Refrigeration equipment intelligent maintenance system based on digital twinning

Abstract

The present application relates to the technical field of intelligent maintenance of refrigeration equipment, in particular to a kind of intelligent maintenance system of refrigeration equipment based on digital twinning;Contain working condition perception, twin model management, stress fault decoupling, transient fault diagnosis, maintenance risk decision and closed-loop control execution unit;System determines transient disturbance index through working condition perception, and dynamically constructs twin model;Its core is to decouple out fault deviation index using dynamic stress baseline model, and diagnose fault occurrence likelihood under specific working condition, and then weighted fusion into comprehensive maintenance risk index;System executes performance optimization, life extension with disease or safety protection strategy according to risk level;The present application realizes the effective distinction of severe working condition disturbance and early fault by dynamic stress baseline model, reduces false alarm and false negative under high load.

Description

A digital twin-based intelligent maintenance system for refrigeration equipment Technical Field

[0001] This invention relates to the field of intelligent maintenance technology for refrigeration equipment, specifically to an intelligent maintenance system for refrigeration equipment based on digital twins. Background Technology

[0002] In the operation and maintenance of refrigeration equipment, the equipment is often subjected to severe operating disturbances and transient shocks. Existing technologies have significant limitations in monitoring and diagnosis. Traditional methods struggle to effectively distinguish between normal performance fluctuations caused by severe operating disturbances and early faults caused by deterioration in equipment health, easily leading to false alarms or missed alarms. When the equipment is under transient shock conditions, the system struggles to accurately capture subtle fault characteristics, resulting in decreased diagnostic accuracy. Furthermore, existing maintenance strategies are mostly passive alarms, lacking proactive, closed-loop, and adaptive control based on real-time risk assessments. Therefore, how to achieve accurate fault identification and forward-looking intelligent maintenance decisions under complex disturbances is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an intelligent maintenance system for refrigeration equipment based on digital twins. Specifically, the technical solution of this invention includes:

[0004] The operating condition sensing unit is used to collect multi-dimensional operating data of the refrigeration equipment, analyze the multi-dimensional operating data to obtain the transient disturbance index, and compare the transient disturbance index with a preset disturbance index threshold to obtain the fidelity level.

[0005] The twin model management unit is used to dynamically construct twin models in response to fidelity levels and parse the twin models to obtain internal state parameters;

[0006] The stress fault decoupling unit is used to combine the transient disturbance index and the preset dynamic stress baseline model to calculate the dynamic stress baseline, and to calculate the deviation between the dynamic stress baseline and the collected real-time sensor measurement values ​​to obtain the fault deviation index. Then, the fault deviation index is judged to obtain the decoupling evaluation result.

[0007] The transient fault diagnosis unit is used to construct a transient fault feature dataset based on internal state parameters and real-time sensor measurements when the fidelity level is extreme impact conditions and the decoupling evaluation result is an early fault warning. The dataset is then sent to a preset transient fault diagnosis model for analysis to obtain the fault occurrence likelihood.

[0008] The maintenance risk decision unit is used to obtain the fault deviation index and the fault occurrence likelihood, perform weighted fusion processing on the two to obtain the comprehensive maintenance risk index, and compare the comprehensive maintenance risk index with the preset risk threshold to obtain the risk level signal.

[0009] The closed-loop control execution unit is used to respond to risk level signals and execute corresponding performance optimization strategies, life extension strategies for existing defects, or safety protection strategies.

[0010] Preferably, the transient disturbance index analysis process is as follows:

[0011] Collect multi-dimensional operational data, including pressure at key measuring points, temperature at key measuring points, and the status of the chamber door opening and closing.

[0012] The time-varying rate of pressure and temperature at key measuring points is calculated, and the time-varying rate is normalized according to a preset reference standard to obtain the normalized values ​​of pressure change rate and temperature change rate.

[0013] The normalized values ​​of pressure change rate, temperature change rate, and door opening / closing status are weighted and summed according to preset contribution weights to obtain the transient disturbance index.

[0014] Preferably, the twin model management unit is further used for:

[0015] When the fidelity level is at steady-state operating condition, the low-fidelity model is called to analyze all components;

[0016] When the fidelity level is transient stress condition, the preset hybrid fidelity model is invoked;

[0017] When the fidelity level is extreme impact condition, the high-fidelity model is called to analyze the preset key components, and the low-fidelity model is called to analyze the preset non-key components.

[0018] Preferably, the deviation calculation process of the stress fault decoupling unit is as follows:

[0019] Based on the transient disturbance index, the dynamic stress baseline model is invoked to predict and generate the dynamic stress baseline;

[0020] Acquire real-time sensor measurements;

[0021] The absolute deviation between the real-time sensor measurement and the dynamic stress baseline is calculated, and the absolute deviation is divided by the dynamic stress baseline to obtain the fault deviation index.

[0022] Preferably, the process for generating the decoupling evaluation result is as follows:

[0023] The fault deviation index is compared and analyzed with the preset first deviation threshold and the preset second deviation threshold;

[0024] When the fault deviation index is less than or equal to the first deviation threshold, a normal stress signal is generated;

[0025] When the fault deviation index is greater than the first deviation threshold and less than or equal to the second deviation threshold, an early fault warning signal is generated.

[0026] When the fault deviation index is greater than the second deviation threshold, a critical fault alarm signal is generated.

[0027] Preferably, the transient fault feature dataset includes:

[0028] Internal state parameters generated by the twin model management unit through high-fidelity model analysis under extreme impact conditions;

[0029] Real-time sensor measurements collected by the stress fault decoupling unit.

[0030] Preferably, the specific process of the maintenance risk decision-making unit is as follows:

[0031] Obtain the largest fault deviation index among all key parameters and set it as the maximum fault deviation index.

[0032] Obtain the highest fault occurrence likelihood among all diagnosed faults and set it as the maximum fault occurrence likelihood.

[0033] The maximum failure deviation index and the maximum failure occurrence likelihood are weighted and summed according to the preset risk weight coefficients to obtain the comprehensive maintenance risk index.

[0034] Preferably, the specific execution logic of the closed-loop control execution unit is as follows:

[0035] When the comprehensive maintenance risk index is less than the preset first risk threshold, it is judged as a normal stress and a performance optimization strategy is executed.

[0036] When the comprehensive maintenance risk index is greater than or equal to the first risk threshold and less than the second risk threshold, it is judged as an early failure and the strategy of extending service life with defects is implemented.

[0037] When the comprehensive maintenance risk index is greater than or equal to the second risk threshold, it is judged as a serious fault, and a safety protection strategy is implemented.

[0038] Preferably, the performance optimization strategy includes adjusting the condenser fan speed; the extended service life strategy includes increasing the evaporator set temperature or limiting the compressor maximum speed.

[0039] Preferably, the safety protection strategy includes reducing the compressor speed to a safe lower limit or executing an emergency shutdown procedure.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. This system constructs a dynamic stress baseline model, which can dynamically predict the benchmark for healthy operation of equipment based on the intensity of real-time operating condition disturbances. By calculating the deviation index between the actual measured value and this dynamic benchmark, it successfully distinguishes between normal performance fluctuations caused by severe operating condition disturbances and abnormal deviations from early equipment faults, significantly reducing false alarms and missed alarms generated by the traditional fixed threshold method under high load.

[0042] 2. This system achieves optimal allocation of computing resources by combining condition perception with dynamic management of twin models. The system only calls on high-fidelity twin models to analyze key components when extreme impact conditions are identified, thereby obtaining internal state parameters that are difficult to measure directly by conventional sensors. This ensures high-efficiency response under stable conditions and retains high-precision diagnostic capabilities under transient impacts, effectively balancing the contradiction between computational accuracy and real-time performance.

[0043] 3. This system solves the technical problem of capturing weak fault features under high disturbance background through transient fault diagnosis unit. The unit is activated under the dual conditions of extreme working conditions and early warning. It uses internal state parameters generated by high-fidelity model and sensor data to construct feature set, which enables the diagnostic model to accurately identify early faults hidden under high stress, and greatly improves the accuracy of transient diagnosis.

[0044] 4. This system achieves a transformation from passive alarm to proactive intelligent maintenance through maintenance risk decision-making and closed-loop control execution. The system integrates the severity of fault deviation with the confidence of fault diagnosis into a comprehensive maintenance risk index, and automatically executes hierarchical control strategies based on this index, including performance optimization during normal operation, life extension during early-stage faults, and safety protection during severe faults, forming a complete closed-loop adaptive maintenance link, which improves the operational reliability and maintenance efficiency of the equipment. Attached Figure Description

[0045] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0046] Figure 1 is a structural diagram of the system of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0048] Example 1:

[0049] Please refer to Figure 1. A digital twin-based intelligent maintenance system for refrigeration equipment includes:

[0050] The operating condition sensing unit is used to collect multi-dimensional operating data of the refrigeration equipment, analyze the multi-dimensional operating data to obtain the transient disturbance index, and compare the transient disturbance index with a preset disturbance index threshold to obtain the fidelity level.

[0051] The twin model management unit is used to dynamically construct twin models in response to fidelity levels and parse the twin models to obtain internal state parameters;

[0052] The stress fault decoupling unit is used to combine the transient disturbance index and the preset dynamic stress baseline model to calculate the dynamic stress baseline, and to calculate the deviation between the dynamic stress baseline and the collected real-time sensor measurement values ​​to obtain the fault deviation index. Then, the fault deviation index is judged to obtain the decoupling evaluation result.

[0053] The transient fault diagnosis unit is used to construct a transient fault feature dataset based on internal state parameters and real-time sensor measurements when the fidelity level is extreme impact conditions and the decoupling evaluation result is an early fault warning. The dataset is then sent to a preset transient fault diagnosis model for analysis to obtain the fault occurrence likelihood.

[0054] The maintenance risk decision unit is used to obtain the fault deviation index and the fault occurrence likelihood, perform weighted fusion processing on the two to obtain the comprehensive maintenance risk index, and compare the comprehensive maintenance risk index with the preset risk threshold to obtain the risk level signal.

[0055] The closed-loop control execution unit is used to respond to risk level signals and execute corresponding performance optimization strategies, life extension strategies for existing defects, or safety protection strategies.

[0056] This embodiment provides an intelligent maintenance system for refrigeration equipment based on digital twins. Its purpose is to solve the technical problems in the prior art, such as the difficulty in distinguishing between severe operating condition disturbances and early faults, the inability to accurately diagnose under transient shocks, and the lack of closed-loop adaptive maintenance strategies. The system realizes intelligent and proactive maintenance of refrigeration equipment by constructing a complete technical link from operating condition perception, model adaptation, stress-fault decoupling, transient diagnosis to risk decision-making and closed-loop control.

[0057] The system includes:

[0058] The operating condition sensing unit aims to quantify the intensity of operational disturbances experienced by the refrigeration equipment in real time and provide a basis for decision-making in subsequent twin model switching. This unit is used to collect multi-dimensional operating data of the refrigeration equipment. In this embodiment, the multi-dimensional operating data may include the pressure at key measuring points obtained through pressure sensors. Temperature at key measuring points obtained through temperature sensors and the door opening / closing status obtained through Hall effect sensors. This unit performs transient perturbation index analysis on multi-dimensional operational data. To integrate dynamic changes in different physical domains with external perturbations into a single scalar, this embodiment introduces a transient perturbation index. ;

[0059] The analytical process includes calculating the pressure at key measuring points. and temperature rate of change over time and The calculated rate of change of time is normalized, and the normalized value is compared with the door opening / closing status. Perform a weighted summation to obtain the transient perturbation index. This unit will also include the transient disturbance index. A comparative analysis is performed with a preset disturbance index threshold; wherein, the preset disturbance index threshold, for example... and This refers to the critical value used to classify the severity of operating conditions. It is derived from the dimensionless values ​​of the system from steady state to transient state and from transient state to extreme shock, as calibrated from historical data. Critical values ​​are determined; specifically, the calibration process may include: acquiring historical operating data. The dataset was analyzed, and the K-means clustering algorithm was used to divide the dataset into three clusters: steady-state, transient stress, and extreme shock. The sample boundary value between the steady-state cluster and the transient stress cluster was taken as... The sample boundary values ​​of the transient stress cluster and the extreme shock cluster are taken as... ;

[0060] when When the system is in a steady-state condition, it is determined to be in a steady-state condition; when When, it is determined to be a transient stress condition; when At that time, it was determined to be an extreme impact condition; the unit obtained the fidelity level based on the comparative analysis results;

[0061] Preset switching function The segmented control logic is adopted, and its specific calculation expression is as follows:

[0062] ;

[0063] in, A system-level twin model built for the current moment; The transient disturbance index; This represents the boundary threshold between steady-state and transient stress. This represents the boundary threshold between transient stress and extreme impact. For all components; This refers to the set of sub-components selected based on sensitivity analysis under transient stress conditions.

[0064] Low-fidelity model ( ): It adopts the lumped parameter method to construct the structure, which simplifies components such as condensers into 0-dimensional nodes, ignores spatial distribution, and uses only ordinary differential equations to describe the conservation of mass and energy. It has a fast calculation speed and is suitable for steady-state analysis.

[0065] High-fidelity model ( ): Constructed using distributed parameter method or finite element analysis. For key components... For example, with a compressor, a three-dimensional geometric model is established and meshed. Partial differential equations such as the Navier-Stokes equations are used to solve the transient coupling field between the fluid and the structure, and the internal state parameters can be output.

[0066] Hybrid fidelity model and medium granularity ( A one-dimensional flow net model is used, discretizing the heat exchanger into several control volumes along the pipeline direction, considering parameter variations along the flow direction but neglecting radial distribution; the selection method for some key components is: calculating the transient disturbance exponent of each component's state parameters. The sensitivity coefficient is used to select components whose sensitivity coefficient exceeds a preset threshold (e.g., 0.8). Medium-grained modeling is performed; the unit finally analyzes the twin model to obtain internal state parameters. These internal state parameters are deep physical quantities that cannot be directly measured by conventional non-invasive sensors in the actual operation scenario of mass-produced commercial refrigeration equipment due to cost and sealing structure limitations.

[0067] The stress-fault decoupling unit's core technology aims to distinguish between normal performance fluctuations caused by severe operating conditions (stress) and abnormal deviations caused by deterioration in equipment health (fault). This unit combines transient disturbance indices from the operating condition sensing unit. and the pre-defined dynamic stress baseline model The dynamic stress baseline was calculated. Among them, the preset dynamic stress baseline model This refers to a system based on a large number of historical health devices in different... The system uses operational data and machine learning techniques, such as Gaussian process regression, to train predictive models; its technological motivation is to provide a dynamic standard of health. Its function is to predict how well a device will respond to a specific disturbance under fault-free conditions. The normal stress response that should occur; this unit will also establish a dynamic stress baseline. Real-time measurements from the collected sensors The deviation is calculated to obtain the fault deviation index. This solution process aims to quantify the relative deviation between the actual state and the state of healthy stress, thereby separating the factors caused by stress. Normal fluctuations caused; this unit's fault deviation index The system performs discrimination processing to obtain decoupling evaluation results, such as normal stress, early fault warning, or serious fault alarm.

[0068] The transient fault diagnosis unit aims to perform in-depth diagnosis using high-fidelity data when the system is under the most severe disturbance and early fault characteristics have appeared, in order to determine the root cause of the specific fault. This unit has a clear triggering logic: when the fidelity level determined by the condition sensing unit is an extreme impact condition, i.e. And when the decoupling evaluation result from the stress fault decoupling unit is an early fault warning, for example... The unit is activated; this logic links the results of condition awareness and decoupled assessment, ensuring that high-cost diagnostics are invoked only when most needed, i.e., when an impact occurs, and most suspicious, i.e., when a deviation occurs; the unit retrieves internal state parameters generated by the twin model management unit, such as and real-time measurements from sensors, such as Construct a transient fault feature dataset ;

[0069] This unit will also include transient fault feature datasets. Send to the preset transient fault diagnosis model Analysis is performed; among which, the preset transient fault diagnosis model is used. This refers to a deep learning model, such as one based on a Convolutional Neural Network (CNN), which is trained offline using historical data, including normal stress data and known fault data under different transient impacts. In a specific implementation, the construction process of this CNN model includes:

[0070] Dataset Construction: Constructing a transient fault feature dataset Divide the time series into fixed time windows; Input tensor: construct a tensor for each segment. A two-dimensional tensor is used as the input to the CNN;

[0071] Network architecture: This CNN model may include: two one-dimensional convolutional layers for extracting temporal features, followed by a max pooling layer for dimensionality reduction, and finally connected to one or more fully connected layers;

[0072] Output layer: The output layer is for For a specific type of fault, adopt Each node has a Sigmoid activation function and outputs... Each value between 0 and 1 corresponds to a specific fault. Fault occurrence likelihood ;

[0073] Its technological advantage lies in its ability to learn under high-stress conditions, i.e., high... Identify the subtle timing characteristics of specific faults, such as valve plate leakage; Model For dataset and transient disturbance index Perform non-linear pattern recognition, for example To obtain the probability of occurrence of a specific fault f. It is a value between 0 and 1;

[0074] The purpose of maintaining the risk decision-making unit is to integrate the system's anomaly degree, i.e., deviation index, and fault certainty, i.e., likelihood, into a single, action-guided risk indicator; this unit acquires the fault deviation index from the stress-fault decoupling unit. And the probability of fault occurrence from the transient fault diagnosis unit The fault deviation index and fault occurrence likelihood are then weighted and fused to obtain the comprehensive maintenance risk index. The fusion process is as follows: This unit will also comprehensively maintain the risk index. Compared with preset risk thresholds, such as A comparative analysis was conducted; among which, a preset risk threshold was used. The source is based on history Risk level classification of values; obtaining risk level signals, such as normal stress, early failure, or severe failure;

[0075] The closed-loop control execution unit aims to automatically execute the optimal control strategy in response to the risk level of the decision-making unit, realizing an intelligent maintenance closed loop from passive prediction to active adaptation. This unit responds to the risk level signal output by the previous unit and executes the corresponding control strategy. Its execution logic may include: when the risk level signal is a normal stress response, i.e. At the same time, performance optimization strategies are implemented, such as adjusting the condenser fan speed. To adapt to high heat loads; when the risk level signal is an early failure, i.e. At that time, implement a strategy to extend the life of the evaporator even with defects, such as proactively increasing the evaporator set temperature. Or limit the compressor's maximum speed This reduces stress on faulty components and generates predictive maintenance work orders; when the risk level signal is a severe fault, i.e. At that time, implement safety protection strategies, such as turning the compressor... Reduce to the safety minimum or execute emergency shutdown procedures to ensure equipment safety;

[0076] This embodiment, through the collaborative work of the aforementioned units, constructs a complete technical chain from condition perception, model adaptation, stress-fault decoupling, transient diagnosis to risk decision-making and closed-loop control. The technical effects of this embodiment are as follows: Through a variable-fidelity twin model, it achieves an optimal balance between computational accuracy and efficiency under different operating conditions; through dynamic stress baseline and fault deviation index, it innovatively solves the technical challenge of distinguishing between severe operating condition disturbances (normal stress) and early faults (abnormal deviation); and by triggering high-fidelity diagnosis under extreme operating conditions, it achieves accurate capture of subtle fault characteristics; simultaneously, through adaptive closed-loop control, it realizes intelligent maintenance from passive alarm to proactive performance optimization, extended service life with defects, or safety protection, significantly improving the operational reliability, safety, and maintenance efficiency of refrigeration equipment.

[0077] Example 2:

[0078] The analysis process of the transient disturbance index is as follows:

[0079] Collect multi-dimensional operational data, including pressure at key measuring points, temperature at key measuring points, and the status of the chamber door opening and closing.

[0080] The time-varying rate of pressure and temperature at key measuring points is calculated, and the time-varying rate is normalized according to a preset reference standard to obtain the normalized values ​​of pressure change rate and temperature change rate.

[0081] The normalized values ​​of pressure change rate, temperature change rate, and door opening / closing status are weighted and summed according to preset contribution weights to obtain the transient disturbance index.

[0082] This embodiment is a refinement of the transient disturbance index analysis process described in Embodiment 1; the process aims to fuse multi-source, heterogeneous operational data into a single scalar characterizing the total intensity of the system's transient disturbances. ;

[0083] In this embodiment, multi-dimensional operational data is collected, including pressure at key measuring points acquired by a pressure sensor. Temperature at key measuring points obtained through temperature sensors and the door opening / closing status obtained through Hall effect sensors. ;

[0084] The time-varying rate of change of pressure and temperature at key measuring points is calculated to obtain the pressure. rate of change over time and temperature rate of change over time The rate of change over time is normalized according to a preset reference standard; the purpose of this normalization is to eliminate... and The physical dimensions are determined for subsequent fusion; its calculation formula is: and ;in, The normalized value of the rate of change of pressure is a dimensionless pure number, derived from... and Calculated; For pressure The rate of change over time, its source being the pressure acquired by the sensor. The sequence is obtained by performing difference or differentiation operations on the sequence; The preset pressure change rate reference benchmark, such as 5000 Pa / s, is based on statistical analysis of historical operating data, such as the 95th percentile or equipment design limit calibration, to represent a benchmark of significant change. The normalized value of the rate of change of temperature is a dimensionless pure number, derived from... and Calculated; For temperature The rate of change over time, its source being the temperature acquired by the sensor. The sequence is obtained by performing difference or differentiation operations on the sequence; The preset temperature change rate reference is, for example, 0.5 K / s, and its source or value is based on the same... ;

[0085] The system will obtain dimensionless values. and Normalized value of pressure change rate Normalized value of temperature change rate and the status of the cabinet door opening and closing The transient disturbance index is obtained by performing a weighted summation based on preset contribution weights. ;

[0086] The technical motivation lies in the dimensionless value calculated in the previous step for the weighted summation formula. and and dimensionless states As input, the final dimensionless perturbation exponent is obtained. In this embodiment, the calculation formula uses linear weighted summation to simplify the calculation. ;in, The transient perturbation exponent is a dimensionless pure number. The predefined dimensionless contribution weight is a dimensionless pure number and satisfies... Its source or value basis is determined through the AHP (Analytic Hierarchy Process) or expert experience method, and is used to reflect... The contribution of the three factors to the total disturbance; in other embodiments, to achieve higher physical fidelity, It can also be achieved through nonlinear functions. To construct a model to characterize the coupling and nonlinear effects between various disturbance sources; The state of the cabinet door is a dimensionless Boolean value, where the definition is... This indicates that the box door is open. This indicates that the cabinet door is closed, and the information is collected by a Hall sensor.

[0087] This embodiment, through the above-described normalization and weighted summation processing methods, successfully integrates multi-source heterogeneous data from pressure, temperature, and door opening / closing into a single, dimensionless transient disturbance index. This index can accurately measure the total intensity of instantaneous disturbances suffered by the system in real time, providing an objective and quantitative basis for classifying the fidelity level of subsequent working condition sensing units.

[0088] Example 3:

[0089] The twin model management unit is also used for:

[0090] When the fidelity level is at steady-state operating condition, the low-fidelity model is called to analyze all components;

[0091] When the fidelity level is transient stress condition, the preset hybrid fidelity model is invoked;

[0092] When the fidelity level is extreme impact condition, the high-fidelity model is called to analyze the preset key components, and the low-fidelity model is called to analyze the preset non-key components.

[0093] This embodiment is a refinement of the logic for dynamically constructing the twin model by the twin model management unit described in Embodiment 1; this unit executes a preset switching function based on the fidelity level output by the working condition sensing unit. The specific switching logic is as follows:

[0094] When the fidelity level is at steady-state operating conditions, for example, as calculated in Example 2 This indicates that the system is in a stable operating state. At this time, the requirement for computational accuracy is not high, but the requirement for response speed is high. The twin model management unit will call the low-fidelity model. Analyze all components; including the low-fidelity model. This refers to a simplified lumped parameter model or a surrogate model trained based on historical steady-state data; its technical characteristics are fast computation speed and high convergence, and it is suitable for describing the macroscopic characteristics of a system under steady-state conditions.

[0095] When the fidelity level is transient stress condition, for example This indicates that the system is experiencing a moderate disturbance; at this point, the system invokes the preset hybrid fidelity model. Among them, the hybrid fidelity model This refers to a model that strikes a trade-off between computational accuracy and speed. For example, it may use a medium-granularity physical model for some key components, while still using a lumped parameter model for other components to cope with medium disturbances.

[0096] When the fidelity level is at extreme impact conditions, for example This indicates that the system has experienced severe disturbances, at which point high-precision simulation of key components is necessary to capture transient details; the twin model management unit will then call upon the high-fidelity model. The key components of the pre-defined set are the parsing. Such as compressors, and calling low-fidelity models. Parse the pre-defined non-critical components, i.e., the set Such as pipelines; among them, high-fidelity models This refers to a refined model constructed based on a three-dimensional physical model and multi-physics coupling, such as thermodynamic-fluid-structure coupling; its technical characteristics are high computational accuracy and the ability to resolve internal state parameters, such as valve plate temperature, but it consumes a lot of computational resources.

[0097] This embodiment, through the aforementioned hierarchical and dynamic model switching logic, achieves on-demand allocation and dynamic focusing of computing resources; in steady state, it uses... It ensures a high response speed; under extreme impact, through and This combination allows valuable computing resources to be focused on analyzing the critical components affected by the impact. At the same time, a low-fidelity model is used. Rapidly calculate non-critical components that occupy most of the system volume This allows for the accurate capture of key local areas without sacrificing the overall system's computational efficiency. This variable-fidelity modeling method perfectly balances the contradiction between simulation accuracy under high disturbances and system real-time performance, ensuring the efficiency and usability of the twin model under all operating conditions.

[0098] Example 4:

[0099] The deviation calculation process of the stress fault decoupling unit is as follows:

[0100] Based on the transient disturbance index, the dynamic stress baseline model is invoked to predict and generate the dynamic stress baseline;

[0101] Acquire real-time sensor measurements;

[0102] The absolute deviation between the real-time sensor measurement and the dynamic stress baseline is calculated, and the absolute deviation is divided by the dynamic stress baseline to obtain the fault deviation index.

[0103] This embodiment is a refinement of the deviation calculation process of the stress fault decoupling unit described in Embodiment 1; the core of this process is to establish a dynamic health standard and quantify the deviation of the actual measurement value from the standard.

[0104] Transient disturbance index generated by the operating condition sensing unit Call the dynamic stress baseline model Predict and generate key parameters dynamic stress baseline Key parameters Specifically, this includes, but is not limited to, core indicators characterizing the thermodynamic cycle of the system, such as compressor suction pressure, compressor discharge temperature, and evaporator coil temperature.

[0105] The technical motivation lies in the fact that the prediction process aims to establish a system that adapts to different operating conditions. The real-time changing normal reference point; its calculation formula can be expressed as: ;in, Key parameters The dynamic stress baseline, which is derived from the model. and It is calculated; its technical meaning is: under the current disturbance Next, this parameter of a health device The normal response value; This represents a parameter vector describing the current quasi-steady-state operating condition, which may include key operating condition parameters such as ambient temperature and evaporator set temperature. For parameters The pre-trained baseline model is derived from or based on a large number of historical health devices in different... and working conditions The combined operational data is trained using machine learning techniques, such as Gaussian process regression. In a specific implementation, the construction process of this Gaussian process regression (GPR) model includes:

[0106] Feature construction: Transient perturbation index and operating condition parameter vector Combined into input feature vector ;

[0107] Model training: using historical health datasets Training the GPR model;

[0108] Kernel function selection: Radial basis function kernel is selected.

[0109] ;

[0110] in, The input feature vector; The signal variance is used to control the vertical scale of the kernel function, characterizing the range of variation of the amplitude of the dynamic stress baseline latent function. The length scale parameter controls the horizontal scale of the kernel function, determining the model's response to the input. Sensitivity to change; if A smaller value means that the baseline model can capture faster fluctuations in operating conditions;

[0111] The hyperparameters of the kernel function are optimized by maximizing the marginal likelihood function. After training, the model... It can then be based on the new input Predicting dynamic stress baseline ; This is the transient disturbance index, which is calculated by the operating condition sensing unit. These are real-time measurements from the sensors, which are collected in real-time by the corresponding sensors.

[0112] Obtain this key parameter Real-time sensor measurements ;

[0113] This process calculates the real-time measurements from the sensor. Compared with dynamic stress baseline The relative deviation between them yields the fault deviation index. The process specifically involves calculating the absolute deviation between the two. And divide the absolute deviation by the dynamic stress baseline. With the preset minimum positive number sum;

[0114] The technical motivation behind this formula lies in quantifying the true state. Health stress The relative deviation between them, thus stripping away the material The resulting normal fluctuations make It only reflects abnormal deviations caused by potential faults; its calculation formula is: ;in, The fault deviation index is a dimensionless pure number, its origin being... and Calculated; It is a preset, extremely small positive number set to ensure computational robustness; its technical motivation is to prevent when... When the value instantaneously approaches 0 under certain operating conditions, it can lead to computational overflow or undefined results due to a denominator of 0. This ensures that the model's behavior conforms to physical principles under all input values; and it also ensures dimensional consistency. The value should have the same as Same physical dimensions; due to and Having the same physical dimensions, such as MPa or ,therefore It is a dimensionless value, which makes it easy to set a uniform evaluation threshold; These are real-time measurements from the sensors, which are collected in real-time by the corresponding sensors.

[0115] This embodiment introduces a random... Dynamically changing baseline And calculate the relative deviation index. Successfully transformed the normal value caused by external operating condition disturbances. Abnormal deviations caused by deterioration of internal health status, i.e., molecular To decouple; By eliminating the influence of operating condition fluctuations, it reflects only the real abnormal deviations caused by potential faults, greatly improving the accuracy of fault identification and avoiding false alarms generated by traditional fixed thresholds under high loads.

[0116] Example 5:

[0117] The process of generating the decoupling evaluation results is as follows:

[0118] The fault deviation index is compared and analyzed with the preset first deviation threshold and the preset second deviation threshold;

[0119] When the fault deviation index is less than or equal to the first deviation threshold, a normal stress signal is generated;

[0120] When the fault deviation index is greater than the first deviation threshold and less than or equal to the second deviation threshold, an early fault warning signal is generated.

[0121] When the fault deviation index is greater than the second deviation threshold, a critical fault alarm signal is generated.

[0122] This embodiment follows the process of generating the decoupling evaluation results from Embodiment 4; after calculating the fault deviation index... Subsequently, the stress fault decoupling unit performs hierarchical discrimination on it;

[0123] This process will include the fault deviation index. Deviation from the preset first threshold and the preset second deviation threshold Perform comparative analysis;

[0124] Among them, the first deviation threshold Second deviation threshold Refers to the division The risk level thresholds are all dimensionless pure numbers; their source or basis for value determination is based on historical early failure data. The statistical distribution settings, for example , This ensures the capture of statistically significant deviations. This defines the scope of serious deviation;

[0125] The logic of this comparative analysis is as follows:

[0126] When the fault deviation index Less than or equal to the first deviation threshold ,Right now When the system determines that the equipment deviates from the normal stress range, it generates a normal stress signal.

[0127] When the fault deviation index Greater than the first deviation threshold And less than or equal to the second deviation threshold ,Right now When the system determines that the equipment has deviated from the normal stress baseline and there is an early failure risk, it generates an early failure warning signal; this signal will serve as one of the conditions for triggering the transient fault diagnosis unit in Embodiment 1.

[0128] When the fault deviation index Greater than the second deviation threshold ,Right now When the system determines that the equipment is severely deviated and has a visible fault, it generates a serious fault alarm signal.

[0129] This embodiment sets... and Two thresholds for the fault deviation index after decoupling A refined classification was implemented, providing clear and actionable assessment results, namely normal stress, early warning, and severe alarm. This not only provides precise triggering conditions for subsequent transient fault diagnosis units but also provides clear input for maintenance risk decisions, realizing the transformation from raw deviation data to specific assessment levels.

[0130] Example 6:

[0131] The transient fault feature dataset includes:

[0132] Internal state parameters generated by the twin model management unit through high-fidelity model analysis under extreme impact conditions;

[0133] Real-time sensor measurements collected by the stress fault decoupling unit.

[0134] This embodiment describes the composition of the transient fault feature dataset described in Embodiments 1 and 5; this dataset The construction of the transient fault diagnosis unit is executed when specific triggering conditions are met, namely, when it is under extreme impact conditions and triggers early fault warning;

[0135] The transient fault feature dataset includes data generated by a twin model management unit under extreme shock conditions via a high-fidelity model. The generated internal state parameters are analyzed; these internal state parameters are deep physical quantities known to those skilled in the art but difficult to obtain directly by conventional sensors, such as compressor valve plate temperature. Mass flow rate of refrigerant at critical nodes These parameters are crucial for identifying subtle temporal characteristics of specific faults such as valve disc leakage; and real-time sensor measurements acquired by the stress fault decoupling unit; the vector of real-time sensor measurements. ,in Representing the Readings from a sensor, for example Represents the pressure at key measurement points. These represent the temperatures at key measurement points; combined with internal state parameters, they constitute multidimensional time-series data describing the transient behavior of the system.

[0136] This embodiment constructs a high-dimensional feature dataset that integrates internal deep states from a high-fidelity twin model and external measurable representations from sensors. Compared to relying solely on external sensor data, this dataset provides richer and more physically accurate fault information, enabling subsequent transient fault diagnosis models to... It can accurately identify the root cause of a malfunction even under high stress conditions, such as This significantly improves the depth and accuracy of diagnosis.

[0137] Example 7:

[0138] The specific process for maintaining the risk decision-making unit is as follows:

[0139] Obtain the largest fault deviation index among all key parameters and set it as the maximum fault deviation index.

[0140] Obtain the highest fault occurrence likelihood among all diagnosed faults and set it as the maximum fault occurrence likelihood.

[0141] The maximum failure deviation index and the maximum failure occurrence likelihood are weighted and summed according to the preset risk weight coefficients to obtain the comprehensive maintenance risk index.

[0142] This embodiment calculates the comprehensive maintenance risk index using the maintenance risk decision-making unit described in Embodiment 1. The process is refined; its technical motivation lies in combining the system's anomaly and fault confidence into a single risk indicator.

[0143] This unit obtains the largest fault deviation index among all key parameters and sets it as the maximum fault deviation index. ,Right now ; Originating from the stress fault decoupling unit, it represents the most severe deviation of the system from the healthy stress baseline.

[0144] This unit also acquires the output of the transient fault diagnosis unit; if this unit is not activated due to failure to meet the triggering conditions, or does not output a valid fault occurrence likelihood, then the maximum fault occurrence likelihood is set. If the unit is activated and outputs the likelihood, the highest fault occurrence likelihood among all diagnosed faults is obtained and set as the maximum fault occurrence likelihood. ,Right now ; The output, derived from the transient fault diagnosis unit, represents the highest confidence level regarding the most likely fault currently occurring in the system.

[0145] This unit will have the maximum fault deviation index. And the likelihood of the maximum failure occurrence Based on the preset risk weight coefficient By performing a weighted summation, we obtain the comprehensive maintenance risk index. ;

[0146] The technical motivation lies in the fact that this weighted sum model is the core of this decision-making unit, and its calculation formula is as follows: ;in, To comprehensively maintain the risk index, it is a dimensionless pure number, and its source is... Calculated; The preset risk weight coefficient is a dimensionless pure number. Its source or value is determined through expert experience or the Analytic Hierarchy Process (AHP) and is used to balance the severity of deviations. and the determinism of fault diagnosis Contribution to total risk;

[0147] This embodiment uses the maximum value, that is... , and weighted summation, i.e. The method of taking the maximum value ensures that the decision-making unit always makes judgments based on the worst-case scenario, i.e., the most severe deviation or the most credible failure, which conforms to the principle of safety redundancy; the weighted fusion unifies the two core indicators from different dimensions, namely deviation and likelihood, into a quantifiable and comparable comprehensive maintenance risk index. This provides a single and clear decision-making basis for the subsequent closed-loop control execution unit to perform hierarchical responses.

[0148] Example 8:

[0149] The specific execution logic of the closed-loop control execution unit is as follows:

[0150] When the comprehensive maintenance risk index is less than the preset first risk threshold, it is judged as a normal stress and a performance optimization strategy is executed.

[0151] When the comprehensive maintenance risk index is greater than or equal to the first risk threshold and less than the second risk threshold, it is judged as an early failure and the strategy of extending service life with defects is implemented.

[0152] When the comprehensive maintenance risk index is greater than or equal to the second risk threshold, it is judged as a serious fault, and a safety protection strategy is implemented.

[0153] This embodiment is a further refinement of embodiment 1 or 7, specifying the comprehensive maintenance risk index for the closed-loop control execution unit response. The execution logic of this unit; Value and preset first risk threshold Second risk threshold Based on the comparison results, hierarchical closed-loop control is executed;

[0154] Among them, the first risk threshold Second risk threshold The source or basis for the value is based on history Risk level classification of values This represents the boundary between normal operation and early-stage failure. This represents the boundary between early-stage and critical failures; specifically, the calibration process may include: acquiring a dataset containing historical data... A labeled dataset containing values ​​and their corresponding manually assigned risk levels was used to train a decision tree classifier. The optimal segmentation threshold for this classifier to distinguish between normal stress and early failure was then extracted as... Extract the optimal segmentation threshold for distinguishing between early and severe faults as... ;

[0155] The execution logic is as follows:

[0156] When the comprehensive maintenance risk index Less than the preset first risk threshold ,Right now At this time, the system determines it as a normal stress response; in this state, even The system determines the device is healthy and a high level of health is detected, at which point a performance optimization strategy is executed.

[0157] When the comprehensive maintenance risk index Greater than or equal to the first risk threshold And less than the second risk threshold ,Right now When the system determines the fault to be in its early stages, it executes a fault-extension strategy and simultaneously reports the fault type and duration. Values ​​are used to generate predictive maintenance work orders;

[0158] When the comprehensive maintenance risk index Greater than or equal to the second risk threshold ,Right now When this occurs, the system determines it to be a serious fault; at this time, the system immediately issues a shutdown alarm and executes safety protection strategies.

[0159] This embodiment defines risk quantification. A clear mapping relationship between the three strategies and specific actions; through and Two thresholds will determine the continuous risk index. It is transformed into three discrete and clear control levels: normal stress, early fault, and severe fault. This makes the execution logic of the closed-loop control clear and reliable, and realizes differentiated, automated and intelligent response to different risk levels, which constitutes the decision-making and execution exit of the entire system.

[0160] Example 9:

[0161] Performance optimization strategies include adjusting the condenser fan speed; strategies for extending the lifespan of a defective compressor include increasing the evaporator set temperature or limiting the compressor's maximum speed.

[0162] This embodiment is an example illustrating the performance optimization strategy and the extended lifespan strategy with defects in Embodiment 8;

[0163] The performance optimization strategy is applied when the system determines that the stress response is normal. The equipment is healthy but may be operating under high load conditions, i.e., high To adapt to high heat loads and improve energy efficiency, this strategy includes adjusting the condenser fan speed. For example, at high temperatures or high disturbances Below, appropriately increase To enhance heat dissipation efficiency;

[0164] The aforementioned strategy for extending lifespan despite defects is applied when the system is identified as having an early-stage failure. The technical objective is to extend the service life of faulty components until planned maintenance by proactively adjusting operating parameters, rather than immediately shutting down the system; this strategy includes increasing the evaporator set temperature. For example, actively raising the temperature by 1-2°C, or limiting the compressor's maximum speed. Both of these methods can effectively reduce the compressor's operating load and discharge pressure, thereby slowing down the rate of failure deterioration.

[0165] This embodiment provides specific and executable control methods for performance optimization and extending the lifespan of equipment with existing defects. The performance optimization strategy ensures that the equipment has the best energy efficiency in a healthy state. The strategy of extending the lifespan of equipment with existing defects reflects one of the core values ​​of this system, namely, after identifying early failures, shifting from passively waiting for damage to actively intervening, and maximizing the lifespan of the equipment by reducing the intensity of operation, which greatly improves the economic benefits and maintainability of the equipment.

[0166] Example 10:

[0167] Safety protection strategies include reducing the compressor speed to a safe minimum or executing an emergency shutdown procedure.

[0168] This embodiment is an example illustrating the security protection strategy in Embodiment 8;

[0169] The aforementioned security protection strategy is applied when the system determines a serious fault, i.e. At this point, the equipment is in a high-risk state, and immediate measures must be taken to prevent catastrophic damage or a safety accident; this strategy includes reducing the compressor speed. Reduce to the safe lower limit, or execute an emergency shutdown procedure; for example, the system will Reduce the speed to the minimum permissible speed to maintain necessary cooling, or immediately cut off the power to perform an emergency shutdown when extreme risks such as compressor stall are detected, and strictly limit the equipment from restarting before maintenance;

[0170] This embodiment provides explicit fallback safety measures for severe fault levels; by reducing the speed to the safe lower limit or performing an emergency shutdown, this strategy ensures that if an unacceptable risk, i.e., a high risk, is detected... When the value is set, the system can execute safety procedures with the highest priority, effectively preventing the escalation of equipment damage and the occurrence of secondary safety accidents, thus ensuring the safety of personnel and equipment.

[0171] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0172] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A digital twin-based intelligent maintenance system for refrigeration equipment, characterized in that, include: The operating condition sensing unit is used to collect multi-dimensional operating data of the refrigeration equipment, analyze the multi-dimensional operating data to obtain a transient disturbance index, and compare the transient disturbance index with a preset disturbance index threshold to obtain a fidelity level. The twin model management unit is used to dynamically construct a twin model in response to the fidelity level, and analyze the twin model to obtain internal state parameters. The stress fault decoupling unit is used to combine the transient disturbance index and a preset dynamic stress baseline model to calculate the dynamic stress baseline, and calculate the deviation between the dynamic stress baseline and the collected real-time sensor measurements to obtain a fault deviation index. Then, the fault deviation index is judged to obtain a decoupling evaluation result. The transient fault diagnosis unit is used to construct a transient fault feature dataset based on internal state parameters and real-time sensor measurements when the fidelity level is extreme impact conditions and the decoupling evaluation result is an early fault warning. The dataset is then sent to a preset transient fault diagnosis model for analysis to obtain the fault occurrence likelihood. The maintenance risk decision unit is used to obtain the fault deviation index and the fault occurrence likelihood, perform weighted fusion processing on the two to obtain the comprehensive maintenance risk index, and compare the comprehensive maintenance risk index with the preset risk threshold to obtain the risk level signal; the closed-loop control execution unit is used to respond to the risk level signal to execute the corresponding performance optimization strategy, fault-extension strategy or safety protection strategy.

2. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 1, characterized in that, The transient disturbance index analysis process is as follows: collect multi-dimensional operating data, including key measuring point pressure, key measuring point temperature, and door opening / closing status; calculate the time change rate of key measuring point pressure and key measuring point temperature, and normalize the time change rate according to a preset reference benchmark to obtain the normalized value of pressure change rate and the normalized value of temperature change rate. The normalized values ​​of pressure change rate, temperature change rate, and door opening / closing status are weighted and summed according to preset contribution weights to obtain the transient disturbance index.

3. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 1, characterized in that, The twin model management unit is also used to: when the fidelity level is steady-state, call the low-fidelity model to analyze all components; when the fidelity level is transient stress, call the preset hybrid fidelity model; when the fidelity level is extreme impact, call the high-fidelity model to analyze the preset key components, and call the low-fidelity model to analyze the preset non-key components.

4. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 1, characterized in that, The deviation calculation process of the stress fault decoupling unit is as follows: Based on the transient disturbance index, the dynamic stress baseline model is called to predict and generate the dynamic stress baseline; real-time sensor measurement values ​​are obtained; the absolute deviation between the real-time sensor measurement values ​​and the dynamic stress baseline is calculated, and the absolute deviation is divided by the dynamic stress baseline to obtain the fault deviation index.

5. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 4, characterized in that, The process of generating the decoupling evaluation result is as follows: the fault deviation index is compared and analyzed with a preset first deviation threshold and a preset second deviation threshold; when the fault deviation index is less than or equal to the first deviation threshold, a normal stress signal is generated. When the fault deviation index is greater than the first deviation threshold and less than or equal to the second deviation threshold, an early fault warning signal is generated. When the fault deviation index is greater than the second deviation threshold, a critical fault alarm signal is generated.

6. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 1, characterized in that, The transient fault feature dataset includes: internal state parameters generated by the twin model management unit through high-fidelity model analysis under extreme impact conditions; and real-time sensor measurements collected by the stress fault decoupling unit.

7. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 1, characterized in that, The specific process of the maintenance risk decision-making unit is as follows: obtain the largest fault deviation index among all key parameters and set it as the maximum fault deviation index; obtain the largest fault occurrence likelihood among all diagnosed faults and set it as the maximum fault occurrence likelihood. The maximum failure deviation index and the maximum failure occurrence likelihood are weighted and summed according to the preset risk weight coefficients to obtain the comprehensive maintenance risk index.

8. A digital twin-based intelligent maintenance system for refrigeration equipment according to claim 1 or 7, characterized in that, The specific execution logic of the closed-loop control execution unit is as follows: when the comprehensive maintenance risk index is less than the preset first risk threshold, it is determined to be a normal stress and the performance optimization strategy is executed; when the comprehensive maintenance risk index is greater than or equal to the first risk threshold and less than the second risk threshold, it is determined to be an early failure and the service life extension strategy is executed. When the comprehensive maintenance risk index is greater than or equal to the second risk threshold, it is judged as a serious fault, and a safety protection strategy is implemented.

9. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 8, characterized in that, The performance optimization strategies include adjusting the condenser fan speed; the life extension strategies for those with defects include increasing the evaporator set temperature or limiting the compressor's maximum speed.

10. The intelligent maintenance system for refrigeration equipment based on digital twins according to claim 8, characterized in that, The safety protection strategy includes reducing the compressor speed to a safe lower limit or executing an emergency shutdown procedure.

Images