Rotating machinery full life cycle maintenance management system based on digital twinning

Abstract

The application relates to a rotating machine full-life cycle maintenance management system based on digital twinning, and relates to the technical field of digital twinning maintenance management, and comprises a data acquisition and virtual-real synchronization module, which is used for acquiring the real-time operation sequence and the operation state feature set of a target rotating machine and the current virtual state of a corresponding digital twinning body; a health state trajectory prediction module, which is used for generating a predicted health state trajectory; a maintenance value evaluation calculation module, which is used for inputting the predicted health state trajectory into a maintenance value evaluation model, combining a risk discount compensation mechanism, and calculating an expected maintenance value; and a maintenance decision optimization generation module, which is used for taking maximizing total value income and minimizing total maintenance downtime as targets, combining maintenance resource constraints, generating a maintenance decision instruction set, and outputting the maintenance decision instruction set to a terminal. The application solves the problem that a traditional maintenance management method relies on artificial experience and reactive processing of maintenance plans, and causes the maintenance work to always lag behind the equipment deterioration process.

Description

Technical Field

[0001] This application relates to the field of digital twin maintenance management, and in particular to a digital twin-based full lifecycle maintenance management system for rotating machinery. Background Technology

[0002] With the development of industrial intelligence, rotating machinery, as the core equipment in various production scenarios, is directly related to production efficiency and safety in terms of stable operation. Full life cycle maintenance has become a key requirement for reducing operational losses.

[0003] Currently, traditional rotating machinery maintenance and management relies on experienced technicians' judgment and fixed-cycle maintenance plans, adopting a reactive approach of "don't repair unless it breaks down, and repair only when it breaks down." This approach cannot predict equipment deterioration trends and is difficult to adapt to dynamic operating conditions. As a result, maintenance work always lags behind the equipment deterioration process, and it also increases the risk of unplanned downtime and maintenance costs. Summary of the Invention

[0004] This application provides a digital twin-based full lifecycle maintenance management system for rotating machinery, which improves the current situation where traditional maintenance relies on experienced technicians' experience and fixed-cycle inspections and reactive handling, resulting in maintenance lagging behind the equipment deterioration process.

[0005] The embodiments of this application disclose the following technical solutions: This application provides a digital twin-based full lifecycle maintenance management system for rotating machinery, the system comprising: The data acquisition and virtual-real synchronization module is used to acquire the real-time operation sequence of the target rotating machinery, extract the set of operation status features, and synchronously acquire the current virtual state of the digital twin corresponding to the target rotating machinery. The health status trajectory prediction module is used to perform twin state synchronization and future multi-step evolution inference based on the set of operating status features and the current virtual state, and generate a predicted health status trajectory containing multiple future time nodes. The maintenance value assessment calculation module is used to input the predicted health status trajectory into a pre-trained maintenance value assessment model, and, in conjunction with a risk discount compensation mechanism, calculate the expected maintenance value of different maintenance actions performed at the current moment. The maintenance decision optimization generation module is used to optimize the solution based on the expected maintenance value, with the goal of maximizing the total value benefit and minimizing the total maintenance downtime, combined with the constraints of currently available maintenance resources, to generate a set of maintenance decision instructions and output them to the maintenance execution terminal.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a digital twin-based full lifecycle maintenance management system for rotating machinery. The system acquires the real-time operating sequence of the target rotating machinery through a data acquisition and virtual-real synchronization module, extracts a set of operating status features, and synchronizes the current virtual state of the digital twin. Consistency verification and adaptive parameter fine-tuning ensure accurate matching between the virtual and real systems. A health status trajectory prediction module generates a predicted health status trajectory containing multiple future time nodes based on the operating status features and the virtual synchronization state through iterative forward simulation. A maintenance value assessment calculation module inputs the predicted health status trajectory into a pre-trained maintenance value assessment model, combining the health status trajectory's discrete coefficient and vibration-temperature correlation deviation coefficient to construct a credibility score and a risk discount compensation mechanism to calculate the expected maintenance value of different maintenance actions. Finally, a maintenance decision optimization generation module aims to maximize total value gains and minimize total maintenance downtime. It uses a multi-objective optimization algorithm to solve for the Pareto optimal solution, considering maintenance resource constraints. After filtering by maintenance preference weights, prioritizing, and allocating time windows, a maintenance decision instruction set is generated and output to the maintenance execution terminal.

[0007] This application's technical solution addresses the problems of traditional rotating machinery maintenance, such as reliance on experience-based judgment, unreasonable resource allocation, and significant downtime losses, achieving intelligent and efficient maintenance management. It provides a comprehensive solution for the entire lifecycle maintenance of rotating machinery, encompassing condition awareness, trend prediction, value assessment, and decision execution, ensuring the stability and reliability of equipment operation. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating the digital twin-based full lifecycle maintenance management system for rotating machinery provided in this application embodiment; Figure 2 This is a schematic diagram of the structure for generating a predicted health status trajectory provided in an embodiment of this application.

[0010] The components represented by each number in the attached diagram are explained below: Data acquisition and virtual-real synchronization module 01, health status trajectory prediction module 02, maintenance value assessment calculation module 03, maintenance decision optimization generation module 04. Detailed Implementation

[0011] This application provides a digital twin-based full lifecycle maintenance management system for rotating machinery to solve the technical problems of traditional rotating machinery maintenance relying on experienced technicians' experience and fixed-cycle inspections and reactive processing, which cannot predict the trend of equipment deterioration, resulting in maintenance lagging behind the equipment deterioration process, low maintenance efficiency and high cost.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0014] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0015] Examples, as shown in the appendix Figure 1 As shown, this application provides a digital twin-based full lifecycle maintenance management system for rotating machinery, which includes the following steps: The data acquisition and virtual-real synchronization module 01 is used to acquire the real-time operation sequence of the target rotating machinery, extract the set of operation status features, and synchronously acquire the current virtual state of the digital twin corresponding to the target rotating machinery. In this embodiment of the application, in order to establish a real-time mapping relationship between physical devices and digital twins and provide a data foundation for subsequent health status prediction and maintenance value assessment, it is necessary to first collect the core data of device operation and extract key features, and then drive the twin to update synchronously and verify consistency, so as to ensure that the virtual state can accurately replicate the actual operation of the physical device.

[0016] Specifically, vibration and temperature sensors deployed on the rotating machinery are used to collect real-time operating data within a continuous time window. The sensor data collection must cover key operating parts of the equipment to ensure that the data comprehensively reflects the equipment's operating status. The continuous time window is set to capture the temporal changes in the operating data and avoid the influence of random data from single points.

[0017] Furthermore, based on the real-time operation sequence, time-domain statistical features are extracted to construct an operational status feature set. This process requires screening core indicators that can characterize the health status of equipment from massive operational data, extracting key information through time-domain statistical analysis, and forming a structured feature set.

[0018] Furthermore, the set of operational status features is input into the digital twin to drive the internal mechanism simulation model to update health status parameters and dynamic load response. Specifically, the mechanism simulation model needs to simulate the physical processes of equipment operation based on key indicators in the feature set, adjust corresponding parameters in real time, and generate a current virtual state that matches the operational state of the physical entity, achieving initial synchronization between virtual and real data.

[0019] Furthermore, consistency verification is performed on the indicators in the operational state feature set and the corresponding simulation outputs of the virtual state. By calculating the deviation of relevant indicators, the degree of matching between the virtual state and the physical entity is determined. If the deviation exceeds a preset threshold, the physical parameters of the mechanism simulation model are adjusted in reverse, the virtual state is regenerated, and the verification is repeated until the deviation meets the requirements, confirming that the virtual state is a valid synchronization state.

[0020] This module acquires core data reflecting the actual operating status of the equipment through data acquisition, feature extraction, virtual-real synchronization, and consistency verification. It also achieves synchronization between the digital twin and the physical equipment, providing data support for the subsequent health status trajectory prediction module to infer future degradation trends based on the synchronized status.

[0021] In the system provided in this application embodiment, the data acquisition and virtual-real synchronization module 01 includes: By deploying vibration and temperature sensors on the target rotating machinery, real-time operation sequences within a continuous time window are collected, and time-domain statistical features are extracted to construct a set of operation status features. The set of operation status features includes, but is not limited to, the effective value of vibration and the average value of bearing temperature. The set of operational status features is input into the digital twin, which drives the internal mechanism simulation model of the twin to update the corresponding health status parameters and dynamic load response, generating the current virtual state that matches the physical entity in real time. The consistency of the indicators in the running state feature set with the corresponding simulation output in the current virtual state is checked. If the deviation exceeds the preset deviation threshold, the twin model parameters are adaptively fine-tuned until the current virtual state is confirmed as a valid synchronization state.

[0022] In this embodiment of the application, in order to avoid deviations in health prediction and maintenance decisions due to incomplete collection of physical equipment operation data and inaccurate mapping of virtual and real states, it is necessary to first comprehensively collect the core data of equipment operation and extract key features, and then drive the digital twin to update the status synchronously. Consistency verification and parameter fine-tuning are used to ensure the matching of virtual and real states.

[0023] Specifically, vibration and temperature sensors deployed on the target rotating machinery are used to collect real-time operation sequences within a continuous time window, laying the foundation for extracting time-domain statistical features and constructing a set of operation status features.

[0024] The deployment of sensors must cover the key operating parts of rotating machinery, such as the areas where core components like bearings and rotors are located, to ensure that vibration and temperature changes during equipment operation can be fully captured, avoiding the omission of critical operating information due to limited data collection areas.

[0025] In addition, the setting of continuous time windows should be reasonably set in combination with the equipment operating cycle and the frequency of data changes. For example, the time window should be set at the minute or hour level to ensure that the collected real-time operating sequence can fully reflect the trend of equipment operating status changes over a period of time, rather than isolated data at a single point in time.

[0026] After acquiring the real-time operation sequence, time-domain statistical features are extracted, transforming multiple raw data sets into a structured set of operational status features. These features must characterize the equipment's operational status and health condition. Among them, the effective vibration value reflects the smoothness of equipment operation, and the average bearing temperature reflects the equipment's thermal operating status. By refining these core features, clear data input is provided for the status update of the digital twin.

[0027] Furthermore, the completed set of operational status features is input into the digital twin, driving its internal mechanism simulation model to update the corresponding health status parameters and dynamic load response, generating the current virtual state that matches the physical entity in real time.

[0028] The mechanism simulation model needs to be built based on the physical operating principles of rotating machinery, and can accurately simulate the changes in the health status and dynamic load feedback of the equipment under different operating characteristic inputs. For example, when the average bearing temperature in the set of operating state characteristics increases, the model needs to update the health status parameters characterizing the thermal degradation of the equipment accordingly, and adjust the dynamic load response to match the load changes in actual operation, so as to ensure that the generated virtual state can truly replicate the current operating condition of the physical equipment and achieve preliminary synchronization between the virtual and real states.

[0029] Furthermore, after generating the current virtual state, it is necessary to verify the consistency between the indicators in the set of operating state features and the corresponding simulation output in the virtual state in order to determine the degree of matching between the virtual and real states, correct model deviations in a timely manner, and ensure that the digital twin can truly replicate the operating state of the physical device.

[0030] In the system provided in this application embodiment, the consistency of the indicators in the running state feature set with the corresponding simulation output in the current virtual state is checked. If the deviation exceeds a preset deviation threshold, the twin model parameters are triggered for adaptive fine-tuning until the current virtual state is confirmed as a valid synchronization state, including: Calculate the first absolute deviation between the effective vibration value in the set of operating state features and the effective vibration value simulated in the current virtual state; Calculate the second absolute deviation between the average bearing temperature in the set of operating state features and the average bearing temperature simulated in the current virtual state; The first absolute deviation and the second absolute deviation are assigned preset weights and then weighted and summed to obtain the comprehensive deviation index. If the overall deviation index does not exceed the preset deviation threshold, the current virtual state will be directly marked as a valid synchronization state. If the comprehensive deviation index exceeds the preset deviation threshold, the physical parameters of the simulation model of the internal mechanism of the digital twin will be adjusted in reverse compensation based on the ratio of the first absolute deviation to the second absolute deviation. The current virtual state is regenerated using the adjusted mechanism simulation model, and iterative verification is performed until the comprehensive deviation index is lower than the preset deviation threshold. The virtual state generated in this iteration is then confirmed as a valid synchronization state.

[0031] First, calculate the first absolute deviation between the effective vibration value in the operating state feature set and the effective vibration value simulated in the current virtual state. At the same time, calculate the second absolute deviation between the average bearing temperature in the operating state feature set and the average bearing temperature simulated in the current virtual state.

[0032] Specifically, the first absolute deviation is obtained by taking the absolute value of the difference between the actual vibration effective value collected in the operating state feature set and the vibration effective value output by the simulation in the current virtual state. The calculation formula is "first absolute deviation = |actual vibration effective value - simulated vibration effective value|".

[0033] For example, if the actual collected effective vibration value is 2.3 mm / s and the virtual state simulation output vibration effective value is 2.26 mm / s, then the first absolute deviation is |2.3-2.26|=0.04 mm / s. This value directly reflects the degree of deviation between the twin and the actual state of the physical device in the mechanical vibration dimension.

[0034] In addition, the second absolute deviation is calculated using the same absolute value method. It is obtained by taking the absolute value of the difference between the actual average bearing temperature collected in the operating state feature set and the average bearing temperature output in the current virtual state. The calculation formula is "Second absolute deviation = |Actual average bearing temperature - Simulated average bearing temperature|".

[0035] For example, the actual average bearing temperature collected is 75℃, and the average bearing temperature output by the virtual state simulation is 73.5℃. Then the second absolute deviation is |75-73.5|=1.5℃. This value clearly shows the difference between the twin in the thermal operation dimension and the actual situation of the physical equipment.

[0036] Furthermore, in order to eliminate the influence of different physical dimensions and enable the deviations of vibration and temperature to be comprehensively evaluated on the same scale, the obtained first and second absolute deviations are dimensionless, and the normalized relative deviations are used as the basis for weighted fusion.

[0037] Specifically, based on the historical operating data of the target rotating machinery, the 3σ principle (that is, under the assumption of normal distribution, with the mean μ as the benchmark and three times the standard deviation 3σ as the upper limit of the fluctuation range, so that the probability of normal data falling within this range is about 99.73%) is used to determine the allowable deviation threshold of the effective value of vibration and the allowable deviation threshold of the average bearing temperature.

[0038] For example, through statistical analysis of historical data, the permissible deviation threshold for the effective value of vibration can be set to 0.05 mm / s, and the permissible deviation threshold for the average bearing temperature can be set to 2℃. These two thresholds reflect the reasonable upper limits of vibration and temperature fluctuations under normal operating conditions, respectively.

[0039] Further, the first relative deviation and the second relative deviation are calculated. Here, "first relative deviation = first absolute deviation / allowable deviation threshold of vibration effective value" and "second relative deviation = second absolute deviation / allowable deviation threshold of bearing temperature average value" are both dimensionless numbers, representing the proportion of the deviation of the simulated vibration and temperature values ​​from the actual physical values ​​relative to the allowable fluctuation range, respectively.

[0040] Furthermore, the obtained first relative deviation and second relative deviation are assigned preset weights and then summed to obtain a comprehensive deviation index. Specifically, the formula for calculating the comprehensive deviation index can be expressed as "Comprehensive Deviation Index = First Relative Deviation × Vibration Weight Coefficient + Second Relative Deviation × Temperature Weight Coefficient", where the sum of the vibration weight coefficient and the temperature weight coefficient is 1.

[0041] The preset weights are determined based on the importance of the indicators in representing the health status of the equipment. Since abnormal vibration often reflects potential faults such as wear and loosening of internal parts earlier and more directly, it is more critical for guiding maintenance decisions. Therefore, the weight coefficient of the effective vibration value (corresponding to the first relative deviation) can be 0.6, and the weight coefficient of the average bearing temperature (corresponding to the second relative deviation) can be 0.4.

[0042] By using a weighted summation method, the deviations of the two dimensions can be integrated into a dimensionless quantitative indicator that can comprehensively reflect the degree of matching between virtual and real states. This avoids the one-sidedness of evaluating deviations of a single indicator and ensures the comprehensiveness and scientific nature of the verification results.

[0043] Furthermore, the preset deviation threshold is set to 1 based on the principle of scale consistency after dimensionless processing. The rationale for this value is that both the first and second relative deviations have been normalized to their respective allowable fluctuation limits, with 1 as the critical point (less than 1 indicates the deviation is within the allowable range, and greater than or equal to 1 indicates it exceeds the limit). The comprehensive deviation index, as the weighted average of the two, is also bounded by 1; if less than 1, it indicates the overall deviation is within a controllable range, and if greater than or equal to 1, it indicates that model correction needs to be triggered.

[0044] If the comprehensive deviation index does not exceed the preset deviation threshold, it means that the difference between the current virtual state and the actual operating state of the physical device is within an acceptable range. It can accurately replicate the operating characteristics of the device and can be directly marked as an effective synchronization state, providing reliable data support for subsequent modules.

[0045] Conversely, if the overall deviation index exceeds the preset deviation threshold, it indicates that the simulation model of the twin deviates significantly from the actual physical equipment and needs to be corrected through parameter adjustment.

[0046] Specifically, based on the proportional relationship between the first absolute deviation and the second absolute deviation, the physical parameters of the simulation model of the internal mechanism of the digital twin are adjusted in reverse compensation to specifically correct simulation deviations related to vibration response or heat conduction, and quickly reduce the difference between virtual and real states.

[0047] For example, if the first absolute deviation accounts for a higher proportion of the overall deviation, it indicates that the twin's simulation of the equipment's vibration state has a greater deviation, and the physical parameters related to vibration response in the model, such as component stiffness and damping coefficient, need to be adjusted using the gradient descent method. Conversely, if the second absolute deviation accounts for a higher proportion, the focus should be on optimizing physical parameters related to heat conduction, such as thermal conductivity and heat dissipation coefficient.

[0048] Furthermore, after the physical parameters are adjusted, the current virtual state is regenerated using the adjusted mechanism simulation model, and the process returns to the initial steps of deviation calculation and comprehensive deviation index evaluation to perform consistency verification again.

[0049] The iterative verification process must continue without exceeding a preset limit on the number of iterations to avoid wasting system resources due to infinite iterations, while ensuring the efficiency and effectiveness of deviation correction. The process continues until the comprehensive deviation index falls below a preset deviation threshold, indicating that the virtual state of the twin has accurately matched the actual operating state of the physical device. At this point, the virtual state generated in this iteration is confirmed as a valid synchronization state.

[0050] Ultimately, through a complete process of deviation calculation, comprehensive evaluation, parameter adjustment, and iterative verification, dynamic calibration of the virtual state of the digital twin and the actual operating state of the physical equipment was achieved. This not only ensured that the virtual state could accurately reflect the real-time operating status of the equipment, but also provided a data foundation for the subsequent health status trajectory prediction module to extrapolate future degradation trends based on the synchronous state.

[0051] The health status trajectory prediction module 02 is used to synchronize the twin state and extrapolate the future multi-step evolution based on the set of running status features and the current virtual state, and generate a predicted health status trajectory containing multiple future time nodes. In this embodiment of the application, in the scenario where the maintenance of rotating machinery requires advance prediction of equipment degradation trends, in order to overcome the lag of traditional maintenance, it is necessary to extract health parameters from the synchronized virtual state, calculate the degradation increment in combination with operating characteristics, and iterate the simulation to generate a predicted health state trajectory that reflects future health changes.

[0052] Specifically, based on the current virtual state, initial health status parameters characterizing the overall health of the equipment are extracted from the digital twin. These initial health status parameters need to comprehensively reflect the current health baseline level of the equipment, taking into account various aspects such as mechanical wear and thermal losses, and provide a reliable starting point for subsequent predictions and extrapolations.

[0053] Furthermore, taking the effective vibration value and the average bearing temperature in the set of operating state characteristics as inputs, and combining them with the dynamic load response in the current virtual state, the preset degradation rate mapping relationship is invoked to calculate the degradation increment of health state parameters within a unit time step.

[0054] Among them, dynamic load response can reflect the current operating load intensity of the equipment. The equipment deterioration rate varies under different loads. By combining it with core operating characteristics, the calculation of deterioration increment can be more in line with the actual operating conditions of the equipment, thus improving the accuracy of incremental data.

[0055] Furthermore, starting from the initial health status parameters and using the increment of health status parameter degradation as the recursive step size, the digital twin is subjected to iterative forward simulation with discrete time steps. By gradually deducing, the natural degradation process of equipment health over time is simulated, so that the prediction process conforms to the actual degradation law of the equipment.

[0056] Furthermore, at the end of each iteration time step, the health status parameter values ​​generated by the simulation are recorded, and the iterations are executed continuously for a preset number of times. The parameter value recorded in each iteration corresponds to the health status at a future time point. Through continuous iteration, multiple future time points are linked with their corresponding health status parameter values ​​to form a complete predicted health status trajectory.

[0057] This module correlates the current state of the equipment with its future degradation trend by extracting initial health parameters, calculating degradation increments, iterating simulations, and recording parameters. The generated predicted health status trajectory can clearly show the path of health changes of the equipment over a period of time, providing predictive data support for the subsequent maintenance value assessment module to quantify the value of different maintenance actions.

[0058] As attached Figure 2 As shown, in the system provided in this application embodiment, the health status trajectory prediction module 02 includes: Based on the current virtual state, extract the initial health status parameters that characterize the overall health of the device from the digital twin; Using the effective vibration value and the average bearing temperature in the set of operating state characteristics as inputs, and combining the dynamic load response in the current virtual state, the preset degradation rate mapping relationship is invoked to calculate the degradation increment of health state parameters within a unit time step. Starting with the initial health state parameters, and using the health state parameter degradation increment as the recursive step size, the digital twin is subjected to iterative forward simulation with discrete time steps. At the end of each iteration time step, the health status parameter values ​​generated by the simulation are recorded. The iterations are executed continuously for a preset number of times to form a sequence consisting of multiple future time nodes and their corresponding health status parameter values, which serves as the predicted health status trajectory.

[0059] In this embodiment of the application, in order to avoid the problem that traditional maintenance cannot predict the future health changes of the equipment and cause maintenance decisions to lag behind the deterioration process, it is necessary to generate a trajectory that reflects the future health evolution trend by quantifying the deterioration law and iteratively simulating based on the current actual operating status and core characteristics of the equipment, so as to provide a forward-looking and data-driven basis for subsequent maintenance value assessment and decision optimization.

[0060] Specifically, based on the current virtual state, initial health status parameters representing the overall health of the device are extracted from the digital twin. The current virtual state has been synchronized with the actual operating state of the physical device through consistency verification and parameter fine-tuning. The extracted initial health status parameters are a quantitative representation of the device's current comprehensive condition, including mechanical wear, thermal losses, and performance degradation, and can objectively reflect the device's health baseline level.

[0061] For example, if the equipment currently has low mechanical wear, the bearing temperature is within the normal range, and the vibration is stable, the initial health status parameter may be 0.92 (ranging from 0 to 1, with the value closer to 1 indicating higher health); if the equipment has slight wear and the vibration is slightly higher than normal, the initial health status parameter may be 0.78, which intuitively quantifies the current overall health status of the equipment.

[0062] Furthermore, taking the effective vibration value and the average bearing temperature in the set of operating state characteristics as inputs, and combining them with the dynamic load response in the current virtual state, the preset degradation rate mapping relationship is invoked to calculate the degradation increment of health state parameters within a unit time step.

[0063] Among them, the effective value of vibration and the average bearing temperature are the core indicators characterizing the operating status of equipment. Abnormal changes in the effective value of vibration are often directly related to potential faults such as wear and loosening of internal parts of the equipment, while the average bearing temperature can reflect the operating status of the equipment's heat transfer system. Together, they constitute the key basis for judging the deterioration trend of the equipment.

[0064] In addition, dynamic load response can reflect the current operating load intensity of the equipment. Under different loads, the stress and heat generation of each component of the equipment are different, and the corresponding degradation rate will also be different. Combining it with the core operating characteristics can make the calculation of degradation increment more in line with the actual operating conditions of the equipment.

[0065] Meanwhile, the preset degradation rate mapping relationship is obtained based on a large amount of historical operating data, which fully covers the degradation patterns of equipment under different operating characteristics and load conditions. By calling this mapping relationship, abstract operating characteristics can be quantitatively associated with specific degradation increments, ensuring that the calculated degradation increments have high accuracy and rationality. For example, when the effective value of vibration increases and the dynamic load is large, the calculated degradation increment will increase accordingly to conform to the actual degradation logic of the equipment.

[0066] Furthermore, starting with the initial health status parameters and using the increment of health status parameter degradation as the recursive step size, iterative forward simulation of the digital twin is performed with discrete time steps. The setting of the discrete time step needs to comprehensively balance prediction accuracy and computational efficiency. An excessively large step size will result in overly coarse prediction results, failing to capture subtle changes in the device's health status; an excessively small step size will increase computational load and reduce system operating efficiency. Therefore, it needs to be reasonably set according to the device type, operating cycle, and characteristics of historical degradation data, for example, by setting the time step in hours or days.

[0067] Furthermore, the iterative forward simulation process simulates the changes in the health status of the equipment at each time step, and gradually extrapolates the future health evolution of the equipment. This process follows the objective laws of equipment degradation, using the degradation increment as a fixed recursive step size to ensure that the predicted trajectory can continuously and smoothly reflect the trend of the equipment's health status from the current baseline to future degradation.

[0068] Furthermore, at the end of each iteration time step, the health status parameter values ​​generated by the simulation are recorded, and the iterations are executed continuously for a preset number of times. After each iteration, the corresponding health status parameter values ​​and time points need to be recorded in a timely manner to form a one-to-one data pair. These data pairs form the basis for predicting the health status trajectory.

[0069] The preset number of times is determined based on the time span requirements of maintenance decisions. It needs to cover the key time intervals in which subsequent maintenance actions may be performed, so as to ensure that the generated predicted trajectory can provide data support for maintenance decisions at different time nodes.

[0070] By continuously executing a preset number of iterations, multiple future time points are linked with their corresponding health status parameter values ​​to form a complete sequence. This sequence is the predicted health status trajectory, which clearly presents the path of health changes of the device over a period of time, such as a gradual decrease from the current health parameter value, intuitively reflecting the deterioration trend of the device.

[0071] Ultimately, through the steps of initial health parameter extraction, degradation increment calculation, iterative forward simulation, and parameter recording, this module generates a predicted health status trajectory that comprehensively reflects the future health evolution trend of the equipment. This provides predictive data support for the subsequent maintenance value assessment module to quantify the expected value of different maintenance actions at each time point.

[0072] The maintenance value assessment calculation module 03 is used to input the predicted health status trajectory into the pre-trained maintenance value assessment model, and combined with the risk discount compensation mechanism, calculate the expected maintenance value of different maintenance actions performed at the current moment. In this embodiment of the application, in order to overcome the problem of value assessment distortion caused by simply relying on model prediction and ignoring uncertainty, it is necessary to first pre-evaluate the maintenance value through a pre-trained model, and then introduce a risk discount compensation mechanism to correct the deviation, so as to ensure that the calculated expected maintenance value can objectively reflect the actual benefits of the maintenance action.

[0073] First, the acquired predicted health status trajectory is input into a pre-trained maintenance value assessment model. The model analyzes the health evolution pattern in the trajectory and combines it with the historical maintenance value mapping relationship to output the initial predicted maintenance value of different maintenance actions, thus laying the foundation for subsequent risk compensation and value quantification.

[0074] In the system provided in this application embodiment, the pre-training steps for maintaining the value assessment model include: Collect multiple sets of sample data of the target rotating machinery during the historical maintenance cycle. Each set of sample data includes the sample health status trajectory, the actual maintenance action performed at the corresponding time, and the sample maintenance value label calculated based on the maintenance action. A maintenance value assessment model is constructed based on a neural network architecture. Using the health status trajectory of the sample as the input feature and the maintenance value label of the sample as the supervision signal, the maintenance value assessment model is trained in a supervised manner until the model loss function converges, thus completing the pre-training of the model.

[0075] Specifically, multiple sets of sample data of the target rotating machinery during historical maintenance cycles are collected first. The collection of sample data needs to cover different operating stages, different health states and different maintenance scenarios of the equipment to ensure the diversity and representativeness of the samples.

[0076] Simultaneously, each set of sample data must completely include the sample's health status trajectory, the actual maintenance actions performed at the corresponding time, and the sample maintenance value label calculated after the maintenance actions were performed. The sample health status trajectory must reflect the health evolution process of the equipment within the corresponding period, such as a complete sequence of health parameters gradually decreasing from 0.9 to 0.6.

[0077] In addition, the specific types of maintenance actions to be performed must be clearly defined, such as predefined actions like bearing lubrication, rotor balancing, and component replacement. The sample maintenance value label must be obtained through quantitative calculation, taking into account the benefits brought by the maintenance action, such as extended equipment life, reduced failure risk, and improved operating efficiency, as well as the various costs incurred during the execution process, to form a quantitative indicator that can objectively reflect the actual value of the maintenance action. For example, if the net benefit after performing a component replacement action is 50,000 yuan, then the corresponding sample maintenance value label is 50,000.

[0078] Furthermore, a value assessment model is constructed and maintained based on a neural network architecture. The model architecture design needs to be adapted to the task requirements to ensure that it can effectively process sequential inputs and output multi-dimensional value prediction results. The model input layer needs to be adapted to the data structure of the sample health status trajectory and adopt network components related to time series data processing, such as the input unit of the Long Short-Term Memory (LSTM) network, which can efficiently capture the temporal features and trend changes in the health trajectory.

[0079] Furthermore, the number of hidden layers and neurons needs to be appropriately set. Through multi-layer nonlinear transformations, the complex mapping relationship between health status trajectory and maintenance value can be deeply explored. For example, 3-5 hidden layers can be set, with the number of neurons in each layer set to 64-256 depending on the dimensionality and complexity of the sample data. The output layer corresponds to the initial predicted maintenance value of multiple predefined maintenance actions, adopting a fully connected layer structure, with the output dimension consistent with the number of predefined maintenance actions. For example, when three types of actions are predefined—routine inspection, component replacement, and major overhaul—the output layer dimension is set to 3 to ensure that the initial predicted value of each type of maintenance action can be output simultaneously.

[0080] During the model architecture construction process, targeted adjustments are also needed based on the characteristics of the sample data and the complexity of the task. For example, if there are many time nodes in the health status trajectory of the samples and the temporal dependencies are complex, the number of LSTM layers or neurons can be increased; if there are many predefined maintenance action types, the number of neurons in the output layer needs to be increased accordingly to ensure that the model can distinguish the value differences of different actions.

[0081] At the same time, it is necessary to select appropriate activation functions and optimizers. The ReLU activation function can be used in the hidden layer to enhance the nonlinear expressive ability of the model. The linear activation function is used in the output layer according to the continuity requirements of value prediction. The Adam optimizer is selected to improve the convergence speed and stability of model training.

[0082] Furthermore, the health status trajectory of the samples is used as the input feature, and the sample maintenance value label is used as the supervision signal to conduct supervised training of the maintenance value assessment model. Before training, the sample data needs to be preprocessed, including data normalization, missing value imputation, and outlier removal, to ensure the quality of the data input to the model.

[0083] Among them, data normalization maps the parameter values ​​of the health status trajectory of the sample to the [0,1] interval, so as to avoid the model training effect being affected by the difference in data volume; missing value filling uses linear interpolation to supplement the missing health parameter values ​​in the trajectory; outlier removal uses the 3σ principle to screen and remove sample data that deviates significantly from the normal range.

[0084] Furthermore, the preprocessed sample data is divided into training set, validation set and test set in a ratio of 7:2:1. The training set is used for iterative updates of model parameters, the validation set is used to monitor model performance during training, and the test set is used to finally evaluate the model's generalization ability.

[0085] During training, the health status trajectories of the training set samples are input into the model, and the model outputs the initial predicted maintenance value. The mean squared error (MSE) is calculated as the loss function to quantify the difference between the predicted value and the sample maintenance value label, i.e., "loss function value = Σ(predicted maintenance value - sample maintenance value label)". 2 " / sample size". Based on the loss function value, the parameters of each layer of the model are continuously adjusted through the backpropagation algorithm to gradually reduce the prediction error.

[0086] During training, the changes in the loss function value of the validation set need to be monitored in real time. If the loss function value of the validation set no longer decreases or shows an upward trend for several consecutive rounds, it indicates that the model has reached convergence or there is a risk of overfitting, and training should be stopped in time. For example, when the fluctuation range of the loss function value of the validation set is less than 0.001 for 10 consecutive rounds, the model loss function is considered to have converged, and training should be stopped.

[0087] Meanwhile, if the validation set loss function value increases during training, measures such as regularization and dropout should be taken to suppress overfitting. For example, a dropout layer can be added to the hidden layer to randomly deactivate some neurons and reduce the model's excessive dependence on the training data.

[0088] In addition, the trained model needs to be evaluated for performance. This involves calculating metrics such as prediction accuracy and mean absolute error using test set data to ensure the model has good generalization ability. For example, if the model's prediction error for the maintenance value of more than 85% of the samples in the test set is controlled within 10%, then the model's performance is considered satisfactory. If the prediction error for a certain type of maintenance action is large, additional sample data for that type of action needs to be added, and the model needs to be retrained and optimized until the model's prediction accuracy for the value of all types of maintenance actions reaches the preset accuracy requirements.

[0089] Ultimately, the pre-trained maintenance value assessment model, completed through sample collection, architecture construction, and supervised training, can output the initial predicted maintenance value for different maintenance actions based on the input health status trajectory, providing a reliable foundation for subsequent calculation of expected maintenance value in conjunction with a risk discount compensation mechanism.

[0090] Furthermore, after the maintenance value assessment model has been pre-trained, the expected maintenance value of different maintenance actions at the current moment is calculated by combining the risk discount compensation mechanism, so as to quantify the actual benefits and potential risks of different maintenance actions.

[0091] In the system provided in this application embodiment, a risk discount compensation mechanism is used to calculate the expected maintenance value of different maintenance actions performed at the current moment, including: Based on the vibration growth rate of the effective vibration value in the set of operating status characteristics compared with the historical benchmark, the number of prediction time nodes is dynamically determined. The larger the vibration growth rate, the fewer prediction time nodes are determined. Based on the number of predicted time points, the twin state is synchronized and multi-step evolution is extrapolated to generate a predicted health state trajectory. The dispersion coefficient of the health status trajectory is obtained by the ratio of the standard deviation of the health status parameter values ​​at multiple future time points to the average value of the health status parameters in the predicted health status trajectory. Identify the equipment's operating conditions based on the current operating status feature set, select the corresponding time period from historical data under the same operating conditions, and calculate the vibration-temperature correlation coefficient between the effective vibration value and the average bearing temperature. The vibration temperature correlation coefficient is compared with the historical benchmark vibration temperature correlation coefficient under the same working conditions, and the vibration temperature correlation deviation coefficient is calculated based on the degree of deviation. The dispersion coefficient of the health status trajectory and the vibration-temperature related deviation coefficient are weighted and fused to generate a preliminary credibility score; The initial credibility score is attenuated and corrected based on the number of predicted time points to generate the final comprehensive credibility score. The fewer the number of predicted time points, the greater the degree of attenuation correction. The overall credibility score is converted into a risk discount factor through a preset nonlinear mapping function, wherein the value range of the risk discount factor is [0, 1]. The predicted health status trajectory is input into the pre-trained maintenance value assessment model, and the output is the initial predicted maintenance value for different maintenance actions. The initial predicted maintenance value is compensated by applying a risk discount factor to generate the compensated expected maintenance value.

[0092] Specifically, the number of prediction time points is dynamically determined first based on the increase in vibration RMS values ​​in the operational status feature set compared to historical benchmarks. The RMS vibration value is an indicator reflecting potential faults such as mechanical wear and loosening of components. Its increase compared to historical benchmarks directly reflects the rate of equipment deterioration. A larger increase indicates a more rapid deterioration of the equipment's condition and a higher uncertainty in its future health trajectory. Excessively extending the prediction time can lead to significant deviations between the prediction results and actual conditions. Therefore, it is necessary to reduce the number of prediction time points and focus on near-term critical maintenance windows.

[0093] Conversely, if the vibration increase is relatively small, the equipment degradation is relatively gradual, and the future state is more predictable. Therefore, the number of prediction time points can be appropriately increased to achieve a longer-term health trajectory projection, ensuring that the prediction range matches the actual degradation trend of the equipment. For example, if the effective vibration value increases by 30% compared to the historical baseline, the number of prediction time points can be set to 5; if the increase is only 5%, it can be set to 10, dynamically adjusting to balance prediction accuracy and decision-making needs.

[0094] Furthermore, based on the determined number of predicted time points, the twin state synchronization and future multi-step evolution projection are performed to generate a predicted health state trajectory. Similarly, this process adopts the same approach as the health state trajectory prediction module 02, using the current virtual state after consistency verification as a basis, combined with operational state characteristics and dynamic load response, and through iterative forward simulation, a sequence containing multiple future time points and corresponding health parameter values ​​is obtained, providing basic data for subsequent risk quantification and value assessment.

[0095] Furthermore, the dispersion coefficient of the health status trajectory is obtained based on the ratio of the standard deviation of the health status parameter values ​​at multiple future time points in the predicted health status trajectory to the average value of the health status parameters.

[0096] The coefficient of variation (COP) of the health status trajectory is an indicator that quantifies the degree of fluctuation in future health status. The standard deviation reflects the dispersion of parameter values, while the mean reflects the overall health level. The ratio of the two directly reflects the stability of the trajectory. A larger COP indicates more drastic fluctuations in future health parameters and lower reliability of the prediction results; a smaller COP indicates a more stable health trajectory and more reliable prediction results.

[0097] For example, if a device predicts health parameters of 0.8, 0.6, 0.7, 0.5, and 0.6 at future nodes, with a standard deviation of approximately 0.11 and an average of 0.64, the coefficient of variation of the health status trajectory is approximately 0.17, indicating that the trajectory fluctuation is small. However, if the parameter values ​​are 0.9, 0.4, 0.8, 0.3, and 0.7, with a standard deviation of approximately 0.25 and an average of 0.62, the coefficient of variation of the health status trajectory is approximately 0.40, indicating significant trajectory fluctuation and a markedly lower reliability.

[0098] Simultaneously, based on the current operating status feature set, the operating conditions of the equipment are identified. From historical data under the same operating conditions, corresponding time periods are selected to calculate the vibration-temperature correlation coefficient between the effective vibration value and the average bearing temperature. The vibration-temperature correlation coefficient is calculated by dividing the covariance of the vibration sequence and the temperature sequence by the product of the standard deviations of the vibration sequence and the temperature sequence, with a value range of [-1, 1] to reflect the synergy between vibration and temperature changes.

[0099] When the vibration-temperature correlation coefficient is close to 1, it indicates that the two are positively correlated, and the equipment operation conforms to the conventional physical logic of increased vibration and temperature rise. When the vibration-temperature correlation coefficient is close to -1, it is negatively correlated, which corresponds to the normal pattern under specific working conditions. When the vibration-temperature correlation coefficient is close to 0, it indicates that the two changes are not significantly related, and there may be problems such as abnormal wear and heat dissipation failure. At this time, the reliability of the prediction results will be affected.

[0100] For example, under normal load conditions, the historical vibration-temperature correlation coefficient is usually around 0.7. If the current calculation result is 0.2, it indicates that the vibration and temperature changes are not synchronized, and there is an abnormality in the operating condition.

[0101] Furthermore, the calculated vibration-temperature correlation coefficient is compared with the historical baseline vibration-temperature correlation coefficient under the same operating conditions, and the vibration-temperature correlation deviation coefficient is calculated based on the degree of deviation. Among them, the historical baseline vibration-temperature correlation coefficient is a standard value obtained from long-term operating data under the same conditions, which can reflect the vibration-temperature coordination law of the equipment under normal conditions.

[0102] Specifically, the formula for calculating the vibration temperature correlation deviation coefficient can be expressed as "Vibration temperature correlation deviation coefficient = |current vibration temperature correlation coefficient - historical benchmark vibration temperature correlation coefficient under the same working condition| / (upper limit of the reasonable range of historical benchmark vibration temperature correlation coefficient - lower limit of the reasonable range of historical benchmark vibration temperature correlation coefficient)".

[0103] The denominator represents the reasonable range of historical baseline vibration-temperature correlation coefficients under the same operating conditions. This range is used to normalize the deviation to the [0, 1] interval to ensure consistency with the quantification dimension of the health state trajectory dispersion coefficient. Since the overall range of the vibration-temperature correlation coefficient is [-1, 1], and considering the vibration-temperature coordination law during normal equipment operation, the reasonable range of historical baseline vibration-temperature correlation coefficients is usually limited to [-0.8, 0.8]. Therefore, the denominator is fixed at 0.8 - (-0.8) = 1.6.

[0104] For example, if the historical baseline vibration-temperature correlation coefficient under the same operating condition is 0.6 (within a reasonable range), and the currently calculated vibration-temperature correlation coefficient is 0.4, the absolute value of the difference between the two is 0.2. Substituting this into the formula, we can obtain the vibration-temperature correlation deviation coefficient = 0.2 / 1.6 = 0.125, indicating that the deviation between the current operating condition and the historical normal operating condition is small, and the prediction model has strong adaptability. If the currently calculated vibration-temperature correlation coefficient is -0.4, and the absolute value of the difference between this and the historical baseline value of 0.6 is 1.0, we can obtain the vibration-temperature correlation deviation coefficient = 1.0 / 1.6 = 0.625, indicating that the vibration-temperature coordination pattern of the current equipment operation is significantly different from the historical normal, and the reliability of the prediction results needs to be reduced accordingly.

[0105] Furthermore, the dispersion coefficient of the health state trajectory and the vibration-temperature correlation deviation coefficient are weighted and fused to generate a preliminary reliability score. The weight allocation needs to be reasonably set based on the degree of influence of both on the prediction reliability. For example, assigning a weight of 0.6 to the dispersion coefficient of the health state trajectory and a weight of 0.4 to the vibration-temperature correlation deviation coefficient highlights the influence of trajectory stability on reliability while taking into account the adaptability to the working conditions.

[0106] Specifically, the weighted fusion formula can be expressed as "Preliminary confidence score = 1 - (health status trajectory dispersion coefficient × weight 1 + vibration-temperature correlation deviation coefficient × weight 2)", with the score ranging from [0, 1]. The closer the score is to 1, the higher the prediction confidence. For example, if the health status trajectory dispersion coefficient is 0.17 and the weight is 0.6, and the vibration-temperature correlation deviation coefficient is 0.625 and the weight is 0.4, then the preliminary confidence score = 1 - (0.17 × 0.6 + 0.625 × 0.4) = 0.648, indicating a medium level of confidence.

[0107] Furthermore, the initial credibility score is attenuated and corrected based on the number of predicted time points to generate the final comprehensive credibility score. Specifically, the correction formula is "final comprehensive credibility score = initial credibility score × (currently dynamically determined number of predicted time points / preset maximum number of predicted time points)".

[0108] The maximum number of preset prediction time nodes is a fixed value set based on equipment maintenance cycles and decision-making time span requirements, such as 10. The fewer the prediction time nodes, the shorter the prediction period, the higher the uncertainty of the future state, and the greater the degree of attenuation correction, thereby quantifying the risk discount of short-term predictions.

[0109] Conversely, the closer the number of predicted time points is to the maximum value, the smaller the correction degree, and the closer the reliability is to the initial score. For example, if the current number of predicted time points is 5, the preset maximum number is 10, and the initial reliability score is 0.818, then the overall reliability score = 0.818 × (5 / 10) = 0.409; if the current number of predicted time points is 8, then the overall reliability score = 0.818 × (8 / 10) = 0.654.

[0110] Furthermore, the overall credibility score is converted into a risk discount factor through a pre-defined nonlinear mapping function to quantify the decision-making risk brought about by the uncertainty of prediction.

[0111] In the system provided in this application embodiment, the comprehensive credibility score is converted into a risk discount factor through a preset nonlinear mapping function, including: Determine the relationship between the overall credibility score and the preset high credibility threshold and preset low credibility threshold; If the overall credibility score is greater than or equal to the preset high credibility threshold, the risk discount factor will be set to a value of 1. If the overall credibility score is less than or equal to the preset low credibility threshold, the risk discount factor will be set to 0. If the overall credibility score is greater than the preset low credibility threshold and less than the preset high credibility threshold, then a risk discount factor between 0 and 1 is calculated based on the overall credibility score using a non-linear decay formula. The lower the overall credibility score, the smaller the calculated risk discount factor value.

[0112] Specifically, the definition and values ​​of the preset thresholds are first clarified. The preset high confidence threshold is 0.8 points, the low confidence threshold is 0.3 points, and the comprehensive confidence score ranges from [0, 1]. The threshold division is based on statistical analysis of a large amount of historical operation and maintenance data. A high confidence threshold of 0.8 corresponds to a match between the predicted health status trajectory and the actual evolution of the equipment exceeding 80%, indicating high reliability of the prediction results. A low confidence threshold of 0.3 corresponds to a match below 30%, indicating extremely low reference value of the prediction results. By using dual thresholds to divide confidence into high, medium, and low ranges, clear boundaries are provided for the differentiated setting of risk discount factors.

[0113] Furthermore, a risk discount factor is set for the high confidence interval. If the overall confidence score is greater than or equal to 0.8, it indicates that the predicted health status trajectory is based on stable operating conditions and a regular deterioration trend, the uncertainty of the prediction result is extremely small, and the expected benefits of maintenance actions are highly guaranteed. Therefore, the risk discount factor is set to 1.

[0114] For example, the overall credibility score of a certain rotating machine is 0.85, which is in the high credibility range. The risk discount factor is directly set to 1, which means that the initial predicted maintenance value does not need to be discounted by risk and can directly reflect the actual expected benefits of the maintenance action.

[0115] Furthermore, for the low confidence interval, if the overall confidence score is less than or equal to 0.3, it indicates that the prediction results are affected by multiple factors such as abnormal operating conditions and trajectory fluctuations, and the reliability is extremely poor. The initial predicted maintenance value can no longer be used as an effective reference for maintenance decisions. In this case, the risk discount factor is set to 0.

[0116] For example, the overall credibility score is 0.28, which is in the low credibility range. The risk discount factor is set to 0, and subsequent value compensation will focus on the cost loss of maintenance actions.

[0117] Furthermore, if the overall credibility score falls within the moderate credibility range of 0.3 to 0.8, the risk discount factor is calculated using a non-linear decay formula, which can be expressed as "risk discount factor = e (-λ×((0.8-综合可信度评分) / (0.8-0.3))) "" Here, λ is a preset attenuation coefficient greater than 0, used to control the attenuation rate of the factor. It is usually set to a value between 1 and 3 according to the risk tolerance of the equipment type and maintenance scenario. For example, for large critical rotating machinery, which has high maintenance costs and high risk sensitivity, λ can be set to 2.5 to achieve a steeper attenuation.

[0118] Furthermore, the natural exponential function ensures that the risk discount factor always falls between 0 and 1, conforming to the value logic of risk quantification. The core advantage of this formula lies in achieving gradient quantification of risk: the closer the overall credibility score is to a high threshold, the closer the factor is to 1, and the lower the risk; the closer it is to a low threshold, the closer the factor is to 0, and the higher the risk. Moreover, its non-linear characteristics can accurately capture the differences in risk levels within the medium credibility range, avoiding the ambiguity in risk assessment caused by linear mapping.

[0119] For example, if λ is 2, the overall credibility score is 0.7 (close to the high threshold). Substituting this into the formula, we can obtain the risk discount factor = e (-2×((0.8-0.7) / 0.5)) =e (-0.4) A score of approximately 0.670 indicates low risk and strong reliability of the initial prediction. If the overall credibility score is 0.5 (mid-range), then the risk discount factor = e (-2×((0.8-0.5) / 0.5)) =e (-1.2)≈0.301, the risk is significantly increased; if the overall credibility score is 0.35 (close to the low threshold), then the risk discount factor = e (-2×((0.8-0.35) / 0.5)) =e (-1.8) The value is approximately 0.165, indicating a relatively high level of risk.

[0120] In practical applications, the value of the attenuation coefficient λ needs to be verified and optimized multiple times. For example, by comparing the actual benefits and expected returns of maintenance decisions under different λ values, λ is adjusted to the optimal value to ensure that the risk discount factor can truly reflect the decision-making risks brought about by prediction uncertainty. At the same time, the threshold and attenuation coefficient need to be updated regularly based on new historical maintenance data and prediction accuracy feedback to ensure that the nonlinear mapping function always adapts to changes in equipment operating status and maintains the accuracy and timeliness of risk assessment.

[0121] Furthermore, after quantifying the risk discount factor, the predicted health status trajectory is input into the pre-trained maintenance value assessment model, outputting the initial predicted maintenance value for different maintenance actions. Simultaneously, the risk discount factor is applied to compensate for the initial predicted maintenance value, generating the compensated expected maintenance value, thereby making the maintenance plan more aligned with the equipment health evolution trend.

[0122] In the system provided in this application embodiment, a risk discount factor is applied to compensate for the initial predicted maintenance value, generating a compensated expected maintenance value, including: If the overall credibility score is greater than or equal to the preset high credibility threshold, then the compensated expected maintenance value is equal to the initial predicted maintenance value. If the overall credibility score is less than or equal to the preset low credibility threshold, the compensated expected maintenance value is equal to the negative maintenance cost. If the overall credibility score is greater than the preset low credibility threshold and less than the preset high credibility threshold, then the compensated expected maintenance value is equal to the risk discount factor multiplied by the expected return, and then minus the maintenance cost. Maintenance operation cost is the total cost incurred in performing the current maintenance operation, including labor costs, spare parts and material costs, and opportunity costs of downtime caused during maintenance. Expected benefits are derived from the initial projected maintenance value.

[0123] First, the compensation rule is based on the high, medium and low ranges of the comprehensive credibility score, combined with the quantitative relationship between the risk discount factor and cost and benefit, to correct the initial prediction maintenance value.

[0124] Specifically, for the high confidence interval where the overall confidence score is greater than or equal to 0.8, the risk discount factor is 1, indicating that the predicted health status trajectory is highly consistent with the actual situation, the initial predicted maintenance value has high reference value, and the expected benefits of maintenance actions can be stably realized. Therefore, the compensated expected maintenance value is directly equal to the initial predicted maintenance value.

[0125] For example, the initial predicted maintenance value of a shutdown overhaul of a rotating machine is 100,000 yuan, with a comprehensive confidence score of 0.85, which is in the high confidence range. The expected maintenance value after compensation is 100,000 yuan, which fully preserves the value information predicted by the model and provides a clear benefit reference for decision-making.

[0126] Furthermore, if the overall credibility score is less than or equal to 0.3, it is in the low credibility range, and the risk discount factor is 0. This means that the prediction results are greatly affected by factors such as abnormal working conditions and trajectory fluctuations. The initial predicted maintenance value can no longer reflect the actual situation. Maintenance actions not only fail to achieve the expected benefits, but may also generate additional cost losses. Therefore, the compensated expected maintenance value is set as the negative maintenance action cost.

[0127] For example, the total maintenance cost of the component replacement action is 80,000 yuan, with a comprehensive credibility score of 0.25, which is in the low credibility range. The expected maintenance value after compensation is -80,000 yuan, which directly reflects the potential loss risk of this maintenance action and reminds the decision-makers to choose carefully.

[0128] Furthermore, when the overall credibility score is in the moderate credibility range of 0.3 to 0.8, the compensation rule needs to balance benefits and risks. Therefore, the calculation formula "risk discount factor × expected benefit - maintenance operation cost" is adopted. Among them, the expected benefit is derived from the initial predicted maintenance value, that is, "expected benefit = initial predicted maintenance value - maintenance operation cost", which reflects the net benefit of the maintenance operation under ideal conditions.

[0129] Meanwhile, the risk discount factor quantifies the revenue loss caused by forecast uncertainty. Multiplying the two yields the actual expected return after considering risk, which is then subtracted from the maintenance cost to obtain the expected maintenance value within that range. The calculation of maintenance costs must cover three core dimensions: labor, spare parts, and downtime, to ensure the comprehensiveness of cost quantification.

[0130] Specifically, labor cost = standard operating hours × comprehensive labor rate. Standard operating hours are determined based on the complexity and process requirements of the maintenance actions, while the comprehensive labor rate is calculated by combining factors such as labor costs and management expenses. For example, the standard operating hours for routine lubrication are 2 hours, the comprehensive labor rate is 150 yuan / hour, and the labor cost is 300 yuan.

[0131] In addition, the cost of spare parts materials = Σ(consumption of the i-th spare part × net unit price of the i-th spare part). It is necessary to accurately calculate the quantity and unit price of various spare parts required for maintenance actions. For example, regular lubrication requires consuming 1 barrel of special lubricating oil with a unit price of 200 yuan, and the cost of spare parts materials is 200 yuan.

[0132] Furthermore, the opportunity cost of downtime = planned downtime × downtime loss per unit time. Planned downtime is related to standard operating hours, while downtime loss per unit time is calculated based on factors such as equipment operating revenue and capacity loss. For example, the planned downtime for routine lubrication is 2 hours, the downtime loss per unit time is 1000 yuan, and the opportunity cost of downtime is 2000 yuan. Therefore, the total maintenance cost for this routine lubrication operation = 300 + 200 + 2000 = 2500 yuan.

[0133] For example, if the initial predicted maintenance value of routine lubrication is 5,000 yuan, the overall credibility score is 0.6 (medium range), and the risk discount factor is 0.449, then the expected benefit = 5,000 - 2,500 = 2,500 yuan. The compensated expected maintenance value = 0.449 × 2,500 - 2,500 ≈ 1,122.5 - 2,500 = -1,377.5 yuan. This indicates that under the current risk level, the actual benefit of this maintenance action cannot cover the cost, and it is not recommended to prioritize its execution.

[0134] Conversely, if the overall credibility score is 0.7 and the risk discount factor is 0.670, the expected maintenance value is 0.670 × 2500 - 2500 = 1675 - 2500 = -825 yuan. Although it is still negative, the risk of loss has been significantly reduced. If the overall credibility score is 0.75 and the risk discount factor is 0.779, the expected maintenance value is 0.779 × 2500 - 2500 = 1947.5 - 2500 = -552.5 yuan, which is closer to the break-even point.

[0135] During the cost accounting process, all parameters must be obtained from the enterprise's asset management platform to ensure the authenticity and authority of the accounting results. Simultaneously, the weighting of cost components needs to be dynamically adjusted based on differences in equipment type and maintenance scenarios. For example, for large, critical equipment, downtime opportunity costs account for a higher proportion and require focused accounting; for smaller equipment, spare parts material costs and labor costs may be more critical.

[0136] Ultimately, by using differentiated compensation rules based on confidence intervals and combining comprehensive cost and benefit quantification, the generated expected maintenance value not only makes full use of the model's predictive capabilities and effectively avoids decision-making risks caused by prediction uncertainty and cost losses, but also objectively reflects the actual benefits of different maintenance actions at the current moment.

[0137] The maintenance decision optimization generation module 04 is used to optimize the solution based on the expected maintenance value, with the goal of maximizing the total value benefit and minimizing the total maintenance downtime, combined with the constraints of currently available maintenance resources, to generate a set of maintenance decision instructions and output them to the maintenance execution terminal.

[0138] In this embodiment of the application, in order to ensure that the maintenance plan is both value-maximizing and practically executable, it is necessary to generate a scientific and reasonable set of maintenance decision instructions by constructing a multi-objective optimization function, clarifying resource constraints, and calling optimization algorithms to solve the problem, so as to achieve the optimal allocation of maintenance resources and the comprehensive improvement of equipment operation and maintenance efficiency.

[0139] Specifically, a multi-objective optimization function is first constructed with the core objective of maximizing total value revenue and minimizing total maintenance downtime. Total value revenue integrates the expected maintenance value after compensation for all optional maintenance actions, and total maintenance downtime summarizes the equipment downtime generated during the execution of all optional maintenance actions. The function construction clarifies the guidance for maintenance decisions.

[0140] Furthermore, establish maintenance resource constraints, covering key dimensions such as the total available maintenance personnel hours, the inventory quantity of each type of spare parts, and the planned time window for different maintenance actions, to define the boundaries of the feasibility of maintenance plans and ensure that decisions do not exceed the actual resource supply capacity.

[0141] Furthermore, a multi-objective optimization algorithm is invoked, guided by the constructed multi-objective optimization function, to solve the problem under the set maintenance resource constraints, and to select a set of Pareto optimal maintenance action execution schemes. These schemes achieve an optimal balance between total value gain and total downtime, and there is no situation where any one scheme is better than the other schemes in all objectives.

[0142] Simultaneously, based on preset maintenance preference weights, the weighted comprehensive fitness of the total value benefit and total downtime of each solution in the Pareto optimal solution set is calculated. The optimal solution is selected as the final execution solution based on the comprehensive fitness, ensuring that the decision result aligns with actual maintenance needs. Furthermore, based on the expected maintenance value of each maintenance action in the final execution solution, execution priorities are generated in descending order to clarify the execution sequence of maintenance actions.

[0143] Finally, specific execution time windows are assigned to maintenance actions with execution priorities. The maintenance action content, execution time windows, and priority information are integrated to form a complete set of maintenance decision instructions, which are then output to the maintenance execution terminal to provide clear and operable guidance for actual maintenance work.

[0144] In the system provided in this application embodiment, the maintenance decision optimization generation module 04 includes: Construct a multi-objective optimization function with the core objective of maximizing total value revenue and minimizing total maintenance downtime. Here, total value revenue is the sum of the expected maintenance value after compensation for all optional maintenance actions, and total maintenance downtime is the sum of the downtime caused by the execution of all optional maintenance actions. Establish maintenance resource constraints, which include at least the total available maintenance personnel man-hours constraints, the inventory quantity constraints of each category of spare parts, and the planned time window constraints for different maintenance actions. A multi-objective optimization algorithm is invoked, guided by the multi-objective optimization function, to solve the problem under the constraint of maintenance resource constraints, thereby obtaining a set of Pareto optimal maintenance action execution schemes. Based on preset operation and maintenance preference weights, the weighted comprehensive fitness of the total value benefit and total downtime of each scheme in the Pareto optimal maintenance action execution scheme set is calculated, and the scheme with the highest comprehensive fitness is selected as the final execution scheme. Based on the expected maintenance value of the maintenance actions in the final execution plan, the execution priorities are generated by sorting them in descending order; Assign specific execution time windows to maintenance actions with execution priorities, and generate a set of maintenance decision instructions that includes the maintenance action content, execution time window, and priority.

[0145] Specifically, a multi-objective optimization function is first constructed with the core objective of maximizing total value revenue and minimizing total maintenance downtime. Total value revenue integrates the compensated expected maintenance value of all optional maintenance actions, reflecting the revenue orientation of maintenance decisions. Total maintenance downtime summarizes the equipment downtime generated during the execution of all optional maintenance actions, reflecting the impact of maintenance on production continuity.

[0146] By constructing a multi-objective optimization function, the two core objectives are quantified into computable mathematical indicators, providing a clear direction for subsequent optimization solutions. This ensures that decisions pursue both maximizing benefits and minimizing downtime losses, achieving synergistic optimization of the dual objectives.

[0147] Furthermore, maintenance resource constraints are established to define feasible boundaries for maintenance decisions. These constraints include at least the total available maintenance personnel man-hours, the inventory quantity of each type of spare parts, and the planned time window constraints for different maintenance actions.

[0148] Specifically, the total available maintenance man-hours constraint is determined based on the current number of on-duty maintenance personnel and their working hours, ensuring sufficient manpower to support the execution of maintenance actions. Furthermore, the inventory quantity constraints for each category of spare parts are set based on real-time inventory data from the enterprise's asset management platform to prevent maintenance actions from being unable to proceed due to spare parts shortages. In addition, the planned time window constraints for different maintenance actions are determined in conjunction with production scheduling to ensure that maintenance work does not conflict with core production tasks.

[0149] Furthermore, a multi-objective optimization algorithm is invoked to solve the problem under the guidance of the multi-objective optimization function and the constraints of maintenance resources, thereby obtaining a set of Pareto-optimal maintenance action execution schemes to find the optimal balance between maximizing revenue and minimizing downtime losses.

[0150] The system provided in this application embodiment invokes a multi-objective optimization algorithm, guided by a multi-objective optimization function, to solve the problem under the constraint of maintenance resource constraints, thereby obtaining a Pareto-optimal set of maintenance action execution schemes, including: Multiple maintenance action execution schemes are randomly generated as the initial population individuals, and each scheme explicitly contains a set of maintenance actions to be executed; For each maintenance action execution plan in the population, the objective function values ​​of total value revenue and total maintenance downtime are calculated to form a multi-objective fitness vector. The feasibility of the plan is verified by maintenance resource constraints, and invalid plans that do not meet the constraints are eliminated. Based on multi-objective fitness vectors, non-dominated sorting and crowding calculation are performed on individuals in the initial population. Selection, crossover and mutation genetic operations are performed by combining the sorting results and crowding values ​​to retain superior individuals in the population and generate a new generation of population. Repeat the steps of calculating the multi-objective fitness vector, verifying feasibility, and performing genetic operations until the preset maximum number of iterations is reached; From the final generation population, all schemes that satisfy the maintenance resource constraints and do not dominate each other are selected to form the Pareto optimal maintenance action execution scheme set.

[0151] Specifically, multiple maintenance action execution plans are first randomly generated as the initial population. Each plan explicitly includes a set of maintenance actions to be executed, and the combination must cover maintenance actions of different types and priorities to ensure the diversity of the initial population.

[0152] For example, some solutions may include a combination of component replacement and routine lubrication, while others may include a combination of downtime overhaul and rotor balancing, thus covering more decision possibilities through a variety of initial solutions.

[0153] Furthermore, for each maintenance action execution plan in the population, the objective function values ​​of total value revenue and total maintenance downtime are calculated to form a multi-objective fitness vector.

[0154] The total value calculation requires summarizing the expected maintenance value after compensation for all optional maintenance actions in the plan, while the total maintenance downtime is the sum of equipment downtime caused by the execution of all maintenance actions. Together, these two constitute the core indicators reflecting the quality of the plan. Simultaneously, the feasibility of the plan is verified by maintenance resource constraints. Each plan is checked to ensure it meets the limitations of available maintenance personnel hours, spare parts inventory quantities, and planned time windows. Invalid plans that exceed resource supply or do not meet time requirements are eliminated, ensuring that the remaining plans are all feasible for actual execution.

[0155] Furthermore, based on the multi-objective fitness vector, non-dominated sorting and crowding calculation are performed on individuals in the initial population. Specifically, the non-dominated sorting needs to be hierarchically divided according to the merits of the objective function values ​​of the schemes. The optimal solution that is not dominated by any other scheme is listed as the first level, and then the non-dominated solutions in the new round of the remaining schemes are listed as the second level, and so on, until all schemes are stratified. The relative advantages of the schemes are clarified through hierarchical division.

[0156] For example, suppose there are three plans, A, B, and C, within a certain level. Plan A has a total value of 80,000 yuan and a total maintenance downtime of 3 hours; Plan B has a total value of 70,000 yuan and a total maintenance downtime of 2 hours; and Plan C has a total value of 90,000 yuan and a total maintenance downtime of 4 hours. Plan A has a higher value and longer downtime than Plan B, but a lower value and shorter downtime than Plan C. It is not completely dominated by any plan and is classified as the first level. If Plan D has a value of 60,000 yuan and a downtime of 1.5 hours, and is not dominated by any other remaining plan, it is classified as the second level. This process is repeated to complete the hierarchical classification of all plans.

[0157] In addition, the Euclidean distance between each scheme and its neighboring schemes in the multi-objective fitness vector space within the same non-dominated level is calculated, and the crowding value of the scheme is obtained by summing them. The distribution sparsity of the schemes within the same non-dominated level is analyzed, and the differences between the schemes are quantified to avoid the selected schemes being too concentrated.

[0158] For example, a first level includes three schemes: E, F, and G, with multi-objective fitness vectors of [70,000 yuan, 2.5 hours], [80,000 yuan, 3 hours], and [90,000 yuan, 3.5 hours], respectively. First, the objective function values ​​are normalized, assuming the normalized profit is [0.3, 0.6, 0.9] and the normalized downtime is [0.2, 0.5, 0.8]. The Euclidean distance between E and F is calculated as follows: The Euclidean distance between F and G is The crowding value of F is 0.424 + 0.424 = 0.848. E is only adjacent to F, with a crowding value of 0.424. G is only adjacent to F, with a crowding value of 0.424. These values ​​clearly show that the area where F is located is more densely distributed, while the areas where E and G are located are more sparsely distributed.

[0159] Furthermore, by combining the non-dominated sorting results with the crowding value, genetic operations such as selection, crossover, and mutation are performed to screen and retain superior individuals in the population, eliminate inferior individuals, and generate a new generation of population with better objective function values.

[0160] Specifically, the selection operation adopts a combination of roulette wheel selection and elite retention strategy. The selection probability is allocated based on the hierarchical priority of non-dominated sorting. The higher the level of the scheme, the greater the selection probability. At the same time, the high-quality schemes with the highest crowding value in each level are directly retained to the new generation of population to ensure the stable inheritance of excellent genes.

[0161] Furthermore, the crossover operation, targeting two selected parent schemes, employs a single-point crossover approach. A crossover point is randomly chosen from the sequence of maintenance action combinations, and the maintenance actions following the crossover point are swapped, generating a offspring scheme that incorporates the advantages of the parent schemes. Simultaneously, the mutation operation randomly selects some schemes from the population and fine-tunes their maintenance action combinations, including replacing individual maintenance actions and adjusting the execution order of maintenance actions. This introduces new decision variables to prevent the population from getting trapped in local optima.

[0162] Furthermore, the calculation of multi-objective fitness vectors, feasibility verification, and genetic operations are repeated until the preset maximum number of iterations is reached. Each iteration optimizes and upgrades the population, gradually increasing the total value of the solution and gradually reducing the total downtime, while always meeting maintenance resource constraints.

[0163] When the preset maximum number of iterations is reached, all solutions that satisfy maintenance resource constraints and are mutually independent are selected from the final generation population, forming a Pareto-optimal set of maintenance action execution solutions. These solutions achieve an optimal balance between total value gain and total maintenance downtime; no single solution is superior to others in both objectives, and each solution represents a unique trade-off.

[0164] Furthermore, after obtaining a set of Pareto optimal maintenance action execution schemes that cover multiple objective trade-off modes and adapt to maintenance resource constraints, the weighted comprehensive fitness of the total value benefit and total downtime of each scheme in the Pareto optimal scheme set is calculated based on the preset operation and maintenance preference weights, and the scheme with the highest comprehensive fitness is selected as the final execution scheme.

[0165] The operation and maintenance preference weights are dynamically adjusted based on the enterprise's emphasis on revenue and sensitivity to downtime losses to ensure that the weight settings align with the core needs of actual operation and maintenance scenarios. Specifically, the calculation of the weighted comprehensive fitness requires first standardizing the total value and total downtime of each solution to eliminate the impact of differences in the dimensions of different indicators. Then, each solution is multiplied by its corresponding operation and maintenance preference weight and summed to obtain a single-dimensional comprehensive evaluation index. This index allows for a direct comparison of the comprehensive advantages of each solution, thereby selecting the final execution solution that best suits the enterprise's operation and maintenance orientation.

[0166] For example, suppose a company is currently in its peak production season and is highly sensitive to downtime losses. The total value gain is weighted at 0.4, and the total downtime weighted at 0.6. A Pareto optimal solution set includes two options, X and Y. Option X has a total value gain of 100,000 yuan and a total downtime of 2 hours, while option Y has a total value gain of 120,000 yuan and a total downtime of 4 hours.

[0167] First, the indicators are standardized (assuming that the total value revenue is standardized to X = 0.5 and Y = 0.6, and the total downtime is standardized to X = 0.2 and Y = 0.4). Then, the weighted comprehensive fitness of option X = 0.5 × 0.4 + (1 - 0.2) × 0.6 = 0.68, and the weighted comprehensive fitness of option Y = 0.6 × 0.4 + (1 - 0.4) × 0.6 = 0.6. Option X has a higher comprehensive fitness and is selected as the final implementation option.

[0168] Conversely, if the company is in a period of equipment idleness, the emphasis is higher on revenue. The weight of total value revenue is adjusted to 0.7 and the weight of total downtime is 0.3. The overall adaptability of option X is recalculated as 0.5×0.7+0.8×0.3=0.59, and that of option Y is 0.6×0.7+0.6×0.3=0.6. In this case, option Y is more in line with the operation and maintenance orientation and becomes the final implementation option.

[0169] Furthermore, based on the expected maintenance value of each maintenance action in the final execution plan, execution priorities are generated by arranging them in descending order. Maintenance actions with higher expected maintenance value contribute more significantly to the overall operation and maintenance benefits, and therefore are given higher execution priorities. This ensures that high-value maintenance actions receive priority support from resources such as manpower and spare parts, and avoids delays or postponements of core benefit actions due to unreasonable priority ranking.

[0170] Furthermore, specific execution time windows are assigned to maintenance actions with execution priority. The allocation process for execution time windows needs to comprehensively consider the planned time window constraints in maintenance resource constraints, the expected execution duration of each maintenance action, and the company's production scheduling plan to ensure that the execution time period of each maintenance action does not conflict with other high-priority actions.

[0171] At the same time, the correlation and connection between maintenance actions must be taken into account. For example, some maintenance actions can only be performed after certain prerequisite actions are completed. Time windows need to be planned reasonably to ensure the smooth progress of the maintenance process.

[0172] Finally, the specific operation content, allocated execution time window, and corresponding execution priority of each maintenance action are integrated to generate a maintenance decision instruction set containing complete information, which is then output to the maintenance execution terminal.

[0173] The maintenance decision instruction set must be clear and explicit, providing maintenance personnel with direct and actionable guidelines to ensure that maintenance work can be carried out in an orderly and efficient manner according to the established plan, ultimately achieving the dual goals of maximizing the total value of operation and maintenance and minimizing the total downtime.

[0174] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects: This application proposes a digital twin-based full lifecycle maintenance management system for rotating machinery. First, a data acquisition and virtual-real synchronization module collects the real-time operating sequence of the target rotating machinery, extracting core operating status features such as effective vibration values ​​and average bearing temperatures. This drives the digital twin to update its virtual state and achieves precise virtual-real synchronization through consistency verification and parameter fine-tuning. Next, a health status trajectory prediction module calculates the degradation increment of health status parameters based on the synchronized virtual state and operating characteristics, generating a predicted health status trajectory containing multiple future time nodes through iterative forward simulation. Subsequently, a maintenance value assessment calculation module inputs this trajectory into a pre-trained maintenance value assessment model. A credibility score is constructed by combining the health status trajectory's dispersion coefficient and vibration-temperature correlation deviation coefficient. This score is then converted into a risk discount factor through nonlinear mapping to compensate for the initial predicted maintenance value, yielding the expected maintenance value for different maintenance actions. Finally, a maintenance decision optimization generation module aims to maximize total value gains and minimize total maintenance downtime. Combining maintenance resource constraints, it uses a multi-objective optimization algorithm to solve for the Pareto optimal solution. After filtering by maintenance preference weights, prioritizing, and allocating time windows, a maintenance decision instruction set is generated and output to the maintenance execution terminal.

[0175] The system provided in this application, through the technical solution of "virtual-real synchronous calibration - health trajectory prediction - maintenance value quantification - decision optimization generation", solves the problem that traditional maintenance management methods rely on manual experience and maintenance plans for reactive processing, resulting in maintenance work always lagging behind the equipment deterioration process. It realizes the intelligentization of the entire process of maintenance management from state perception and trend prediction to value assessment and decision execution, and minimizes maintenance costs and downtime losses. It provides technical support for the full life cycle operation and maintenance of key rotating machinery in industrial production.

[0176] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0177] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0178] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A digital twin-based full lifecycle maintenance management system for rotating machinery, characterized in that, The system includes: The data acquisition and virtual-real synchronization module is used to acquire the real-time operation sequence of the target rotating machinery, extract the set of operation status features, and synchronously acquire the current virtual state of the digital twin corresponding to the target rotating machinery. The health status trajectory prediction module is used to perform twin state synchronization and future multi-step evolution inference based on the set of operating status features and the current virtual state, and generate a predicted health status trajectory containing multiple future time nodes. The maintenance value assessment calculation module is used to input the predicted health status trajectory into a pre-trained maintenance value assessment model, and, in conjunction with a risk discount compensation mechanism, calculate the expected maintenance value of different maintenance actions performed at the current moment. The maintenance decision optimization generation module is used to optimize the solution based on the expected maintenance value, with the goal of maximizing the total value benefit and minimizing the total maintenance downtime, combined with the constraints of currently available maintenance resources, to generate a set of maintenance decision instructions and output them to the maintenance execution terminal.

2. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 1, characterized in that, Obtain the real-time operation sequence of the target rotating machinery and extract the set of operation state features. Simultaneously obtain the current virtual state of the digital twin corresponding to the target rotating machinery, including: By deploying vibration and temperature sensors on the target rotating machinery, real-time operating sequences within a continuous time window are collected, and time-domain statistical features are extracted to construct an operating state feature set, wherein the operating state feature set includes, but is not limited to, the effective value of vibration and the average value of bearing temperature; The set of operating status features is input into the digital twin, which drives the internal mechanism simulation model of the twin to update the corresponding health status parameters and dynamic load response, and generates the current virtual state that matches the physical entity in real time. The consistency of the indicators in the set of running state features with the corresponding simulation output in the current virtual state is checked. If the deviation exceeds the preset deviation threshold, the twin model parameters are adaptively fine-tuned until the current virtual state is confirmed as a valid synchronization state.

3. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 2, characterized in that, The consistency of the indicators in the set of operating state features with the corresponding simulation output in the current virtual state is checked. If the deviation exceeds a preset deviation threshold, adaptive fine-tuning of the twin model parameters is triggered until the current virtual state is confirmed as a valid synchronization state, including: Calculate the first absolute deviation between the effective vibration value in the set of operating state features and the effective vibration value simulated in the current virtual state; Calculate the second absolute deviation between the average bearing temperature in the set of operating state features and the average bearing temperature simulated in the current virtual state; The first absolute deviation and the second absolute deviation are assigned preset weights, and then weighted summation is performed to obtain the comprehensive deviation index. If the comprehensive deviation index does not exceed the preset deviation threshold, the current virtual state is directly marked as a valid synchronization state; If the comprehensive deviation index exceeds the preset deviation threshold, then based on the ratio of the first absolute deviation to the second absolute deviation, the physical parameters of the simulation model of the internal mechanism of the digital twin are adjusted in reverse compensation. The current virtual state is regenerated using the adjusted mechanism simulation model, and iterative verification is performed until the comprehensive deviation index is lower than the preset deviation threshold. The virtual state generated in this iteration is then confirmed as a valid synchronization state.

4. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 1, characterized in that, Based on the aforementioned set of operational state features and the current virtual state, twin state synchronization and future multi-step evolution projection are performed to generate a predicted health state trajectory containing multiple future time nodes, including: Based on the current virtual state, extract the initial health status parameters representing the overall health of the device from the digital twin; Using the effective vibration value and the average bearing temperature in the set of operating state features as inputs, and combining the dynamic load response in the current virtual state, the preset degradation rate mapping relationship is invoked to calculate the degradation increment of health state parameters within a unit time step. Starting with the initial health state parameters, and using the degradation increment of the health state parameters as the recursive step size, the digital twin is subjected to iterative forward simulation with discrete time steps. At the end of each iteration time step, the health status parameter values ​​generated by the simulation are recorded, and the iteration is continuously executed for a preset number of times to form a sequence consisting of the multiple future time nodes and their corresponding health status parameter values, which serves as the predicted health status trajectory.

5. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 1, characterized in that, The pre-training steps of the maintenance value assessment model include: Multiple sets of sample data of the target rotating machinery during the historical maintenance cycle are collected. Each set of sample data includes the sample health status trajectory, the maintenance action actually performed at the corresponding time, and the sample maintenance value label calculated based on the maintenance action. A maintenance value assessment model is constructed based on a neural network architecture. Using the health status trajectory of the sample as input features and the maintenance value label of the sample as a supervision signal, supervised training is performed on the maintenance value assessment model until the model loss function converges, thus completing the pre-training of the model.

6. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 1, characterized in that, Using a risk discount compensation mechanism, calculate the expected maintenance value of different maintenance actions performed at the current moment, including: Based on the vibration growth rate of the effective vibration value in the set of operating state features compared with the historical benchmark, the number of prediction time nodes is dynamically determined. The larger the vibration growth rate, the fewer prediction time nodes are determined. Based on the number of predicted time points, the twin state synchronization and future multi-step evolution prediction are performed to generate the predicted health state trajectory; The dispersion coefficient of the health status trajectory is obtained by the ratio of the standard deviation of the health status parameter values ​​at multiple future time points to the average value of the health status parameters in the predicted health status trajectory. Identify the equipment's operating conditions based on the current operating status feature set, select the corresponding time period from historical data under the same operating conditions, and calculate the vibration-temperature correlation coefficient between the effective vibration value and the average bearing temperature. The vibration temperature correlation coefficient is compared with the historical benchmark vibration temperature correlation coefficient under the same working conditions, and the vibration temperature correlation deviation coefficient is calculated based on the degree of deviation. The discrete coefficient of the health state trajectory and the deviation coefficient of the vibration temperature correlation are weighted and fused to generate a preliminary credibility score; The initial credibility score is attenuated and corrected based on the number of predicted time points to generate a final comprehensive credibility score. The fewer the number of predicted time points, the greater the degree of attenuation correction. The comprehensive credibility score is converted into a risk discount factor through a preset nonlinear mapping function, wherein the value range of the risk discount factor is [0, 1]. The predicted health status trajectory is input into the pre-trained maintenance value assessment model, and the initial predicted maintenance value of different maintenance actions is output. The risk discount factor is applied to compensate the initial predicted maintenance value to generate the compensated expected maintenance value.

7. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 6, characterized in that, The comprehensive credibility score is converted into a risk discount factor through a preset nonlinear mapping function, including: Determine the relationship between the overall credibility score and the preset high credibility threshold and the preset low credibility threshold; If the overall credibility score is greater than or equal to the preset high credibility threshold, the risk discount factor is set to a value of 1. If the overall credibility score is less than or equal to the preset low credibility threshold, the risk discount factor is set to a value of 0. If the overall credibility score is greater than the preset low credibility threshold and less than the preset high credibility threshold, then a risk discount factor between 0 and 1 is calculated based on the overall credibility score using a nonlinear decay formula. The lower the overall credibility score, the smaller the calculated risk discount factor value.

8. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 6, characterized in that, The risk discount factor is applied to compensate the initial predicted maintenance value to generate the compensated expected maintenance value, including: If the overall credibility score is greater than or equal to the preset high credibility threshold, then the compensated expected maintenance value is equal to the initial predicted maintenance value. If the overall credibility score is less than or equal to the preset low credibility threshold, then the compensated expected maintenance value is equal to the negative maintenance action cost. If the overall credibility score is greater than the preset low credibility threshold and less than the preset high credibility threshold, then the compensated expected maintenance value is equal to the risk discount factor multiplied by the expected return, minus the maintenance cost. The maintenance cost is the total cost incurred in performing the current maintenance action, including labor costs, spare parts and material costs, and opportunity costs of downtime during the maintenance period. The expected revenue is derived from the initial predicted maintenance value.

9. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 1, characterized in that, Based on the expected maintenance value, with the objectives of maximizing total value revenue and minimizing total maintenance downtime, and considering the constraints of currently available maintenance resources, an optimization solution is performed to generate a maintenance decision instruction set, including: Construct a multi-objective optimization function with the core objective of maximizing total value revenue and minimizing total maintenance downtime. Here, total value revenue is the sum of the expected maintenance value after compensation for all optional maintenance actions, and total maintenance downtime is the sum of the downtime caused by the execution of all optional maintenance actions. Establish maintenance resource constraints, wherein the maintenance resource constraints include at least the total available maintenance personnel man-hours constraint, the inventory quantity constraint of each type of spare parts, and the planned time window constraint for different maintenance actions; A multi-objective optimization algorithm is invoked, guided by the multi-objective optimization function, to solve the problem under the constraints of the maintenance resources, thereby obtaining a set of Pareto-optimal maintenance action execution schemes. Based on preset operation and maintenance preference weights, the weighted comprehensive fitness of the total value benefit and total downtime of each scheme in the Pareto optimal maintenance action execution scheme set is calculated, and the scheme with the highest comprehensive fitness is selected as the final execution scheme. Based on the expected maintenance value of the maintenance actions in the final execution plan, the execution priorities are generated by sorting them in descending order; Assign specific execution time windows to maintenance actions with the aforementioned execution priority, and generate a set of maintenance decision instructions that includes the maintenance action content, execution time window, and priority.

10. The rotating machinery lifecycle maintenance management system based on digital twins according to claim 9, characterized in that, A multi-objective optimization algorithm is invoked, guided by the multi-objective optimization function, to solve the problem under the constraints of the maintenance resources, thereby obtaining a set of Pareto-optimal maintenance action execution schemes, including: Multiple maintenance action execution schemes are randomly generated as the initial population individuals, and each scheme explicitly contains a set of maintenance actions to be executed; For each maintenance action execution plan in the population, the objective function values ​​of the total value revenue and the total maintenance downtime are calculated respectively to form a multi-objective fitness vector. The feasibility of the plan is verified by the maintenance resource constraints, and invalid plans that do not meet the constraints are eliminated. Based on the multi-objective fitness vector, non-dominated sorting and crowding calculation are performed on the individuals in the initial population. Selection, crossover and mutation genetic operations are performed by combining the sorting results and crowding values ​​to retain the superior individuals in the population and generate a new generation of population. Repeat the calculation, feasibility verification, and genetic operation steps of the multi-objective fitness vector until the preset maximum number of iterations is reached; From the final generation population, select all schemes that satisfy the aforementioned maintenance resource constraints and are mutually independent to form a Pareto optimal maintenance action execution scheme set.

Images