, the valve stem speed is obtained by valve stem displacement differential, the stuffing box temperature is collected by a temperature sensor attached to the outer wall of the stuffing box, the medium pressure difference is collected and calculated by pressure sensors set on both sides of the sealing area, the leakage is converted by the leakage collection chamber flow meter, the gland displacement is collected by a gland displacement sensor, and the pressure during the stop and pressure holding stage is continuously collected by a stop stage pressure sensor.

[0061] In this embodiment, instead of directly inputting all historical operational data into the model, bidirectional paired samples are first identified. For example... Figure 2 As shown, the entire stroke of the moving rod is first divided into multiple axial segments. For example, when the valve stem's total stroke is 400 mm, it can be divided into 8 axial segments, each 50 mm long. Figure 2This paper illustrates the pairing method of samples in opposite directions within the same axial segment, and the relationship between these bidirectional paired samples and the four directional separation feature groups. This demonstrates that this embodiment does not directly use the original time-series data, but first constructs a sample base with spatial consistency and directional comparability. Subsequently, the axial segment number of each sampling period is determined based on the valve stem displacement, and the corresponding movement direction is determined based on the sign of the valve stem velocity. For two operating samples within the same maintenance cycle that pass through the same axial segment and have opposite movement directions, they are considered as candidate sample pairs. The average medium pressure difference and average stuffing box temperature difference of the candidate sample pairs are further calculated. Only candidate sample pairs with an average medium pressure difference not exceeding a preset pressure difference threshold and an average stuffing box temperature difference not exceeding a preset temperature difference threshold are retained, while candidate sample pairs with medium pressure fluctuation rates and stuffing box temperature fluctuation rates exceeding their respective preset thresholds are removed, thus obtaining bidirectional paired samples. Through the above processing, it is possible to ensure that the two operating data segments within the same bidirectional paired sample have high comparability in terms of spatial location, operating cycle, and main boundary conditions, providing a basis for separating recoverable sealing losses and irrecoverable real wear during maintenance. For example, the preset differential pressure threshold can be from 0.05 MPa to 0.30 MPa, and in this embodiment, it is 0.15 MPa; the preset temperature difference threshold can be from 1 degree Celsius to 5 degrees Celsius, and in this embodiment, it is 3 degrees Celsius; the medium differential pressure fluctuation rate threshold can be from 5% to 12%, and in this embodiment, it is 8%; the stuffing box temperature fluctuation rate threshold can be from 3% to 8%, and in this embodiment, it is 5%.

[0062] To clearly understand the source of the aforementioned thresholds, in this embodiment, the preset differential pressure threshold, preset temperature threshold, media differential pressure fluctuation rate threshold, and stuffing box temperature fluctuation rate threshold are all obtained from historical normal operation samples. Specifically, firstly, normal operation samples without abnormal leakage, without unplanned shutdowns, and with stable conditions before and after maintenance are selected from historical operation data. Then, the statistical distributions of the average media differential pressure difference, average stuffing box temperature difference, media differential pressure fluctuation rate, and stuffing box temperature fluctuation rate between candidate sample pairs are calculated respectively. Finally, the 95th percentile of each distribution is taken as the corresponding preset threshold. This ensures that the vast majority of normal samples can enter the subsequent pairing process, while effectively eliminating incomparable samples caused by excessive deviations in boundary conditions.

[0063] In this embodiment, a directional separation feature set is extracted for each bidirectional paired sample. This directional separation feature set includes at least the difference in starting force upon departure, the difference in stable frictional work density, the leakage recovery hysteresis, and the difference in pressure drop rate during stationary holding pressure. To ensure clarity, the calculation formulas are explained below.

[0064] The difference in starting force after leaving the seat is denoted as the difference in starting force after leaving the seat. The formula for its calculation is:

[0065]

[0066] in, This indicates the starting force at the moment of positive start-up. This represents the starting force upon reversing the engine. The difference in starting force upon reversing the engine. It is used to characterize the degree of asymmetry in starting resistance in opposite directions within the same axial section.

[0067] The steady friction work density difference is denoted as the steady friction work density difference. The formula for its calculation is:

[0068]

[0069] in, This represents the frictional work of a positive sample during the steady-state operation phase. This represents the frictional work of the reverse sample during the steady-state operation phase. This represents the distance traveled by the positive samples. This represents the running distance of the reverse sample. The difference in stable frictional work density. Used to characterize the difference in frictional energy consumption per unit operating distance.

[0070] The leakage recovery hysteresis is denoted as the leakage recovery hysteresis. The formula for its calculation is:

[0071]

[0072] in, This indicates the recovery time required for the leakage to return to the preset recovery threshold after the positive operation ends. This indicates the recovery time required for the leakage to return to the preset recovery threshold after the reverse run is completed. Leakage recovery hysteresis. This is used to characterize the difference in re-adhesion of the sealing surface under opposite directions. In this embodiment, the preset recovery threshold is taken as 1.10 times the average leakage of nearly 30 normal samples in the corresponding axial section. The preset recovery threshold can also be other values ​​in the range of 1.05 times to 1.20 times. In this embodiment, 1.10 times is selected as a value that takes into account both recovery sensitivity and noise tolerance.

[0073] The difference in pressure drop rate during the stop-holding phase is denoted as the pressure drop rate difference during the stop-holding phase. The formula for its calculation is:

[0074]

[0075] in, and These represent the pressure values ​​at two sampling moments during the forward stop and pressure holding phase. and These represent the pressure values ​​at two sampling moments during the reverse stop and pressure holding phase. , , and These represent the corresponding sampling times. The difference in the rate of pressure drop during the stop-hold period. Used to characterize the difference in seal attenuation during the pressure holding phase.

[0076] To enable those skilled in the art to implement this directly, a set of specific calculation examples are given below. Suppose that in a bidirectional pairing sample of a certain axial section, the starting force at the forward start is 186 N, and the starting force at the reverse start is 143 N. Then the difference in starting force is... for:

[0077]

[0078] Let the frictional work during the steady-state operation of the forward sample be 152 joules and the forward travel distance be 380 mm. Let the frictional work during the steady-state operation of the reverse sample be 108 joules and the reverse travel distance be 360 ​​mm. Then, what is the difference in the density of the steady-state frictional work? for:

[0079]

[0080] Assuming the recovery time required for the leakage to return to the preset recovery threshold after the forward operation is 26 seconds, and the recovery time required for the leakage to return to the preset recovery threshold after the reverse operation is 15 seconds, then the leakage recovery hysteresis is... for:

[0081]

[0082] Let the pressure during the forward stop-holding phase be 5.42 MPa and 5.28 MPa at 0 seconds and 20 seconds respectively, and the pressure during the reverse stop-holding phase be 5.39 MPa and 5.33 MPa at 0 seconds and 20 seconds respectively. Then, the pressure drop rate difference during the stop-holding phase is... for:

[0083]

[0084] Therefore, the above four types of characteristics can characterize the degradation differences of sealing packing under opposite conditions from four perspectives: starting force, friction work, leakage recovery, and pressure holding at the stop.

[0085] In this embodiment, after obtaining the directional separation feature set, reversible seal loss tags and irreversible wear tags are further constructed based on the re-tightening event and the replacement event. For the reversible seal loss tag, bidirectional paired samples in the same axial segment and located before and after the re-tightening event are selected, and the leakage change, the recovery of the off-seating starting force, and the decrease in stable frictional work density before and after the re-tightening event are calculated respectively. The leakage change is denoted as the leakage change amount. The formula for its calculation is:

[0086]

[0087] in, This indicates the average leakage amount before re-tightening. This indicates the average leakage amount after re-tightening.

[0088] The amount of force recovery after dismounting is denoted as the amount of force recovery after dismounting. The formula for its calculation is:

[0089]

[0090] in, This indicates the difference in starting force before and after the seat is tightened. This indicates the difference in starting force after re-tightening and disengagement.

[0091] The amount of steady friction work density decrease is denoted as the steady friction work density decrease. The formula for its calculation is:

[0092]

[0093] in, This represents the difference in stable frictional work density before re-tightening. This represents the difference in stable frictional work density after re-tightening.

[0094] In this embodiment, the leakage change amount , recovery amount of starting force after leaving seat and the amount of stable frictional work density decrease After standardization, reversible seal loss labels are constructed using a weighted method. These reversible seal loss labels are denoted as reversible seal loss labels. The formula for its calculation is:

[0095]

[0096] in, , and These represent the standardized values ​​of the corresponding quantities. , and Represents the weighting coefficient, and .

[0097] In order to ensure that the weighting coefficients have a clear source, in this embodiment, the weighting coefficients... , and The validation set optimization method is used to determine the optimal solution. Specifically, under the condition that... Given that all weight coefficients are non-negative, a grid search is performed with a step size of 0.05, resulting in a reversible sealing loss label. The reversible sealing loss prediction error is minimized on the validation set after training. For example, after searching, the following is selected: , , Correspondingly, The selectable range is from 0.20 to 0.60. The selectable range is from 0.10 to 0.40. The selectable range is 0.10 to 0.40, as long as the sum of the three is 1.

[0098] For example, suppose the average leakage rate before and after a re-tightening event is 8.2 ml / min and 3.6 ml / min, respectively; the difference in the start-up force before and after re-tightening is 43 N and 17 N, respectively; and the difference in the steady-state frictional work density before and after re-tightening is 0.100 J / mm and 0.036 J / mm, respectively. Further suppose the standardized leakage change... The standardized recovery amount of the off-seat starting force is 0.82. The normalized stable frictional work density decreases by 0.65. The weighting coefficient is 0.58. , and Taking values ​​of 0.45, 0.30, and 0.25 respectively, the reversible seal loss label... for:

[0099]

[0100] This indicates that the sample had a significant recoverable seal loss before re-tightening.

[0101] For irreversible wear tags, in this embodiment, the maintenance cycle in which the replacement event occurs is selected, and the bidirectional paired samples are sorted according to the cumulative running volume within that maintenance cycle. Let the irreversible wear tag at the beginning of the maintenance cycle be 0, and the irreversible wear tag corresponding to the last sample before replacement be 1. Then, interpolation is performed on intermediate samples based on their cumulative running volume position to obtain the irreversible wear tag. For example, if the total cumulative running volume of a maintenance cycle is 80 kilometers, and the cumulative running volume corresponding to the current sample is 52 kilometers, then when using linear interpolation, the irreversible wear tag is recorded as 0. The formula for its calculation is:

[0102]

[0103] Therefore, irreversible wear labels It characterizes the irreversible wear process of the current sample throughout the entire maintenance cycle.

[0104] like Figure 3 As shown, after obtaining the directional separation feature group, reversible seal loss label and irreversible wear label, a seal packing failure prediction and health management model is further constructed. Figure 3 The left side shows the effect of re-tightening event samples and replacement event samples on label construction; the middle shows the shared feature encoding layer; the right side shows the first and second prediction branches; and the bottom shows the role of the same-period trajectory consistency constraint in the training process. This illustrates that this embodiment does not use a single-output model, but rather a dual-branch structure to respectively handle the prediction of two types of degradation states. The sealing packing failure prediction and health management model includes a shared feature encoding layer, a first prediction branch, and a second prediction branch. The shared feature encoding layer is used for feature mapping of the directional separation feature group. The first prediction branch outputs the current reversible sealing loss state, and the second prediction branch outputs the current irreversible wear state. During model training, in addition to the conventional error term, a same-period trajectory consistency constraint is introduced to ensure that samples before and after the re-tightening event still fall on the same irreversible wear trajectory after deducting the influence of reversible sealing loss. Let the total training loss be denoted as the total training loss. The prediction error of the reversible sealing loss corresponding to the first prediction branch is denoted as the reversible sealing loss prediction error. The irreversible wear prediction error corresponding to the second prediction branch is denoted as the irreversible wear prediction error. Consistent constraint terms for trajectories with the same period are denoted as consistent constraint terms for trajectories with the same period. The constraint weights are denoted as constraint weights. Then the total training loss for:

[0105]

[0106] Among them, the consistency constraint term of the trajectory in the same period This constraint prevents non-physical jumps in the output of the second prediction branch for samples before and after a re-tightening event within the same maintenance cycle. This approach allows the model output to better reflect the physical mechanism of packing re-tightening, which alters recoverable seal loss without changing the actual wear trajectory. Furthermore, in this embodiment, the supervision target of the first prediction branch is the reversible seal loss label. The monitoring target of the second prediction branch is irreversible wear labels. During the online operation phase, the output of the first prediction branch is the current reversible sealing loss state. The output of the second prediction branch is the current irreversible wear state. Therefore, the labels in the training phase and the status in the online phase are distinct from each other in terms of both technical meaning and symbol usage.

[0107] In this embodiment, the training sample set, validation sample set, and test sample set are divided into 3024 groups, 648 groups, and 648 groups respectively. Furthermore, samples within each maintenance cycle are divided according to the entire maintenance cycle, avoiding simultaneous allocation of samples from the same maintenance cycle to different sets to prevent data leakage. During model training, the orientation separation feature set is used as input, and the reversible sealing loss label is used. and irreversible wear labels As supervised output, training is performed using a mini-batch iterative approach. For example, the number of samples per batch can be 32 to 128, and in this embodiment, 64; the maximum number of training epochs can be 100 to 300, and in this embodiment, 200; the initial learning rate can be 0.0001 to 0.01, and in this embodiment, 0.001. The total training loss on the validation sample set is then used. If the model fails to decrease its parameters for 20 consecutive rounds, training is terminated, and the model parameters with the minimum total training loss on the validation set are selected as the final model parameters. (Constraint weights) In this embodiment, it can be determined by optimizing the verification sample set, for example in The value that minimizes the error of the validation sample set is selected from the given values. For example, a value could be selected as... The selectable range is from 0.1 to 2.0. Those skilled in the art can then train the model accordingly. Furthermore, the shared feature encoding layer can employ a multi-layer fully connected network, a convolutional network, or other network structures capable of feature mapping. The first and second prediction branches can each employ a single-output regression head. Model training can be completed on a general-purpose server, an industrial edge computing device, or a cloud training platform.

[0108] In this embodiment, to improve on-site adaptability, local sample adaptive correction is performed before inputting the current directional separation feature group into the sealing packing fault prediction and health management model. The basic idea of ​​local sample adaptive correction is to select a subset of local samples from historical training samples that share the same packing material type, gland structure type, moving rod surface roughness level, and media type as the equipment. The statistical distribution of this local sample subset is then used to perform scale balancing and bias correction on the current directional separation feature group. In this embodiment, local sample adaptive correction is implemented using a mean-variance correction method. That is, for each feature quantity in the current directional separation feature group, it is standardized according to the mean and standard deviation of the corresponding local sample subset, and then restored to a unified model input scale based on the mean and standard deviation of the global training samples. This method reduces the input domain offset caused by differences in packing material, gland structure, valve stem surface roughness, and media type.

[0109] During the online operation phase, the system first acquires the current operating data, identifies the current bidirectional paired samples and extracts the current directional separation feature group according to the aforementioned method. After local sample adaptive correction, the current directional separation feature group is input into the sealing packing failure prediction and health management model to obtain the current reversible seal loss state. and the current irreversible wear state Current reversible seal loss state This reflects the portion of the current decrease in sealing capability that can be recovered through a re-tightening event, indicating the current irreversible wear state. It is used to reflect the irrecoverable parts caused by material degradation and structural wear at the current moment.

[0110] Given the current reversible sealing loss state and the current irreversible wear state Then, the equivalent wear age is further constructed. See [link / reference] Figure 4 In this embodiment, the equivalent wear age is used to uniformly characterize the combined effects of "irreversible wear accumulation" and "reversible seal loss attenuation after re-tightening" on the current degradation state. Figure 4 The diagram illustrates how, after the current directional separation feature group enters the model, it outputs the current reversible seal loss state and the current irreversible wear state, respectively. These are then combined with the cumulative operating data since the most recent re-tightening event to construct the equivalent wear age, ultimately relating to the projection of the non-re-tightening condition and the maintenance decision output. The basic wear age is denoted as the basic wear age. The amount of re-tightening compensation is recorded as the re-tightening compensation amount. The equivalent wear age is recorded as the equivalent wear age. The calculation formulas are as follows:

[0111]

[0112]

[0113]

[0114] in, This represents the cumulative amount of data generated since the most recent replacement event. This indicates the current state of irreversible wear and tear. This indicates the current state of reversible seal loss. This represents the cumulative amount of operations since the most recent tightening event. , , and These represent the coefficients obtained by fitting historical samples. This represents the natural exponential function.

[0115] To clarify the source of the fitting coefficients, in this embodiment, the coefficients are... , , and The remaining lifespan is determined using a combination of least squares fitting and validation set correction. Specifically, firstly, the target remaining lifespan value is backfitted using the actual maintenance time, cumulative operation amount, current reversible seal loss state, and current irreversible wear state from the training sample set to obtain initial coefficients; then, these initial coefficients are substituted into the validation sample set to calculate the mean absolute error of the target remaining lifespan, and finally, the set of coefficients with the smallest error is selected as the final coefficients. For example, in this embodiment, the following can be used: Every kilometer per day sky, sky, Kilometers. Correspondingly, The selectable range is 0.03 to 0.15 days per kilometer. The available time range is 10 to 30 days. The available time range is 5 to 15 days. The selectable range is 10 to 40 kilometers.

[0116] For example, let's say the cumulative runtime since the most recent replacement event. The current wear and tear is 52 kilometers and is in an irreversible state. The current reversible seal loss state is 0.65. The cumulative amount since the most recent tightening event is 0.709. It is 6 kilometers, and the coefficient is 6. , , and Taking 0.08 days per kilometer, 18 days, 10 days, and 20 kilometers respectively, the foundation wear age is then... , Tightening compensation amount and equivalent wear age They are respectively:

[0117]

[0118]

[0119]

[0120] Therefore, the equivalent wear age corresponding to the current sample It takes 21.11 days.

[0121] After obtaining the equivalent wear age Next, the untightened condition projection is further executed. The untightened condition projection means that no new tightening events will be introduced in future operation phases, but only the current irreversible wear state will be considered. The determined irreversible wear trajectory is extrapolated forward to obtain a more conservative remaining lifespan that is more suitable for maintenance planning. To perform this extrapolation, it is first necessary to obtain an estimate of the future cumulative operating volume. Here, the estimate of the future cumulative operating volume is not estimated independently of the current operating conditions, but is determined based on the operating conditions corresponding to the current bidirectional pairing sample. Specifically, historical operating segments with the same or similar operating conditions as the current bidirectional pairing sample are first selected from historical operating data. The operating conditions are determined by at least the axial section where the current bidirectional pairing sample is located, the average medium differential pressure range corresponding to the current bidirectional pairing sample, the average stuffing box temperature range corresponding to the current bidirectional pairing sample, and the operating frequency range corresponding to the current bidirectional pairing sample. Then, the daily cumulative operating volume of this type of historical operating segment within the most recent statistical window is calculated, thereby obtaining an estimate of the future cumulative operating volume that matches the operating conditions corresponding to the current bidirectional pairing sample. Let the estimate of the future cumulative operating volume be denoted as the future cumulative operating volume estimate. ,recent The cumulative operating volume under the same operating conditions for each statistical day is as follows: , , , Then the estimated cumulative amount of future operation for:

[0122]

[0123] The statistical window can be the most recent 5 to 30 days; in this embodiment, the most recent 7 days are used. The value can range from 5 to 30; in this embodiment, 7 is chosen. Subsequently, based on the estimated cumulative future operating amount... Extrapolating forward along the irreversible wear trajectory, while also considering the equivalent wear age. Projecting future leakage changes. Let the time when the leakage threshold is reached be denoted as the leakage threshold arrival time. The time when the wear threshold is reached is denoted as the wear threshold arrival time. The target remaining lifetime value is denoted as the target remaining lifetime value. Then the target's remaining lifetime value for:

[0124]

[0125] For example, if the leakage threshold is calculated to arrive at the time... The wear threshold is reached after 12 days. If the remaining lifespan is 18 days, then the target's remaining lifespan value is... for:

[0126]

[0127] This indicates that, without the introduction of new re-tightening events, the remaining maintenance lifespan corresponding to the current state is 12 days.

[0128] In this embodiment, based on the current reversible seal loss state and the current irreversible wear state Further output the maintenance decision window. A first preset threshold is used to determine the current reversible seal loss state. Whether the required tightening level has been reached, the second preset threshold is used to determine the current irreversible wear state. Has the level reached the point where replacement is required? This is based on the current state of reversible seal loss. The first preset threshold has been reached and the current state of irreversible wear is reached. If the second preset threshold is not reached, output the tightening decision window; when the current irreversible wear state... When the second preset threshold is reached, a replacement decision window is output. Therefore, maintenance actions no longer simply depend on a remaining lifetime value, but rather correspond to specific sources of degradation.

[0129] To ensure that the first and second preset thresholds have a clear basis, in this embodiment, the first and second preset thresholds are jointly determined using a validation sample set. Specifically, with the goal of maximizing the hit rate of reinforcement suggestions and the hit rate of replacement suggestions while minimizing the false alarm rate, different threshold combinations are traversed through the validation sample set, and the threshold combination with the optimal comprehensive evaluation index is finally selected as the official threshold. For example, the first preset threshold can be 0.60, and the second preset threshold can be 0.75. Correspondingly, the selectable range of the first preset threshold can be from 0.45 to 0.75, and the selectable range of the second preset threshold can be from 0.60 to 0.90.

[0130] To prevent drawing overly strong conclusions under conditions of insufficient sample support, this embodiment also introduces a prediction confidence level. Let the prediction confidence level be denoted as prediction confidence level. The coupling deviation score is denoted as the coupling deviation score. The sample density score is denoted as the sample density score. The coverage completeness score is recorded as the coverage completeness score. The weighting coefficients are denoted as follows: , and Then the prediction confidence for:

[0131]

[0132] in, For example, when the coupling deviation score... The sample density score is 0.84. The coverage score is 0.76. It is 0.88, and the weighting coefficient is... , and When the values ​​are 0.40, 0.30, and 0.30 respectively, the prediction confidence level is... for:

[0133]

[0134] If the preset confidence threshold is 0.75, then due to the prediction confidence... A value greater than 0.75 indicates a normal maintenance decision window can be output; if the prediction confidence level is... If the score is below the preset confidence threshold, only a restrictive maintenance prompt will be output, such as suggesting a shorter review cycle or recommending manual review. The preset confidence threshold can also be determined by iterating through the verification sample set, and can be 0.75 for example, with a selectable range of 0.60 to 0.85.

[0135] During long-term model operation, incremental updates are performed on the sealing packing failure prediction and health management model upon acquiring new actual maintenance results. If an actual maintenance result corresponds to a re-tightening event, and it can be confirmed that the gland displacement has changed and the leakage has decreased after re-tightening, then this sample is added to the incremental training set to update the first prediction branch. If an actual maintenance result corresponds to a replacement event, and it can be confirmed that the sample exhibits baseline state reset characteristics after replacement, then this sample is added to the incremental training set to update the second prediction branch. In this way, the model can continuously absorb new maintenance samples during long-term use, while ensuring that the direction of sample updates is consistent with its physical meaning.

[0136] In one abnormal operating condition embodiment, if the average medium pressure difference value corresponding to the current bidirectional paired sample exceeds a preset pressure difference threshold, or the average stuffing box temperature difference value exceeds a preset temperature difference threshold, then the sample is considered not to meet the operating condition consistency requirements and will not enter the directional separation feature group calculation process; instead, it will be directly discarded as an incomparable sample. In another abnormal operating condition embodiment, if the current bidirectional paired sample meets the pressure difference and temperature difference conditions, but its medium pressure difference fluctuation rate or stuffing box temperature fluctuation rate exceeds the corresponding threshold, then the current sample is in a state of significant disturbance, and it will also not enter the directional separation feature group calculation process. In a boundary operating condition embodiment, if the coverage integrity of the current bidirectional paired sample in each axial segment is insufficient, for example, only covering fewer samples than the preset minimum number of segments, then although the system can still output the current reversible seal loss state... and the current irreversible wear state The estimated value, but due to the low confidence level of the prediction Often, the values ​​fall below the preset confidence threshold, therefore only a restrictive maintenance prompt is output, without a clear re-tightening or replacement decision window. In another boundary condition embodiment, if the gland displacement changes after a re-tightening event, but the leakage does not decrease after re-tightening, or the difference in the starting force and the difference in stable frictional work density do not show a decline after re-tightening, it indicates that the maintenance event lacks positive recovery evidence. In this case, the sample is not written into the incremental training set to avoid invalid maintenance events interfering with the update of the first prediction branch. In yet another boundary condition embodiment, if the sample does not show baseline state reset characteristics after a replacement event, such as the leakage, the difference in the starting force, and the difference in stable frictional work density remaining high, the event is not written into the incremental training set used to update the second prediction branch, thereby ensuring the sample quality of incremental updates.

[0137] To more intuitively illustrate the technical effects of the present invention, exemplary experimental data and curves are given below. Table 1 shows the experimental platform and sample composition.

[0138] Table 1 Experimental Platform and Sample Composition

[0139]

[0140] Table 1 shows that the sample size used in this embodiment covers multiple maintenance cycles, and the number of re-tightening events and replacement events is sufficient to support the construction of reversible seal loss labels, irreversible wear labels, and model training, verification, and testing.

[0141] Table 2 provides a set of example raw data used for feature calculation and label construction.

[0142] Example Original Data Table 2

[0143]

[0144] Table 2 shows that the calculation of the separation feature groups in each direction in this embodiment, as well as the construction of reversible sealing loss labels, irreversible wear labels, and equivalent wear ages, can all be obtained step by step from the original sampling data, with clear data sources and traceability.

[0145] Table 3 presents a comparison of the performance of the proposed solution and the comparative solution on the test set. It should be noted that the comparative solution, following its publicly disclosed technical approach of "directly predicting remaining lifetime after extracting general features from operational data," was constructed using the same experimental platform, the same historical operational data, the same training / validation / test set partitioning method, and the same target remaining lifetime evaluation method as the proposed solution. In other words, the main difference lies in the fact that the proposed solution separates and characterizes recoverable seal loss and irrecoverable real wear, while the comparative solution does not perform this separation, thus ensuring comparability and fairness in the comparison. Furthermore, all results in Table 3 are the average values ​​obtained after three independent training and testing sessions on the same test set. The mean absolute error and mean relative error of the target remaining lifetime are statistically obtained based on the actual maintenance time corresponding to the samples in the test set. The hit rate of the re-tightening suggestion and the hit rate of the replacement suggestion are statistically obtained based on subsequent actual and confirmed effective re-tightening or replacement events in the test set. The false alarm rate in low-confidence scenarios is statistically obtained based on the proportion of samples that should have output restrictive maintenance prompts but mistakenly output explicit maintenance decision windows.

[0146] Table 3 compares the results of this embodiment with the comparative embodiment.

[0147]

[0148] Table 3 shows that, under the same test platform and data conditions, the scheme in this embodiment outperforms the comparative scheme in terms of mean absolute error of target remaining life, mean relative error, hit rate of re-tightening recommendation, hit rate of replacement recommendation, and false alarm rate in low confidence scenarios. This indicates that the technical approach in this embodiment that separates and characterizes the maintenance of recoverable seal loss from unrecoverable real wear has a significant effect.

[0149] like Figure 5 As shown, the current reversible seal loss state within the same maintenance cycle. A significant decline occurred after the re-tightening event, while the current state of irreversible wear... Then it will continue its monotonous upward trend, and before and after the re-tightening event... The magnitude of the change is significantly greater than The variation range indicates that this embodiment can characterize recoverable sealing losses separately from irrecoverable actual wear, rather than combining the two into a single degradation amount. For example... Figure 6As shown, the remaining life prediction curve of this embodiment is closer to the actual remaining life curve. The comparative embodiment exhibits more significant life estimation fluctuations in the stage following the re-tightening event, indicating that without separating and characterizing maintenance recoverable factors, the re-tightening event interferes with life characterization. Figure 7 As shown, under the untightened operating condition projection, the predicted leakage curve reaches the preset maintenance upper limit before the wear threshold curve, therefore the target remaining life value... Driven by leakage risk, this result demonstrates that this embodiment can provide a basis for developing more conservative maintenance plans in the field.

[0150] From the above implementation process, calculation examples, tabular data, and curve comparisons, it can be seen that this embodiment has at least the following beneficial effects: First, by using bidirectional paired samples and directional separation feature groups, it is possible to extract feature information that is closer to the degradation mechanism from the directional friction behavior, leakage recovery behavior, and stationary pressure holding behavior of the sealing packing; Second, by constructing reversible seal loss labels respectively... and irreversible wear labels By combining the first and second prediction branches for modeling, it is possible to separate and characterize the maintenance recoverable seal loss from the unrecoverable real wear, avoiding the interference of the two on the remaining life characterization; thirdly, by constructing an equivalent wear age And by performing the unre-tightened condition projection, the target remaining lifetime value can be... Fourth, it better meets the requirements for maintaining upper limit control; Joint output can improve the pertinence, robustness, and engineering applicability of maintenance decisions.

[0151] In addition, in other embodiments, the moving rod can also be a piston rod, plunger rod, or shaft; the cumulative running amount can be the cumulative running distance, the cumulative running time, or a weighted combination of the cumulative running distance and the cumulative running time; in addition to the four types of features in this embodiment, the directional separation feature group can also add media pulsation related features, gland displacement response related features, or sealing surface contact stiffness related features according to the equipment type. As long as it still revolves around the separation characterization of recoverable sealing loss and irrecoverable real wear, the technical idea of ​​this invention can be adopted.

[0152] Furthermore, the historical operation data acquisition unit, bidirectional paired sample identification unit, directional separation feature group extraction unit, tag construction unit, model training and inference unit, equivalent wear age construction unit, untightened condition projection unit, and maintenance decision output unit in this embodiment can respectively implement the historical data preparation step, bidirectional paired sample identification step, directional separation feature group extraction step, reversible seal loss tag construction step, irreversible wear tag construction step, model training step, online state prediction step, equivalent wear age construction step, untightened condition projection step, and maintenance decision output step in the above method implementation, thereby providing corresponding support for subsequent system implementations.

[0153] In another embodiment, the present invention can also constitute a sealing packing wear life prediction system. The sealing packing wear life prediction system includes a data acquisition module, a pairing identification module, a feature extraction module, a tag construction module, a model construction module, a state prediction module, an equivalent wear age construction module, and an untightened condition projection and maintenance decision module. Specifically, the data acquisition module is used to acquire historical operating data, historical maintenance event data, historical life result data, and current operating data; the pairing identification module is used to identify bidirectional paired samples based on maintenance cycle, axial section number, operating direction, average medium pressure difference, and average stuffing box temperature difference; the feature extraction module is used to extract directional separation feature groups; and the tag construction module is used to construct reversible seal loss tags based on tightening events. And build irreversible wear tags based on replacement events. The model building module is used to train the sealing packing failure prediction and health management model; the state prediction module is used to output the current reversible seal loss state. and the current irreversible wear state The equivalent wear age construction module is used to determine the current reversible seal loss state. Current irreversible wear and tear state And the equivalent wear age is constructed from the cumulative operating volume following the most recent tightening event. The untightened condition projection and maintenance decision module is used to base maintenance decisions on equivalent wear age. Perform untightened condition projection to obtain the target remaining lifetime value. The system outputs a maintenance decision window. In this embodiment, the above modules can be implemented using hardware circuits, software programs, firmware logic, or any combination thereof, or they can be implemented using industrial controllers, edge computing devices, servers, or cloud platforms. As long as they can complete the corresponding functions in the aforementioned method flow, they all belong to the system implementation method of this invention.

[0154] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for predicting the wear life of sealing packing, characterized in that, The method is applied to a device having a moving rod passing through a sealing packing assembly, and the method includes: S1. Obtain historical operating data, historical maintenance event data, and historical life result data of the sealing packing assembly in multiple maintenance cycles, and mark the re-tightening event and replacement event in the historical maintenance event data in the corresponding maintenance cycle; S2. From the historical operation data of each maintenance cycle, identify bidirectional paired samples in which the moving rod passes through the same axial section and moves in opposite directions, and extract directional separation feature groups from each bidirectional paired sample. S3. Based on the historical maintenance event data and the historical lifespan result data, establish reversible sealing loss labels and irreversible wear labels for each of the directional separation feature groups, and construct a training sample set; S4. Use the training sample set to train a machine learning-based sealing packing failure prediction and health management model. The sealing packing failure prediction and health management model includes at least a first prediction branch for outputting reversible sealing loss state and a second prediction branch for outputting irreversible wear state. S5. Obtain the current operating data, identify the current bidirectional pairing samples in the current operating data, and extract the current direction separation feature group. Input the current direction separation feature group into the sealing packing failure prediction and health management model to obtain the current reversible seal loss state and the current irreversible wear state. S6. Based on the current reversible seal loss state, the current irreversible wear state, and the cumulative amount of operation since the most recent re-tightening event, construct the equivalent wear age corresponding to the current maintenance cycle; S7. Based on the equivalent wear age, perform untightened condition projection to obtain the target remaining life value, and output the maintenance decision window according to the current reversible seal loss state and the current irreversible wear state.

2. The method for predicting the wear life of sealing packings according to claim 1, characterized in that, The bidirectional paired samples are two operating samples that meet the following conditions: they are located within the same maintenance cycle; they pass through the same axial section; their corresponding directions of movement are opposite; the difference in their corresponding average medium pressure difference is not greater than a preset pressure difference threshold; and the difference in their corresponding average stuffing box temperature is not greater than a preset temperature difference threshold.

3. The method for predicting the wear life of sealing packings according to claim 2, characterized in that, Identifying the bidirectional paired samples includes: dividing the entire stroke of the moving rod into multiple axial segments; determining the axial segment number to which each sampling time period belongs based on the displacement signal; pairing sampling time periods with the same segment number and opposite motion directions; eliminating sampling time periods where the medium pressure difference fluctuation rate and stuffing box temperature fluctuation rate exceed their respective preset thresholds; and determining the filtered pairing results as the bidirectional paired samples.

4. The method for predicting the wear life of sealing packings according to claim 2, characterized in that, The directional separation feature group includes at least: the difference in starting force corresponding to the bidirectional pairing sample, the difference in stable frictional work density corresponding to the bidirectional pairing sample, the leakage recovery hysteresis corresponding to the bidirectional pairing sample, and the difference in pressure drop rate corresponding to the stop pressure holding of the bidirectional pairing sample.

5. The method for predicting the wear life of sealing packings according to claim 4, characterized in that, Establishing the reversible seal loss label includes: for maintenance cycles in which a re-tightening event occurs, extracting bidirectional paired samples located in the same axial segment before and after the re-tightening event; and determining the reversible seal loss label of the corresponding training sample based on the leakage change before and after re-tightening, the recovery of the off-seat starting force, and the decrease in stable friction work density.

6. The method for predicting the wear life of sealing packings according to claim 4, characterized in that, Establishing the irreversible wear label includes: for maintenance cycles in which a replacement event occurs, sorting bidirectional paired samples according to the cumulative amount of operation within the maintenance cycle; determining the wear state corresponding to the end sample before the replacement event as the terminal irreversible wear state; determining the wear state corresponding to the new filler reference sample after the replacement event as the reset reference state; and determining the irreversible wear label for each training sample based on the terminal irreversible wear state and the reset reference state.

7. The method for predicting the wear life of sealing packings according to claim 1, characterized in that, The sealing packing failure prediction and health management model includes: a shared feature encoding layer for feature mapping of the directional separation feature group; a first prediction branch for outputting the reversible sealing loss state; and a second prediction branch for outputting the irreversible wear state; wherein the first prediction branch and the second prediction branch share the output of the shared feature encoding layer.

8. The method for predicting the wear life of sealing packings according to claim 7, characterized in that, When training the sealing packing failure prediction and health management model, the method also includes applying a periodic trajectory consistency constraint. The periodic trajectory consistency constraint is used to ensure that samples before and after the re-tightening event within the same maintenance cycle correspond to the same irreversible wear trajectory after deducting the influence of reversible seal loss.

9. The method for predicting the wear life of sealing packings according to claim 7, characterized in that, Before inputting the current direction separation feature group into the sealing packing failure prediction and health management model, the method further includes performing local sample adaptive correction; the local sample adaptive correction is performed based on historical training samples that have the same packing material category, the same gland structure category, the same moving rod surface roughness level, and the same medium category as the device.

10. A system for predicting the wear life of sealing packings, characterized in that, The system is applied to a device having a moving rod passing through a sealing packing assembly, and the system includes: The data acquisition module is used to acquire historical operation data, historical maintenance event data, historical lifespan result data, and current operation data; The pairing identification module is used to identify bidirectional pairing samples that are located within the same maintenance cycle, pass through the same axial segment, and move in opposite directions based on the historical operation data, and to identify the current bidirectional pairing sample based on the current operation data. The feature extraction module is used to extract directional separation feature groups from the bidirectional pairing samples and extract current directional separation feature groups from the current bidirectional pairing samples; The tag construction module is used to establish reversible seal loss tags and irreversible wear tags based on the historical maintenance event data and the historical lifespan result data, and to construct a training sample set in combination with the directional separation feature group; The model building module is used to train a machine learning-based sealing packing failure prediction and health management model using the training sample set. The sealing packing failure prediction and health management model includes at least a first prediction branch and a second prediction branch. The state prediction module is used to input the current direction separation feature group into the sealing packing failure prediction and health management model to output the current reversible seal loss state and the current irreversible wear state. The equivalent wear age construction module is used to construct the equivalent wear age based on the current reversible seal loss state, the current irreversible wear state, and the cumulative amount of operation since the most recent re-tightening event. The untightened working condition projection and maintenance decision module is used to perform untightened working condition projection based on the equivalent wear age to obtain the target remaining life value, and output the maintenance decision window according to the current reversible seal loss state and the current irreversible wear state.

Images