A digital twin driven prognostic maintenance decision system for decommissioned wind turbines

Abstract

The present application relates to the technical field of wind turbine maintenance decision, and discloses a kind of digital twin driven retired wind turbine predictive maintenance decision system.The system includes operation data acquisition, component data analysis, digital twin model construction, adaptation compatibility evaluation and maintenance decision generation module.Core maintenance parameters and its priority sequence are formulated based on operation and component data;By fusing core and alternative component materials to construct a digital twin model prototype, analyze its signal detection characteristics;According to the durability test data of the model, the life prediction equivalent value is calculated, and the adaptation compatibility with the maintenance system is evaluated;Comprehensive screening of the best maintenance material and generating decisions.This method realizes joint simulation and compatibility quantitative evaluation of various material replacement schemes in a virtual environment, and can optimize signal sensing ability and system matching degree before actual maintenance, thereby improving the feasibility and decision reliability of the life extension operation and maintenance of retired units.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine maintenance decision, in particular to a digital twin driven retired wind turbine predictive maintenance decision system. BACKGROUND

[0002] In the aspect of data acquisition, the existing technology focuses more on the collection of conventional parameters during the operation of the wind turbine, and lacks a systematic and comprehensive acquisition mechanism for operating condition parameters and dynamic performance parameters during the retirement stage. Most maintenance schemes only rely on partial basic operating data and fail to develop targeted core maintenance parameters in combination with the aging and performance degradation of retired unit components, resulting in a lack of clear target orientation for subsequent maintenance work and often falling into the dilemma of "comprehensive overhaul but failing to grasp the key points".

[0003] In the component analysis link, the traditional method usually only focuses on the surface damage or a single performance indicator of the component, and fails to deeply schedule the intrinsic data and structural design data of the component. The physical performance representation corresponding to the intrinsic data of the component lacks systematic analysis, and the importance and maintenance urgency of different components in the retired unit cannot be accurately judged, making it difficult to form a scientific parameter priority sequence. This causes maintenance resources to be often allocated to non-critical components, resulting in waste of manpower and material resources, and also possibly ignoring the potential failure risk of core components.

[0004] Although the application of digital twin technology in industrial equipment maintenance has gradually increased, its application in the field of retired wind turbines is still not mature enough. The existing digital twin model construction is mostly based on material and structural data under the new state of the unit, without fully considering the aging loss of the core component materials of the retired unit, and rarely incorporating alternative component materials for comparison simulation. After the model is built, only the basic functions are simply verified, and there is a lack of systematic collection and analysis of sensing accuracy data and durability test data, which cannot accurately grasp the signal detection characteristics of the model, making it difficult for the model to truly reflect the actual operating state of the retired unit and provide reliable simulation support for maintenance decision-making.

[0005] In terms of adaptation compatibility evaluation, the current technology often ignores the adaptability between the digital twin model and the maintenance system. In most cases, only the performance indicators of the model are concerned, without calculating the life prediction equivalent value of the model combined with the durability test data, and without deeply analyzing the adaptation compatibility of the model in the actual maintenance system. This makes the constructed digital twin model may not match the existing maintenance process and equipment, cannot effectively integrate into the maintenance decision-making system, and is difficult to play its due role.

[0006] In the maintenance decision generation stage, the existing scheme relies on manual experience, and the combined parameters are relatively single. Often, only the damage degree or running time of the component is used to develop a maintenance strategy, without considering key factors such as signal detection characteristics, performance stability and adaptation compatibility of the digital twin model. At the same time, in the material selection, there is a lack of comprehensive comparison and screening of core component materials and alternative component materials, making it difficult to determine the most suitable maintenance material for the retired unit, resulting in the generated maintenance decision lacking scientificity and rationality, which may not only fail to completely solve the problems of the unit, but also increase the subsequent maintenance cost and safety hazards due to improper material selection. SUMMARY

[0007] The purpose of the present application is to provide a digital twin driven predictive maintenance decision system for retired wind turbine units to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application provides a digital twin driven predictive maintenance decision system for retired wind turbine units, which comprises:

[0009] An operating data acquisition module for acquiring operating condition parameters and dynamic performance parameters of the retired wind turbine unit and developing core maintenance parameters of the wind turbine unit;

[0010] A component data analysis module for scheduling component intrinsic data and structural design data of the wind turbine unit, analyzing physical performance characteristics corresponding to the component intrinsic data, and determining a parameter priority sequence of the core maintenance parameters;

[0011] A digital twin model construction module for querying core component materials and alternative component materials of the wind turbine unit, simulating the construction of the wind turbine unit based on the core component materials and the alternative component materials, obtaining a digital twin model prototype, collecting sensing accuracy data and durability test data of the digital twin model prototype, and analyzing signal detection characteristics of the digital twin model prototype based on the sensing accuracy data;

[0012] An adaptation compatibility evaluation module for calculating a life prediction equivalent value of the digital twin model prototype based on the durability test data, and analyzing the adaptation compatibility of the digital twin model prototype in the maintenance system;

[0013] A maintenance decision generation module for screening the best maintenance material from the core component materials and the alternative component materials based on the signal detection characteristics, the performance stability and the adaptation compatibility, and generating a maintenance decision for the wind turbine unit based on the core maintenance parameters, the parameter priority sequence and the best maintenance material.

[0014] Preferably, the operation data acquisition module determines a working environment factor of the wind turbine based on the operation condition parameters and the dynamic performance parameters, analyzes application scene features of the wind turbine based on the working environment factor, collects existing fault information of the wind turbine, and formulates core maintenance parameters of the wind turbine based on the application scene features and the existing fault information.

[0015] The operation data acquisition module formulates the core maintenance parameters of the wind turbine based on the application scene features and the existing fault information, including:

[0016] feature extraction is performed on the application scene features to obtain scene feature factors;

[0017] information classification processing is performed on the existing fault information to obtain a fault category set;

[0018] encoding processing is performed on the scene feature factors to obtain a scene feature encoding vector;

[0019] encoding processing is performed on the fault category set to obtain a fault category encoding vector;

[0020] cluster analysis is performed on the scene feature encoding vector to obtain a scene feature cluster center;

[0021] cluster analysis is performed on the fault category encoding vector to obtain a fault category cluster center;

[0022] an association relationship between the scene feature cluster center and the fault category cluster center is analyzed to obtain an association mapping matrix;

[0023] based on the association mapping matrix, key scene features and key fault categories of the wind turbine are determined;

[0024] based on the key scene features and the key fault categories, design constraint conditions of the wind turbine are analyzed;

[0025] based on the design constraint conditions, the core maintenance parameters of the wind turbine are formulated.

[0026] Preferably, the component data analysis module analyzes physical performance representations corresponding to the component intrinsic data, determines a signal propagation topology of the wind turbine according to the structural design data, evaluates a synergistic coupling effect between the signal propagation topology and the physical performance representations, and determines a parameter priority sequence of the core maintenance parameters according to the synergistic coupling effect.

[0027] The component data analysis module analyzes the physical performance representations corresponding to the component intrinsic data, including:

[0028] Standardize the component intrinsic data to obtain standardized component data;

[0029] Extract the component physical properties corresponding to the standardized component data, and perform screening processing on the component physical properties to obtain key physical properties;

[0030] Calculate the physical performance indicators corresponding to the key physical properties, and generate the physical performance representation corresponding to the component intrinsic data based on the physical performance indicators.

[0031] Preferably, the component data analysis module determines the signal propagation topology of the wind turbine according to the structure design data, comprising:

[0032] Perform data preprocessing on the structure design data to obtain target structure design data;

[0033] Extract the structure parameter set of the wind turbine from the target structure design data;

[0034] Perform material attribute association processing on the structure parameter set to obtain an attribute association parameter set;

[0035] Based on the attribute association parameter set, construct a signal propagation numerical model corresponding to the wind turbine;

[0036] Perform simulation calculation processing on the signal propagation numerical model to obtain signal propagation dynamic data;

[0037] Perform topology abstraction processing on the signal propagation dynamic data to generate a preliminary signal propagation topology;

[0038] Obtain the topology structure data of the wind turbine, and based on the topology structure data, perform spatial topology expression optimization on the preliminary signal propagation topology to obtain an optimized signal propagation topology.

[0039] Preferably, the component data analysis module evaluates the synergistic coupling effect between the signal propagation topology and the physical performance representation, comprising:

[0040] Extract the signal propagation features corresponding to the signal propagation topology, and perform dimensionality reduction processing on the signal propagation features to obtain reduced dimension signal propagation features;

[0041] Calculate the feature similarity index between the reduced dimension signal propagation features, and calculate the representation similarity index between the physical performance representations;

[0042] Calculate the correlation factor between the signal propagation topology and the physical performance representation;

[0043] Calculate the synergistic coupling degree between the signal propagation topology and the physical performance characterization based on the correlation factor, the characteristic similarity index, and the characterization similarity index.

[0044] Evaluate the synergistic coupling effect between the signal propagation topology and the physical performance characterization based on the synergistic coupling degree.

[0045] Preferably, the calculation of the correlation factor between the signal propagation topology and the physical performance characterization comprises:

[0046] Vectorize the signal propagation topology and the physical performance characterization respectively to obtain a propagation topology vector and a performance characterization vector;

[0047] Calculate the vector cosine value between the propagation topology vector and the performance characterization vector;

[0048] Calculate the vector mutual information between the propagation topology vector and the performance characterization vector;

[0049] Calculate the correlation factor between the signal propagation topology and the physical performance characterization by combining the vector cosine value and the vector mutual information through a weight adjustment coefficient.

[0050] Preferably, the digital twin model construction module analyzes the signal detection characteristics of the digital twin model prototype based on the sensing accuracy data, comprising:

[0051] Data cleaning processing is performed on the sensing accuracy data to obtain cleaned sensing accuracy data;

[0052] Analyze and extract the accuracy time domain features and accuracy frequency domain features corresponding to the cleaned sensing accuracy data;

[0053] Generate a signal characteristic descriptor corresponding to the digital twin model prototype based on the accuracy time domain features and the accuracy frequency domain features;

[0054] Analyze the signal detection characteristics of the digital twin model prototype based on the signal characteristic descriptor.

[0055] Preferably, the adaptation compatibility evaluation module calculates the life prediction equivalent value of the digital twin model prototype according to the durability test data, evaluates the performance stability of the digital twin model prototype based on the life prediction equivalent value, determines the installation location requirements and working condition constraints of the wind turbine in the maintenance system, collects the adaptability parameters of the digital twin model prototype, and analyzes the adaptation compatibility of the digital twin model prototype in the maintenance system by combining the installation location requirements, the working condition constraints, and the adaptability parameters.

[0056] The life prediction equivalent value of the digital twin model prototype is calculated according to the durability test data in the adaptation compatibility evaluation module, comprising:

[0057] The abnormal behavior response label group is called to obtain time data of the wind turbine, including fault start time, fault development time and fault recovery time, to calculate time intervals and generate fault key time interval data;

[0058] Based on the fault key time interval data, fault density distribution information is analyzed;

[0059] According to the fault density distribution information, a fault density offset characteristic value is calculated, the position interval of the fault density abnormal section on the time axis is identified, and a fault density offset time period is generated;

[0060] Based on the fault density offset time period, the correlation between fault development and time period is evaluated, the continuous segment with fault development behavior offset is screened, and a fault behavior delay segment is generated;

[0061] Based on the fault behavior delay segment, the life prediction equivalent value of the digital twin model prototype is calculated.

[0062] Preferably, the adaptation compatibility evaluation module combines the installation location requirement, the working condition constraint and the adaptability parameter to evaluate the adaptation compatibility of the digital twin model prototype in the maintenance system, comprising:

[0063] Based on the fault behavior delay segment, maintenance time data of the wind turbine is extracted to obtain maintenance lag information;

[0064] It is judged whether the maintenance lag value exceeds the maintenance completion threshold, the component number of which maintenance is not completed is screened, and the waiting time of the corresponding component is extracted, sorted according to the waiting time of the unfinished components, and a component maintenance waiting priority sequence is generated;

[0065] The sorting information in the component maintenance waiting priority sequence is called to sequentially configure additional maintenance time windows for the components, adjust the maintenance time length within the total maintenance time range, record the component number and the corresponding adjusted maintenance time, and generate a maintenance time configuration table;

[0066] Based on the maintenance time configuration table, the installation location requirement, the working condition constraint and the adaptability parameter are combined to analyze the adaptation compatibility of the digital twin model prototype in the maintenance system.

[0067] Preferably, the maintenance decision generation module generates the maintenance decision of the wind turbine, comprising:

[0068] The maintenance time configuration table is called to detect real-time wind turbine operation data and identify component maintenance time points;

[0069] determining whether the maintenance time point meets the maintenance condition, and generating a maintenance start instruction if the maintenance condition is met;

[0070] generating a maintenance decision for the wind turbine based on the maintenance start instruction.

[0071] Compared with the prior art, the present application has the following advantages:

[0072] The operation data acquisition module focuses on the acquisition of operation condition parameters and dynamic performance parameters of the retired wind turbine, can accurately capture performance change data of the retired unit components due to aging, wear and tear, etc., and the core maintenance parameters formulated on this basis closely meet the actual maintenance needs of the retired unit, avoiding the problem that the maintenance parameters are disconnected with the actual needs due to one-sided data acquisition in traditional technology, so that the maintenance work has clear pertinence and target from the beginning.

[0073] The component data analysis module can clearly understand the role, performance state and potential fault risk of different components in the overall operation of the retired unit by scheduling component intrinsic data and structural design data, and further determine a scientific and reasonable core maintenance parameter priority sequence. This process enables the maintenance resources to be reasonably allocated according to the importance and maintenance urgency of the components, avoiding the situation that resources are wasted or key components are ignored in traditional maintenance, and improving the efficiency and effectiveness of maintenance work.

[0074] The digital twin model construction module fully considers the actual situation of the retired wind turbine during model building, and includes both the core component materials and the alternative component materials in the simulation range, so that the digital twin model prototype built can more truly reflect the structural characteristics and material state of the retired unit. At the same time, by collecting the sensing accuracy data and durability test data of the model prototype, and analyzing the signal detection characteristics based on the sensing accuracy data, the staff can accurately understand the perception ability and accuracy of the model to the unit operation state in advance, ensuring that the model can provide reliable simulation basis for subsequent maintenance decision, solving the problem that the simulation result deviates greatly from the actual situation due to the lack of consideration of the characteristics of the retired unit in traditional digital twin model.

[0075] The adaptation compatibility evaluation module calculates the life prediction equivalent value of the digital twin model prototype according to the durability test data, and analyzes the adaptation compatibility of the model in the maintenance system, which can judge in advance whether the model is matched with the existing maintenance process and equipment, and the stability and applicability of the model in the long-term use. This link effectively avoids the situation that the model cannot be used after being built due to incompatibility with the maintenance system, ensuring that the digital twin model can be smoothly integrated into the maintenance decision system and fully play its role in state simulation, fault prediction, etc.

[0076] The maintenance decision generation module comprehensively considers factors such as signal detection characteristics, performance stability and adaptation compatibility, and performs comprehensive comparison and screening on core component materials and alternative component materials, and selects the best maintenance material most suitable for the decommissioned unit. On this basis, the maintenance decision generated in combination with the core maintenance parameters and the parameter priority sequence takes into account the pertinence of maintenance, the applicability of materials and the reliability of the model, breaks away from the limitations of traditional maintenance decisions relying on artificial experience and single parameters, makes the maintenance decision more scientific and reasonable, can effectively solve the problems existing in the decommissioned wind turbine, reduce the maintenance cost, and ensure the subsequent safe and stable operation of the unit. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 a timing diagram of the digital twin driven decommissioned wind turbine predictive maintenance decision system described in the present application;

[0078] Figure 2 a working principle flowchart of the wind turbine core maintenance parameter for the operation data acquisition module;

[0079] Figure 3 a working principle flowchart of the core maintenance parameter priority sequence for the component data analysis module. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0081] Please refer to Figure 1 The present application provides a digital twin driven decommissioned wind turbine predictive maintenance decision system, which comprises an operation data acquisition module, a component data analysis module, a digital twin model construction module, an adaptation compatibility evaluation module and a maintenance decision generation module.

[0082] The operation data acquisition module collects unit operating condition parameters and dynamic performance parameters, and formulates core maintenance parameters accordingly. The component data analysis module schedules component intrinsic data and structural design data, analyzes physical performance characteristics, and determines the priority sequence of core maintenance parameters. The digital twin model construction module queries core component materials and alternative component materials, generates a digital twin model prototype through simulation, and analyzes its signal detection characteristics based on sensing accuracy data. The compatibility evaluation module calculates the life prediction equivalent value based on durability test data and evaluates the model's compatibility with the maintenance system. The maintenance decision generation module comprehensively considers signal detection characteristics, performance stability, and compatibility to select the best maintenance materials, and generates the final maintenance decision by combining the core maintenance parameters and their priority sequence.

[0083] Example 1: See Figure 2 The module collects operating parameters and dynamic performance parameters of decommissioned wind turbines. Operating parameters include real-time monitoring data such as rotor speed, gearbox oil temperature, generator winding temperature, and nacelle vibration amplitude. Dynamic performance parameters cover performance indicators such as power output curves, energy conversion efficiency, and load response characteristics. This data is continuously collected and transmitted through a sensor network installed in various parts of the turbine. Based on the acquired operating parameters and dynamic performance parameters, the module further analyzes the wind turbine's operating environment factors. These factors include external factors such as atmospheric temperature change cycles, relative humidity distribution, wind speed fluctuation frequency, and salt spray corrosion concentration. These factors are obtained through data fusion analysis of environmental monitoring stations and the turbine's built-in sensors.

[0084] Based on a clear understanding of the working environment factors, the module conducts an in-depth analysis of the application scenario characteristics of wind turbines. These characteristics include specific climate patterns resulting from geographical location, such as the high salinity and high humidity environment of coastal areas, or the low temperature and strong wind conditions of plateau regions. It also considers the material aging and performance degradation trends caused by years of operation. These characteristics are extracted through long-term data accumulation and pattern recognition technology. The module simultaneously collects existing fault information of the wind turbines. This information comes from fault diagnosis reports in historical maintenance records, abnormal alarm logs from the real-time monitoring system, and lists of potential defects discovered during regular inspections. Common fault types include blade surface corrosion, increased bearing clearance, decreased insulation resistance, and control system signal drift.

[0085] In combination with the application scene characteristics and existing fault information, the module starts the core maintenance parameter formulation process. The process first extracts the scene characteristic factors from the application scene characteristics. The feature extraction uses principal component analysis method to reduce the dimensionality of the key influencing factors such as corrosion acceleration factor, fatigue load cycle number and thermal stress fluctuation coefficient from multi-dimensional environmental data. The existing fault information is classified and processed to obtain a fault category set. The classification processing uses hierarchical clustering algorithm to divide the faults into electrical system faults, mechanical structure faults and control system faults according to the fault mechanism and influence degree, and each category is further subdivided into specific fault modes such as cable joint oxidation, gear tooth surface pitting, sensor precision failure, etc.

[0086] The scene characteristic factors are encoded to obtain a scene characteristic encoding vector. The encoding process uses one-hot encoding technology to convert discrete feature factors into binary vectors, and at the same time, combined with normalization processing, the continuous feature factors are scaled to a standard numerical range to realize unified data expression. The fault category set is encoded to obtain a fault category encoding vector. The encoding process introduces a fault severity weight coefficient, which is calculated according to the downtime and maintenance cost caused by the fault, so that the encoding vector can not only reflect the fault type but also reflect the criticality of the fault. The scene characteristic encoding vector is clustered to obtain a scene characteristic clustering center. The clustering analysis uses K-means algorithm to aggregate the feature vectors under similar environmental conditions into clusters, and the center point of each cluster represents a typical environmental scene such as high temperature and high humidity cluster or low temperature and high load cluster. The fault category encoding vector is clustered to obtain a fault category clustering center. The clustering process is iteratively calculated according to the fault frequency and maintenance urgency, and finally forms a clustering group with high-frequency faults and high-risk faults as the core.

[0087] The correlation mapping matrix is obtained by analyzing the correlation between the scene feature cluster centers and the fault category cluster centers. The correlation analysis uses the canonical correlation analysis method to calculate the statistical correlation between the environmental scene and the fault type, and the matrix element value reflects the probability intensity of the occurrence of a specific fault under a specific environmental condition. Based on the correlation mapping matrix, the key scene features and key fault categories of the wind turbine are determined. The key scene features focus on environmental factors that are significantly correlated with high-frequency faults, such as the combination of salt mist concentration and vibration intensity. The key fault categories are selected as the fault modes that are most affected by environmental factors, such as bearing corrosion and insulation breakdown. Based on the key scene features and key fault categories, the design constraints of the wind turbine are analyzed. The design constraints include material selection that must meet the requirements of salt mist corrosion resistance, structural design that needs to withstand specific vibration spectrum loads, and electrical system that should have humidity protection capability. Based on the design constraints, the core maintenance parameters of the wind turbine are developed. The core maintenance parameters are ultimately determined as quantifiable monitoring indicators such as vibration acceleration threshold, insulation resistance lower limit, lubricating oil acidity index, and blade surface roughness grade. These parameters will serve as the direct basis for predictive maintenance decisions.

[0088] Example 2: Referring to Figure 3 The processing component intrinsic data, which is derived from the original attribute records of each component of the wind turbine, including material composition certificates, mechanical performance test reports, and multi-dimensional information such as service history parameters. Standardization processing uses Z-score algorithm combined with range standardization method. Z-score algorithm converts data to a distribution with mean zero and standard deviation one, and range standardization linearly maps data to the interval of zero to one. This double processing ensures the comparability of parameters with different dimensions and orders of magnitude, such as unifying the units of GPa for Young's modulus and ppm / ℃ for thermal expansion coefficient to dimensionless values. When extracting the standardized component data corresponding to the component physical properties, the module calls the physical property database for matching and mapping. The physical properties include the intrinsic characteristics of the material such as density, yield strength, fatigue limit, thermal conductivity, resistivity, etc. These properties are determined by both material science theory definition and experimental measurement data.

[0089] The key physical properties are obtained by screening the physical properties of the components. The screening process is based on the correlation analysis of the properties and the typical failure modes of the wind turbine. The Pearson correlation coefficient is used to calculate the correlation strength of each property with the historical failure records. For example, the correlation between vibration failure and material damping coefficient, the correlation between overheating failure and thermal conductivity. The properties with correlation coefficients exceeding the threshold value are retained as the key physical properties. When calculating the physical performance indicators corresponding to the key physical properties, the module defines the indicator calculation rules based on the operating characteristics of the wind turbine. For example, the fatigue life index is calculated through the stress-life curve and the load spectrum. The thermal efficiency coefficient considers the material thermal conductivity and the component geometry. These indicators are calculated by combining physical models and empirical formulas. Based on the physical performance indicators, the physical performance representation corresponding to the component intrinsic data is generated. The physical performance representation is constructed as a multi-dimensional vector. Each dimension represents the value of a performance indicator. The similarity between vectors is measured by the Euclidean distance, which is used for subsequent collaborative coupling analysis.

[0090] The signal propagation topology of the wind turbine is determined based on the structural design data, including CAD drawings, finite element mesh models, and assembly relationship tables. The target structural design data is obtained by preprocessing the structural design data. The preprocessing includes data denoising using median filtering to eliminate measurement errors, and missing value interpolation using K-nearest neighbor algorithm to fill incomplete fields, ensuring the continuity and consistency of the data. The structural parameter set of the wind turbine is extracted from the target structural design data. The structural parameter set includes geometric dimensions such as blade length, shaft diameter, gear tooth number, bolt distribution coordinates, as well as the type of connecting nodes and constraint conditions. These parameters are automatically identified from the three-dimensional model through feature extraction algorithms.

[0091] The attribute association parameter set is obtained by associating the structural parameter set with the material properties. The association process binds the geometric parameters of each component with its material properties by querying the material database. For example, the modulus of the gear is associated with the elastic modulus of the alloy steel, and the size of the bearing is associated with the hardness of the ceramic ball, forming a parameter set containing both geometric and material attributes. Based on the attribute association parameter set, the signal propagation numerical model corresponding to the wind turbine is constructed. The numerical model uses the finite element method to establish multi-physical field coupling equations, simulating the transmission path of mechanical vibration waves in the structure, the diffusion process of heat flow between components, and the transmission characteristics of electromagnetic signals in the circuit. The model mesh is divided using adaptive refinement technology to ensure the calculation accuracy of key areas.

[0092] The signal propagation dynamic data is obtained by simulating and calculating the signal propagation numerical model. The simulation and calculation includes transient analysis for simulating the signal response under impact load, harmonic response analysis for calculating the steady propagation under periodic load, and random vibration analysis for evaluating the signal characteristics under uncertain load. The output data includes displacement contour, stress distribution and temperature gradient of each node. The preliminary signal propagation topology is generated by topological abstraction of the signal propagation dynamic data. The topological abstraction adopts graph theory algorithm to abstract components as nodes and signal transmission paths as edges. The node weight represents the importance of components, and the edge weight represents the signal transmission efficiency, forming a weighted directed graph structure. The topology structure data of the wind turbine is obtained, including the hierarchical structure in the assembly relationship diagram and the causal chain in the signal flow diagram. These data are extracted from the control system logic diagram and the sensor layout diagram. Based on the topology structure data, the preliminary signal propagation topology is optimized by spatial topological expression to obtain the optimized signal propagation topology. The optimization process adopts Delaunay triangulation algorithm to reconstruct the node connection relationship to ensure the connectivity and robustness of the network. Meanwhile, virtual nodes are introduced to handle the cross-component signal jump phenomenon, and finally the topology network model accurately reflecting the internal signal transmission rule of the wind turbine is generated.

[0093] In embodiment 3, key features are extracted from the constructed signal propagation topology. The signal propagation features include node degree distribution, path propagation efficiency, signal attenuation coefficient and network clustering coefficient, which are indices describing the topology structure and are calculated by graph theory algorithm. The signal propagation features are processed by dimension reduction to obtain reduced signal propagation features. The principal component analysis method is used to retain the principal components with the highest contribution rate, thereby reducing the data dimension while retaining most of the original information. The feature vectors after dimension reduction are easier for subsequent similarity calculation. The feature similarity index between the reduced signal propagation features is calculated. The feature similarity index measures the directional consistency between different feature vectors by improved cosine similarity algorithm. The algorithm introduces a weight factor to adjust the contribution of different feature dimensions, so that the key features have higher discriminant weight. The representation similarity index between the physical performance representations is calculated. The similarity of the physical performance representations as multi-dimensional vectors is quantified by the reciprocal of the standardized Euclidean distance. The normalization of different physical dimensions is considered in the distance calculation to ensure the fairness of distance measurement. The correlation factor between the signal propagation topology and the physical performance representation is calculated. This step first vectorizes the signal propagation topology and the physical performance representation. The graph embedding technology is used to map the topology structure into a low-dimensional vector, and the physical performance representation is converted into a fixed-dimensional vector expression by eigenvalue decomposition. Finally, the propagation topology vector and the performance representation vector are obtained.

[0094] The vector cosine value between the propagation topology vector and the performance characterization vector is calculated, and the cosine value calculation adopts the dot product formula divided by the product of the module length, reflecting the similarity of the two vectors in the direction. The vector mutual information between the propagation topology vector and the performance characterization vector is calculated, and the mutual information calculation is based on the KL divergence of the joint probability distribution and the marginal probability distribution of the two vectors, measuring the statistical dependence between the vectors. In combination with the vector cosine value and the vector mutual information, the correlation factor is calculated through the weight adjustment coefficient, and the weight adjustment coefficient is dynamically allocated according to the feature importance, and the importance is determined by the random forest feature importance evaluation algorithm. The calculation formula of the correlation factor is:

[0095]

[0096] Among them: represents the correlation factor, represents the vector cosine value, represents the vector mutual information, and are the weight coefficients of the cosine value and the mutual information, respectively, satisfying the constraint condition . The weight coefficients are determined by the grid search cross-validation method to ensure that the correlation factor can balance the direction similarity and statistical dependence.

[0097] The synergistic coupling degree between the signal propagation topology and the physical performance characterization is calculated in combination with the correlation factor, the feature similarity index and the characterization similarity index, and the calculation process integrates the three indexes by using the weighted arithmetic average formula, and the weight distribution is dynamically adjusted based on the correlation strength of each index and the system performance. The synergistic coupling effect between the signal propagation topology and the physical performance characterization is evaluated based on the synergistic coupling degree, and the synergistic coupling degree value range is mapped to the range of zero to one, and the closer the value is to one, the stronger the synergy between the topology structure and the physical performance. The system provides an objective basis for subsequent maintenance decisions through quantitative evaluation.

[0098] In the process of processing sensing accuracy data, the digital twin model construction module first collects original signals from the sensor network installed in the key parts of the wind turbine, including vibration accelerometers, temperature sensors, strain gauges and current transformers, etc. The collected data includes timestamp, measurement value, sensor ID and state code, etc. The sensing accuracy data is cleaned to obtain cleaned sensing accuracy data. The cleaning process uses a sliding window anomaly detection algorithm to identify and eliminate abnormal values that are obviously outside the physical range, such as data points with a rotational speed exceeding 300% of the rated value or a temperature instantaneous jump of more than 100°C. At the same time, Kalman filtering algorithm is applied to smooth the data to eliminate random noise interference.

[0099] The precision time domain features and precision frequency domain features corresponding to the precision data of the extraction cleaning induction are analyzed. The time domain feature calculation includes mean, variance, peak-to-peak value, kurtosis index, and waveform factor, etc. These features reflect the amplitude distribution and fluctuation characteristics of the signal in the time dimension. The frequency domain feature extraction converts the signal to the frequency domain through fast Fourier transform, and calculates the power spectral density, center of gravity frequency, frequency variance, and harmonic distortion, etc. These features reveal the energy distribution law of the signal in the frequency dimension. The signal characteristic descriptor corresponding to the digital twin model prototype is generated based on the precision time domain features and precision frequency domain features. The descriptor construction uses feature fusion technology to combine the time domain and frequency domain features into a high-dimensional feature vector, and normalizes each dimension to eliminate dimension differences. Finally, a digital description that can fully characterize the signal characteristics is formed. Based on the signal characteristic descriptor, the signal detection characteristics of the digital twin model prototype are analyzed. The analysis process uses pattern recognition algorithm to match and compare the current signal characteristics with the standard signal template. The evaluation indexes include sensitivity resolution reflecting the ability of the system to detect weak signals, linearity error measuring the deviation degree of the input-output relationship, dynamic range representing the maximum span of the detectable signal amplitude, and repeatability error evaluating the consistency of multiple measurement results. These characteristics together constitute the signal detection capability atlas of the digital twin model.

[0100] The durability test data is used to calculate the life prediction equivalent value of the digital twin model prototype by the adaptation compatibility evaluation module. The abnormal behavior response label group is called to define the severity level and influence range of the fault. The label group includes fault code, affected components, and emergency level fields. The time data of the wind turbine generator is obtained, including fault start time, fault development time, and fault recovery time. These time stamps are extracted from the historical event log of the unit monitoring system. The time interval is calculated to generate fault key time interval data. The time interval reflects the whole life cycle of the fault from occurrence to repair. Based on the fault key time interval data, the fault density distribution information is analyzed. The analysis uses the kernel density estimation method to construct the probability density function of fault occurrence on the time axis. The density function curve shows the concentration degree of the fault in a certain time period. According to the fault density distribution information, the fault density offset characteristic value is calculated. The offset characteristic value is obtained by comparing the difference between the actual density distribution and the reference distribution. The position interval of the fault density anomaly segment on the time axis is identified to generate the fault density offset time segment. These time segments correspond to the dangerous period of fault anomaly concentration.

[0101] The correlation between fault development and time section is evaluated based on the fault density offset time period. The correlation coefficient between different time sections and fault development is calculated using the grey correlation analysis method. The continuous segment with fault development behavior offset is screened to generate the fault behavior delay section. The delay section represents the time lag phenomenon between fault occurrence and actual performance. The life prediction equivalent value of the digital twin model prototype is calculated based on the fault behavior delay section. The Weibull distribution model is used to fit the fault interval time data in the calculation process. The probability distribution of the remaining life is estimated by the shape parameter and the scale parameter. The median of the distribution is taken as the prediction reference value. Table 1 shows the fault time interval data.

[0102] Table 1: Fault time interval and density distribution data of wind turbine

[0103] Fault number Time of fault initiation Time of fault development Time of fault recovery Time interval (hours) Density value F2023001 2023-01-1208:30 2023-01-1214:45 2023-01-1309:20 24.83 0.087 F2023002 2023-02-0516:20 2023-02-0608:10 2023-02-0711:40 43.33 0.124 F2023003 2023-03-1810:15 2023-03-1909:30 2023-03-2014:00 51.75 0.156 F2023004 2023-04-2213:40 2023-04-2307:55 2023-04-2408:30 42.83 0.118 F2023005 2023-05-3009:00 2023-05-3110:20 2023-06-0116:45 55.75 0.192 F2023006 2023-07-1214:50 2023-07-1308:30 2023-07-1410:15 43.42 0.126 F2023007 2023-08-0811:30 2023-08-0909:45 2023-08-1012:20 48.83 0.142 F2023008 2023-09-1908:20 2023-09-2010:10 2023-09-2115:30 55.17 0.185 F2023009 2023-10-2515:40 2023-10-2608:50 2023-10-2709:30 41.83 0.115 F2023010 2023-11-1410:10 2023-11-1509:30 2023-11-1614:45 52.58 0.162

[0104] Through multi-source data fusion and time series analysis, a complete evaluation chain from signal detection to life prediction is established, providing a data basis for the adaptability evaluation of the digital twin model in the maintenance system. The fault density value is calculated using Gaussian kernel function for density estimation. The bandwidth parameter is adaptively determined according to the standard deviation of the time interval data. The density value reflects the aggregation degree of faults on the time axis. Higher density value indicates that the fault occurrence frequency in this time period is significantly higher than the average level. The life prediction equivalent value takes into account multiple factors such as fault occurrence frequency, duration and repair effect, forming a quantitative evaluation of the durability performance of the digital twin model.

[0105] In the implementation example 5, when the adaptation compatibility evaluation module processes the system adaptability of the digital twin model prototype, the maintenance time data of the wind turbine is first extracted based on the fault behavior delay section. The maintenance time data comes from the unit maintenance log database, including the planned start time, actual start time, completion time and delay reason record of historical maintenance work order. The maintenance lag information is calculated through these timestamp data. The maintenance lag information specifically represents the difference sequence between the planned maintenance time and the actual execution time, as well as the average delay time distribution of different component categories. It is judged whether the maintenance lag value exceeds the maintenance completion benchmark threshold. The maintenance completion benchmark threshold is set according to the wind farm operation and maintenance specification. It usually requires that the non-critical component delay does not exceed 48 hours, and the critical component delay does not exceed 24 hours. The system automatically compares the actual delay value of each component with the threshold value, screens out the component numbers whose maintenance is not completed, and extracts the waiting time of the corresponding components. The waiting time is calculated by subtracting the planned completion time from the current system time.

[0106] The component maintenance waiting priority sequence is generated by sorting the waiting time of the unfinished components. The sorting algorithm uses a multi-factor weighted evaluation, considering not only the absolute value of the waiting time, but also the component criticality coefficient and the failure risk index. The criticality coefficient is divided according to the impact of component failure on the whole machine, and the failure risk index is calculated by the historical failure frequency and severity. Finally, a priority list is generated in descending order of maintenance urgency. The sorting information in the component maintenance waiting priority sequence is called to configure additional maintenance time windows for the components. The configuration process uses a dynamic programming algorithm to seek the optimal allocation scheme within the total maintenance time range, ensuring that high-priority components have sufficient time resources while minimizing the impact on low-priority components. The maintenance time length is adjusted within the total maintenance time range, and the adjustment strategy includes extending the single maintenance time, increasing the maintenance frequency, or arranging an emergency maintenance window. A maintenance time configuration table is generated by recording the component number and the corresponding adjusted maintenance time, which includes fields such as component ID, original planned time, new allocated time window, adjustment reason, and expected completion time.

[0107] Based on the maintenance time configuration table, the compatibility of the digital twin model prototype in the maintenance system is analyzed in combination with the installation location requirements, work condition constraints, and adaptability parameters. The installation location requirements include size limitations of physical installation space, interface type matching requirements, and safety distance regulations of adjacent components, which are extracted from the unit design drawings and installation specifications. The work condition constraints involve environmental temperature range, maximum allowed vibration amplitude, electromagnetic compatibility indicators, and other operating boundary conditions, which are obtained from the unit technical manual. The adaptability parameters include communication protocol version, data sampling rate matching degree, signal accuracy error range, and other digital interface parameters between the digital twin model and the physical system. The compatibility analysis uses a multi-objective optimization algorithm to calculate the weighted adaptation index. The algorithm takes installation matching degree, work condition compliance, and parameter consistency as three optimization objectives, and obtains the optimal compatibility score through Pareto frontier solution. The score result is mapped to the range of 0-100, and the higher the score, the better the integration adaptability of the digital twin model in the maintenance system.

[0108] The maintenance decision generation module calls the maintenance time configuration table after starting, detects real-time wind turbine operation data, and the operation data is collected in real time through the SCADA system, including key monitoring parameters such as unit power output, bearing temperature, gear box oil pressure, vibration spectrum, etc. The system continuously compares the deviation of the current data from the normal operation range, identifies the component maintenance time point, and the time point is determined not only according to the planned arrangement of the maintenance time configuration table, but also introduces the real-time equipment state evaluation result. When the monitoring parameter exceeds the early warning threshold, the maintenance time point is automatically advanced. Determine whether the maintenance time point meets the maintenance condition, which includes that the equipment must be in a shutdown state, the environmental wind speed is lower than the safe operation limit, the on-site maintenance resource availability and personnel qualification matching, etc. Comprehensive factors, the system obtains the wind farm meteorological data, maintenance team location information and tool equipment state through the Internet of Things platform in real time, and verifies the multi-dimensional condition compliance.

[0109] If the maintenance condition is met, a maintenance start instruction is generated. The instruction generation process adopts a combination of rule engine and machine learning. The rule engine generates a basic instruction framework based on predefined safety procedures and operation processes, and the machine learning model optimizes the instruction details according to historical maintenance success cases and real-time environmental variables. The final maintenance decision includes specific maintenance operation steps, required tool list, safety precautions and expected completion time node. The maintenance decision is automatically issued to the mobile terminal of the on-site maintenance personnel in the form of a work order, and is simultaneously updated to the digital twin platform of the wind farm management center, realizing the whole-process visual monitoring and dynamic adjustment of the maintenance process.

[0110] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin-driven predictive maintenance decision-making system for decommissioned wind turbines, characterized in that, The system includes: The operation data acquisition module is used to acquire the operating condition parameters and dynamic performance parameters of decommissioned wind turbines and to formulate the core maintenance parameters of wind turbines. The component data analysis module is used to schedule the intrinsic data of wind turbine components and structural design data, analyze the physical performance characteristics corresponding to the intrinsic data of the components, and determine the parameter priority sequence of the core maintenance parameters. The digital twin model construction module is used to query the core component materials and alternative component materials of the wind turbine, combine the core component materials and alternative component materials to simulate and build the wind turbine to obtain a digital twin model prototype, collect the sensing accuracy data and durability test data of the digital twin model prototype, and analyze the signal detection characteristics of the digital twin model prototype based on the sensing accuracy data. The compatibility evaluation module is used to calculate the lifetime prediction equivalent value of the digital twin model prototype based on the durability test data, and analyze the compatibility of the digital twin model prototype in the maintenance system. The maintenance decision generation module is used to combine the signal detection characteristics, performance stability and compatibility to select the best maintenance material from the core component materials and the alternative component materials, and generate the maintenance decision of the wind turbine based on the core maintenance parameters, the parameter priority sequence and the best maintenance material. The compatibility evaluation module calculates the lifetime prediction equivalent value of the digital twin model prototype based on the durability test data, evaluates the performance stability of the digital twin model prototype based on the lifetime prediction equivalent value, determines the installation location requirements and working condition constraints of the wind turbine in the maintenance system, collects the compatibility parameters of the digital twin model prototype, and analyzes the compatibility of the digital twin model prototype in the maintenance system by combining the installation location requirements, the working condition constraints and the compatibility parameters. The compatibility evaluation module calculates the lifetime prediction equivalent value of the digital twin model prototype based on the durability test data, including: Call the abnormal behavior response tag group to obtain the time data of the wind turbine, including the fault start time, fault development time and fault recovery time, calculate the time interval, and generate fault key time interval data; Based on the data from the critical time intervals of the fault, analyze the fault density distribution information; Based on the fault density distribution information, calculate the fault density offset feature value, identify the location interval of the fault density abnormal segment on the time axis, and generate the fault density offset time period. Based on the fault density offset time period, the correlation between fault development and time segment is evaluated, continuous segments with offset fault development behavior are screened, and fault behavior delay segments are generated. Based on the fault behavior delay segment, the lifetime prediction equivalent value of the digital twin model prototype is calculated.

2. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 1, characterized in that, The operation data acquisition module determines the working environment elements of the wind turbine based on the operating condition parameters and the dynamic performance parameters. Based on the working environment elements, it analyzes the application scenario characteristics of the wind turbine, collects the existing fault information of the wind turbine, and formulates the core maintenance parameters of the wind turbine by combining the application scenario characteristics and the existing fault information. The operational data acquisition module combines the application scenario characteristics and existing fault information to formulate the core maintenance parameters of the wind turbine, including: Feature extraction is performed on the application scenario features to obtain scenario feature factors; The existing fault information is classified to obtain a set of fault categories; The scene feature factors are encoded to obtain a scene feature encoding vector; The fault category set is encoded to obtain a fault category encoding vector; Cluster analysis is performed on the scene feature encoding vectors to obtain scene feature cluster centers; Cluster analysis is performed on the fault category encoding vectors to obtain the fault category cluster centers; The correlation between the cluster centers of the scene features and the cluster centers of the fault categories is analyzed to obtain the correlation mapping matrix; Based on the correlation mapping matrix, the key scenario characteristics and key fault categories of the wind turbine are determined; Based on the key scenario characteristics and key fault categories, the design constraints of the wind turbine are analyzed. Based on the aforementioned design constraints, the core maintenance parameters for the wind turbine are determined.

3. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 1, characterized in that, The component data analysis module analyzes the physical performance characterization corresponding to the intrinsic data of the component, determines the signal propagation topology of the wind turbine based on the structural design data, evaluates the synergistic coupling effect between the signal propagation topology and the physical performance characterization, and determines the parameter priority sequence of the core maintenance parameters based on the synergistic coupling effect. The component data analysis module analyzes the physical performance characterization corresponding to the intrinsic data of the component, including: The intrinsic data of the component are standardized to obtain standardized component data; Extract the physical attributes of the components corresponding to the standardized component data, and filter the physical attributes to obtain key physical attributes; Calculate the physical performance index corresponding to the key physical attributes, and generate the physical performance characterization corresponding to the intrinsic data of the component based on the physical performance index.

4. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 3, characterized in that, The component data analysis module determines the signal propagation topology of the wind turbine based on the structural design data, including: The structural design data is preprocessed to obtain the target structural design data; Extract the set of structural parameters of the wind turbine from the target structural design data; The structural parameter set is subjected to material property association processing to obtain a property association parameter set; Based on the attribute association parameter set, construct the signal propagation numerical model corresponding to the wind turbine; The signal propagation numerical model is simulated and processed to obtain dynamic data of signal propagation. The signal propagation dynamic data is subjected to topological abstraction processing to generate a preliminary signal propagation topology; The topology data of the wind turbine is acquired, and the spatial topology representation of the preliminary signal propagation topology is optimized based on the topology data to obtain the optimized signal propagation topology.

5. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 4, characterized in that, The component data analysis module evaluates the synergistic coupling effect between the signal propagation topology and the physical performance characterization, including: Extract the signal propagation features corresponding to the signal propagation topology, and perform dimensionality reduction processing on the signal propagation features to obtain dimensionality-reduced signal propagation features; Calculate the feature similarity index between the reduced-dimensional signal propagation features, and calculate the representation similarity index between the physical performance representations; Calculate the correlation factor between the signal propagation topology and the physical performance characterization; By combining the correlation factor, the feature similarity index, and the representation similarity index, the degree of synergistic coupling between the signal propagation topology and the physical performance representation is calculated. Based on the aforementioned cooperative coupling degree, the cooperative coupling effect between the signal propagation topology and the physical performance characterization is evaluated.

6. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 5, characterized in that, The calculation of the correlation factor between the signal propagation topology and the physical performance characterization includes: The signal propagation topology and the physical performance representation are vectorized respectively to obtain the propagation topology vector and the performance representation vector. Calculate the vector cosine between the propagation topology vector and the performance characterization vector; Calculate the vector mutual information between the propagation topology vector and the performance representation vector; By combining the vector cosine value and the vector mutual information, the correlation factor between the signal propagation topology and the physical performance characterization is calculated using a weight adjustment coefficient.

7. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 1, characterized in that, The digital twin model construction module analyzes the signal detection characteristics of the digital twin model prototype based on the sensing accuracy data, including: The sensing accuracy data is cleaned to obtain cleaned sensing accuracy data. Analyze and extract the time-domain and frequency-domain features of the cleaning sensing accuracy data; Based on the precision time-domain features and the precision frequency-domain features, a signal characteristic descriptor corresponding to the digital twin model prototype is generated; Based on the signal characteristic descriptor, the signal detection characteristics of the digital twin model prototype are analyzed.

8. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 7, characterized in that, The compatibility evaluation module analyzes the compatibility of the digital twin prototype in the maintenance system by combining the installation location requirements, the working condition constraints, and the compatibility parameters, including: Based on the fault behavior delay segment, the maintenance time data of the wind turbine is extracted to obtain maintenance lag information; Determine whether the maintenance lag value exceeds the maintenance completion baseline threshold, filter the component numbers that have not been maintained, extract the waiting time of the corresponding components, sort them according to the waiting time of the incomplete components, and generate a component maintenance waiting priority sequence; Call the sorting information in the component maintenance waiting priority sequence, configure additional maintenance time windows for the components in sequence, adjust the maintenance time length within the total maintenance time range, record the component number and the corresponding adjusted maintenance time, and generate a maintenance time configuration table; Based on the maintenance time configuration table, combined with the installation location requirements, the working condition constraints, and the adaptability parameters, the compatibility of the digital twin model prototype in the maintenance system is analyzed.

9. The predictive maintenance decision-making system for decommissioned wind turbines driven by digital twins as described in claim 1, characterized in that, The maintenance decision generation module generates maintenance decisions for the wind turbine, including: Call the maintenance time configuration table, detect real-time wind turbine operation data, and mark component maintenance time points; Determine whether the maintenance time point meets the maintenance conditions; if so, generate a maintenance start command. Based on the maintenance initiation command, a maintenance decision for the wind turbine is generated.

Images