Data and knowledge double-driven wheel set multi-parameter comprehensive state evaluation method and system

Abstract

The application discloses a data and knowledge double-driven wheel set multi-parameter comprehensive state evaluation method and system, and belongs to the technical field of railway vehicle maintenance. The method comprises the following steps: firstly, a mapping relationship between geometric parameters and dynamic performance indexes is constructed through sampling and dynamic simulation, and global sensitivity analysis is performed to screen key dynamic performance indexes; secondly, the subjective and objective weights of the indexes are determined in combination with an analytic hierarchy process and an entropy weight method, and a dynamic optimization model based on Bellman equation is introduced to generate final combination weights; then, a multi-dimensional state space division model is constructed by using adaptive kernel density estimation and fuzzy C-means clustering, and the probability density and membership function of each index under different health grades are determined; finally, a simulation prediction is performed on a target wheel set, the predicted value of the target wheel set is input into the state space model, and the dynamic combination weights and D-S evidence theory are fused to calculate a comprehensive health index and output a grading result, so that the wheel set health state can be accurately and efficiently evaluated.

Description

Technical Field

[0001] This invention relates to the field of railway vehicle maintenance technology, and in particular to a data- and knowledge-driven multi-parameter comprehensive condition assessment method and system for wheelsets. Background Technology

[0002] Wheelsets are core components of railway vehicles, and their performance and condition directly affect train operation safety, stability, and passenger comfort. In railway transportation, wheelsets bear complex wheel-rail forces, and their health plays a crucial role in train dynamics (such as smoothness and derailment safety) and passenger experience. Therefore, wheelset monitoring and maintenance have always been a key focus of railway vehicle maintenance.

[0003] The multidimensional geometric parameters of wheelsets (such as flange thickness, flange height, rim thickness, qR value, tread wear, etc.) and their derived parameters (such as equivalent taper) are key indicators determining the wheel-rail contact geometry. The wear and changes of these parameters during operation directly affect the distribution of wheel-rail contact points, the motion stability of the wheelset, and the overall vehicle dynamics. By comprehensively analyzing the multi-parameter state of the wheelset and its impact on the wheel-rail dynamic response, the health condition of the wheelset can be accurately assessed, providing a basis for scientific inspection and maintenance decisions.

[0004] The existing technology has the following main defects and shortcomings in the assessment and maintenance of wheelset condition for conventional passenger buses:

[0005] 1. Lack of comprehensive evaluation methods: Currently, the evaluation of wheelset condition relies heavily on experience-based judgment or measurement of single parameters, such as manual inspection using basic tools like the fourth type of inspection device. This approach cannot comprehensively and systematically reflect the overall impact of wheelset geometric parameter changes on vehicle dynamics performance, resulting in inaccurate evaluation results.

[0006] 2. Outdated Maintenance Model: Current wheelset maintenance largely adopts a planned maintenance model, which involves turning wheels according to fixed mileage (e.g., 600,000 kilometers) or fixed cycles. This "one-size-fits-all" model does not fully consider the actual condition of the wheelsets, potentially leading to two problems: first, wheelsets in good condition are over-maintained, increasing maintenance costs and wasting resources; second, deteriorating wheelsets are not detected in time, missing the optimal maintenance opportunity and posing safety hazards. Furthermore, some maintenance decisions rely on alarms from the Vehicle Performance Data System (TPDS) or the Train Control System (TCDS), but these systems typically collect data during train operation, resulting in delayed alarms. Moreover, replacing a single wheelset can trigger a chain reaction of alarms on other wheelsets, leading to unnecessary repeated disassembly and assembly and additional maintenance costs.

[0007] 3. Insufficient Detection Methods and Intelligence Level: Existing detection methods mainly rely on manual inspection or simple automated equipment. These methods can only obtain the surface geometric parameters of the wheelsets, making it difficult to deeply analyze the dynamic response of wheel-rail contact and its impact on dynamic performance. Evaluations relying on dynamic detection systems such as TPDS or TCDS need to be conducted during train operation, which is complex, costly, and limited in terms of data real-time performance and accuracy. Current technologies generally lack comprehensive evaluation methods based on artificial intelligence and data-driven approaches, making it impossible to achieve efficient and accurate evaluations under static conditions within the depot, and also difficult to automatically generate intelligent maintenance recommendations based on evaluation results.

[0008] Therefore, there is an urgent need to develop an intelligent evaluation method that can comprehensively consider the multidimensional geometric parameters of wheelsets and their impact on dynamic performance under static conditions in the depot, so as to achieve accurate determination of the health status of wheelsets and condition-based maintenance. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a data- and knowledge-driven multi-parameter comprehensive condition assessment method and system for wheelsets. This method, through comprehensive analysis of the multi-dimensional geometric parameters and derived parameters of the wheelset, combined with high-fidelity dynamic simulation (knowledge-driven) and multi-source data analysis and machine learning (data-driven), can achieve accurate and quantitative assessment of the health status of wheelsets under static conditions within a storage facility, providing a scientific basis for condition-based maintenance and intelligent management of wheelsets.

[0010] In a first aspect, the present invention provides a data- and knowledge-driven multi-parameter comprehensive state evaluation method for wheelsets, comprising the following steps:

[0011] S1: Based on the preset multidimensional geometric parameter space of wheelsets, parameter combination samples are generated through sampling methods, and the dynamic performance indicators corresponding to the parameter combination samples are calculated using a rigid-flexible coupling dynamic simulation platform to obtain an input-output dataset; global sensitivity analysis is performed based on the input-output dataset.

[0012] S2: Based on the results of the global sensitivity analysis, a subset of key dynamic performance indicators is selected from the aforementioned dynamic performance indicators;

[0013] S3: Based on the aforementioned subset of key dynamic performance indicators, the subjective and objective weights of each key dynamic performance indicator are determined using the analytic hierarchy process (AHP) and the entropy weight method, respectively.

[0014] S4: Construct a dynamic weight optimization model with the goal of minimizing the weighted deviation between subjective and objective weights, and introduce a dynamic adjustment mechanism based on the Bellman equation to fuse and iteratively optimize the subjective and objective weights to generate the final combined weights.

[0015] S5: Combining historical data and simulation data, the probability density function of each key dynamic performance index is fitted by the kernel density estimation algorithm, and the membership function of each index under multiple preset health levels is determined by the fuzzy C-means clustering algorithm, thereby constructing a multi-dimensional state space partitioning model.

[0016] S6: Obtain the actual geometric parameters of the wheelset to be evaluated, and use the dynamic simulation platform to calculate the predicted values ​​of various key dynamic performance indicators of the wheelset to be evaluated under preset operating conditions.

[0017] S7: Standardize the predicted values ​​and input the standardized data into the multidimensional state space partitioning model to calculate the membership value and probability density value of each indicator for each health level.

[0018] S8: By integrating the final combined weights, membership values, and probability density values, and using a decision fusion algorithm based on evidence theory, the comprehensive health index of the wheel pair to be evaluated is calculated, and its health status classification result is output according to a preset threshold.

[0019] As an optional implementation of the first aspect of this application, in step S1: the sampling method is Latin hypercube sampling to generate uniformly distributed parameter combination samples in the multidimensional geometric parameter space of the wheelset; the global sensitivity analysis is a global sensitivity analysis method based on variance decomposition, used to quantitatively calculate the first-order sensitivity index and full-order sensitivity index of each geometric parameter to each dynamic performance index, and to screen according to these indices in step S2.

[0020] As an optional implementation of the first aspect of this application, step S4, specifically includes: minimizing the sum of squared weighted deviations between subjective weights and objective weights, and obtaining a set of initial combination coefficients using the Lagrange multiplier method; constructing a dynamic programming model based on the Bellman equation, and iteratively optimizing the combination coefficients through value iteration until convergence, thereby generating the final combination weights.

[0021] As an optional implementation of the first aspect of this application, in step S5: the kernel density estimation algorithm is an adaptive kernel density estimation algorithm, which adjusts the smoothing bandwidth by calculating the local bandwidth factor for each data point; the initial cluster centers of the fuzzy C-means clustering algorithm are determined according to the health level boundary values ​​of each predefined key dynamic performance index, so as to ensure the physical meaning and convergence speed of the clustering results.

[0022] As an optional implementation of the first aspect of this application, in step S7: the standardization process adopts the Z-score standardization method, and the mean and standard deviation of the Z-score are derived from the full historical data and simulation data used to construct the multidimensional state space partitioning model in step S5; the standardized data are input into the membership function and probability density function to calculate the membership degree value and probability density value.

[0023] As an optional implementation of the first aspect of this application, in step S8: the decision fusion algorithm of the evidence theory is DS evidence theory, specifically including: treating each key dynamic performance index as an independent source of evidence, and constructing a basic probability allocation function for each source of evidence; wherein, for any health level, its basic probability allocation value is jointly determined by the final combined weight corresponding to the index and its membership value to the health level.

[0024] As an optional implementation of the first aspect of this application, step S8 further includes: iteratively fusing the basic probability allocation functions of all evidence sources using the Dempster combination rule to obtain the final fused evidence; and, presetting utility scores for each health level, and obtaining the comprehensive health index by calculating the weighted expected value of the beliefs of each health level in the final fused evidence.

[0025] Secondly, embodiments of this application provide a data- and knowledge-driven multi-parameter comprehensive condition assessment system for wheelsets, including:

[0026] The data acquisition and key indicator screening module is configured to generate parameter combination samples based on a preset multidimensional geometric parameter space of wheelsets through a sampling method, and calculate the dynamic performance indicators corresponding to the parameter combination samples using a rigid-flexible coupling dynamic simulation platform to obtain an input-output dataset; perform global sensitivity analysis based on the input-output dataset; and screen a subset of key dynamic performance indicators from the dynamic performance indicators based on the results of the global sensitivity analysis.

[0027] The indicator weight determination and dynamic optimization module is configured to determine the subjective and objective weights of each key dynamic performance indicator based on the subset of key dynamic performance indicators, using the analytic hierarchy process (AHP) and entropy weight method respectively; construct a weight dynamic optimization model with the goal of minimizing the weighted deviation between subjective and objective weights, and introduce a dynamic adjustment mechanism based on the Bellman equation to fuse and iteratively optimize the subjective and objective weights to generate the final combined weights.

[0028] The state space partitioning model construction module is configured to combine historical data and simulation data, use kernel density estimation algorithm to fit the probability density function of each key dynamic performance index, and use fuzzy C-means clustering algorithm to determine its membership function under multiple preset health levels, thereby constructing a multidimensional state space partitioning model.

[0029] The real-time state assessment and parameter calculation module is configured to obtain the actual geometric parameters of the wheelset to be assessed, and use the dynamic simulation platform to calculate the predicted values ​​of various key dynamic performance indicators of the wheelset to be assessed under preset operating conditions; the predicted values ​​are standardized, and the standardized data is input into the multi-dimensional state space partitioning model to calculate the membership degree value and probability density value of each indicator for each health level.

[0030] The decision fusion and health status output module is configured to fuse the final combined weights, membership values, and probability density values, use an evidence-based decision fusion algorithm to calculate the comprehensive health index of the wheel pair to be evaluated, and output its health status classification result according to a preset threshold.

[0031] Thirdly, embodiments of this application provide an electronic device, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of the method described in the first aspect.

[0032] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0033] Compared with existing technologies, this invention proposes a data- and knowledge-driven multi-parameter comprehensive state evaluation method for wheelsets, which has the following significant advantages:

[0034] 1. Comprehensiveness and accuracy of the assessment: This invention constructs a multi-parameter comprehensive state assessment system for wheelsets, which integrates multi-dimensional geometric parameters and their derived parameters, and comprehensively considers their deterioration effects on dynamic performance in multiple dimensions such as safety, stability, and comfort. This breaks through the limitations of traditional reliance on visual inspection or single parameter measurement, and the assessment results are more comprehensive and accurate.

[0035] 2. High Efficiency and Low Cost of Evaluation: This invention can accurately determine the dynamic performance of wheelsets under static conditions within a storage facility simply by measuring the wheelset's geometry and combining this with simulation calculations. This avoids costly, time-consuming, and time-consuming dynamic online detection, greatly improving evaluation efficiency and reducing maintenance costs.

[0036] 3. Intelligent and Scientific Decision-Making: This invention integrates advanced technologies such as machine learning (KDE, FCM), dynamic weight optimization (Bellman equation), and decision fusion (DS theory) to achieve full automation and intelligence from data acquisition to condition rating. By outputting a quantified comprehensive health index and clear health levels, it provides managers with direct and reliable scientific basis for making decisions such as condition-based maintenance, overhaul, or replacement, promoting the transformation of railway vehicle maintenance from "planned maintenance" to "condition-based maintenance." Attached Figure Description

[0037] Figure 1 This is a technical roadmap for a data- and knowledge-driven multi-parameter integrated state assessment method for wheelsets according to an embodiment of the present invention.

[0038] Figure 2 This is a hierarchical decomposition structure diagram in a data and knowledge-driven multi-parameter integrated state evaluation method for wheelsets according to an embodiment of the present invention.

[0039] Figure 3 This is a visualization diagram of the comprehensive weight of indicators in a data and knowledge-driven multi-parameter comprehensive state evaluation method for wheelsets according to an embodiment of the present invention.

[0040] Figure 4 This is a radar chart predicting the dynamic performance of the wheelset to be evaluated in a data and knowledge-driven multi-parameter integrated state evaluation method for wheelsets according to an embodiment of the present invention.

[0041] Figure 5 This is a belief allocation evolution diagram in the evidence fusion process of a data and knowledge-driven multi-parameter integrated state assessment method for wheelsets according to an embodiment of the present invention.

[0042] Figure 6 This is a schematic diagram of the structure of a data and knowledge-driven multi-parameter integrated state assessment system for wheelsets provided in an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0045] Example 1

[0046] Please see Figure 1 This is a flowchart illustrating a data- and knowledge-driven multi-parameter integrated state evaluation method for wheelsets, provided by an embodiment of the present invention. The method may include the following steps:

[0047] S1: Based on the preset multidimensional geometric parameter space of wheelsets, parameter combination samples are generated by sampling method, and the dynamic performance index corresponding to the parameter combination samples is calculated by using a rigid-flexible coupling dynamic simulation platform to obtain the input-output dataset; global sensitivity analysis is performed based on the input-output dataset.

[0048] This step, through systematic numerical experiments and data-driven methods, delves into the complex nonlinear relationship between multidimensional geometric wear parameters of wheelsets and their resulting vehicle dynamic response under dynamic operating conditions. Its core lies in establishing a nonlinear sensitivity index model to quantitatively assess the influence of each geometric parameter on key dynamic performance indicators, providing a scientific basis for subsequent screening of strongly correlated indicators.

[0049] I. Definition of wheelset geometric parameters in space

[0050] Define a vector containing 6 key geometric parameters of the wheelset. , which serves as the input variable space for the model.

[0051]

[0052] In the formula, each component represents a specific geometric parameter:

[0053] Flange thickness (Sd)

[0054] Flange height (Sh)

[0055] : Rim thickness

[0056] qR value

[0057] tread wear index

[0058] Equivalent taper (λe)

[0059] For each parameter The range of its values ​​is determined based on historical wear data, maintenance limits, and expert experience; that is, a p-dimensional hyperrectangular parameter space is defined. .

[0060]

[0061] in, and The first i The lower and upper limits of a geometric parameter.

[0062] II. Sample Matrix Generation Based on Latin Hypercube Sampling

[0063] In parameter space To efficiently and uniformly acquire sample points, the Latin hypercube sampling method is employed. With a total sample size of N, the LHS method assigns each parameter... The range of values segmentation N There are several equally probable subintervals. By analyzing these... N×6 The sub-intervals are randomly arranged and combined to generate N Each sample point is 6-dimensional. The final input sample matrix is ​​then constructed. :

[0064]

[0065] In the formula, It is the first j A combination sample of the geometric parameters of each wheelset.

[0066] III. High-fidelity dynamic simulation and response data acquisition

[0067] sample matrix Each sample vector in As input, a pre-established high-fidelity, nonlinear train-track coupled rigid-flexible dynamics simulation model is substituted. The model is capable of simulating the dynamic behavior of trains under specific track conditions and operating speeds.

[0068] For each input sample The simulation model outputs a... Response vectors of key dynamic performance indicators .

[0069]

[0070] In the formula, each component represents a specific dynamic index.

[0071] For all N Batch simulation of each sample is performed to obtain the output response matrix. :

[0072]

[0073] IV. Global Sensitivity Analysis

[0074] Based on the constructed input-output dataset A global sensitivity analysis method based on variance decomposition (Sobol' method) is used to quantitatively evaluate each input geometric parameter. For a specific output dynamic index The impact.

[0075] For the output Its total variance It can be decomposed into the sum of variances contributed by individual parameters and the interactions between parameters. Calculate the first-order sensitivity index. and full-order sensitivity index :

[0076] First-order sensitivity index ( ): Measure a single parameter Changes in output The direct contribution of variance.

[0077]

[0078] in, It is in a fixed position hour Expected value Indicates except All input parameters other than those specified.

[0079] Full-order sensitivity index ( ): Measurement parameter The main effect and its interaction with all other parameters on the output The total contribution of variance.

[0080]

[0081] By calculating the relationship between all input parameters and all key output metrics and The results of the model (i.e., the sensitivity index values ​​of each parameter) will be directly used in step S2 to screen out key dynamic performance indicators that are strongly correlated with the overall condition of the wheelset.

[0082] S2: Based on the results of the global sensitivity analysis, a subset of key dynamic performance indicators is selected from the dynamic performance indicators; S3: Based on the subset of key dynamic performance indicators, the subjective weight and objective weight of each key dynamic performance indicator are determined by using the analytic hierarchy process (AHP) and the entropy weight method, respectively.

[0083] I. Constructing the Hierarchical Analysis Model

[0084] To implement the Analytic Hierarchy Process (AHP), the evaluation objectives are first decomposed into a structured form (e.g., Figure 2 As shown, construct a top-down three-level hierarchical model.

[0085] 1. Target Layer (G): The highest layer of the model, which is the ultimate goal—the accurate evaluation of the overall health status of the wheel pair.

[0086] 2. Criterion Layer (T): The intermediate layer, based on the physical meaning and engineering properties of the indicators, divides the key dynamic indicators selected in step S2 into five criteria, T={T1,T2,T3,T4,T5}:

[0087] T1 (Safety): An indicator directly related to the safety of train operation.

[0088] T2 (Smoothness): An indicator reflecting the quality of train operation and the risk of instability.

[0089] T3 (Comfort and Service Performance): An indicator that relates to passenger experience and component service condition.

[0090] T4 (Structural Reliability and Durability): An indicator reflecting the stress and long-term wear trend of key components.

[0091] T5 (Stability): An index reflecting the lateral stability in wheel-rail response.

[0092] 3. Index Layer (I): The lowest level of the model, i.e., the key dynamic performance index vectors selected in step S2. Each specific indicator is assigned to a corresponding criterion layer. In this embodiment, the selected key dynamic performance indicators n=12, and their specific affiliations are as follows:

[0093] Belongs to T1: [Derailment coefficient, wheel load reduction rate, critical speed]

[0094] Belongs to T2: [lateral stability, vertical stability]

[0095] Belongs to T3: [Lateral Comfort, Vertical Comfort]

[0096] Belongs to T4: [Tread wear, wear power]

[0097] Belongs to T5: [Wheel and axle lateral force, wheel and rail lateral force, frame vibration acceleration]

[0098] II. Determining Subjective Weight Vectors Based on the Analytic Hierarchy Process (AHP)

[0099] By using expert scoring, the relative importance of each element at the same level with respect to the target at the next higher level is compared pairwise to construct a judgment matrix.

[0100] 1. Construct the judgment matrix:

[0101] Construct the judgment matrix of the criterion layer T relative to the target layer G. .

[0102] For each criterion Construct the judgment matrix of its subordinate indicator layer I relative to the criterion. Elements in the matrix Represents element i Relative to element j The importance ratio is calculated using a 1-9 scale.

[0103] 2. Calculate the weights and perform a consistency check:

[0104] For each judgment matrix, the weight vector is calculated using either the square root method or the eigenvalue method. Taking the square root method as an example, the geometric mean of each row of elements in the matrix is ​​calculated. :

[0105]

[0106] For vectors After normalization, the weight vector for this level is obtained:

[0107]

[0108] Calculate the largest eigenvalue And a consistency check is performed to ensure the logical consistency of the expert's judgment.

[0109] The local weights of each indicator layer relative to its corresponding criterion layer are multiplied by the weights of its superior criteria layer and then summed to obtain the global subjective weights of each indicator, ultimately forming a subjective weight vector. .

[0110] III. Determining the Objective Weight Vector Based on the Entropy Weight Method (EWM)

[0111] 1. Construct the original data matrix

[0112] The dynamic response simulation data matrix generated in step S1 In the process, extract the columns corresponding to the n=12 key indicators selected in step S2, and select... m Construct the original data matrix using representative simulation samples (operating conditions). ,in For the first i The first sample j Individual indicator values.

[0113] 2. Data standardization:

[0114] For matrix Standardization is performed to eliminate the influence of dimensions and orders of magnitude, resulting in a standardized matrix. .

[0115] 3. Calculate information entropy and weights:

[0116] Calculate the first j Under this indicator, the first i The proportion of each sample value to this indicator :

[0117]

[0118] Calculate the first Information entropy of the indicator :

[0119]

[0120] Where, constant ensure .

[0121] Information entropy The smaller the value, the better the indicator. The greater the degree of variation, the more information it provides, and therefore the greater its weight should be. Information redundancy is... Therefore, we obtain the first... Objective weight of each indicator :

[0122]

[0123] The objective weights of all indicators constitute the objective weight vector. .

[0124] S4: Construct a dynamic weight optimization model with the goal of minimizing the weighted deviation between subjective and objective weights, and introduce a dynamic adjustment mechanism based on the Bellman equation to fuse and iteratively optimize the subjective and objective weights to generate the final combined weights.

[0125] I. Establishment of the Initial Combined Weight Model

[0126] To integrate subjective prior knowledge with objective data, a static combined weight optimization model is first established. The combined weight vector is defined. yes and Linear weighted combination:

[0127]

[0128] In the formula, and These are the combination coefficients to be solved, representing the relative importance of subjective and objective weights, and satisfying the following constraints: .

[0129] The objective function is constructed with the goal of minimizing the sum of squared weighted deviations between the combined weights and the subjective and objective weights. :

[0130]

[0131] In the formula, This is a balancing factor used to adjust the relative importance of the deviation between subjective and objective weights. Substituting the definition, the optimization problem can be formalized as:

[0132]

[0133] The solution is obtained using the Lagrange multiplier method. The Lagrange function is constructed. :

[0134]

[0135] In the formula, These are Lagrange multipliers. Through the analysis of... By taking the partial derivatives and setting them equal to zero, and solving the system of equations, we can obtain a set of initial optimal combination coefficients. And calculate the initial combined weight vector. .

[0136] II. Introduction of a dynamic weight adjustment mechanism

[0137] Considering the dynamic changes of the system state over time, a dynamic programming method based on the Bellman equation is introduced to handle the combination coefficients. Iterative optimization is performed to minimize the long-term cumulative bias of the weights.

[0138] Define the state transition function for the weight vector, i.e., the weights from time step [step 1] to [time step 2]. arrive The evolutionary rules. The goal of weight updates is to make it approximate the evolution from the current... The ideal combination is defined. The combination weight at time t is defined as... The update process is as follows:

[0139]

[0140]

[0141] In the formula, To adjust the step size or learning rate, control the magnitude of dynamic updates.

[0142] The denominator is a normalization operation to ensure that the sum of the weights is 1. This state transition process can be denoted as: .

[0143] III. Dynamic optimization of combination coefficients based on Bellman equations

[0144] To find the optimal combination coefficients that minimize long-term deviation Construct the Bellman dynamic programming model.

[0145] 1. Define the state-value function: Define the state-value function. To start from the current weight state The minimum of the discounted cumulative sum of deviations from all future steps. The Bellman optimality equation is expressed as:

[0146]

[0147] In the formula, In the state Below, selection coefficient The resulting instantaneous deviation (i.e., the objective function value). This is a discount factor used to balance the importance of current bias and future bias.

[0148] 2. Iterative Solution: The Bellman equation is solved through value iteration or policy iteration, and the process is as follows:

[0149] Initialization: Set the iteration count Initialize the value function .

[0150] Iterative update: For each iteration (t=0,1,2,…), update the value function:

[0151]

[0152] In each step, a numerical optimization method (gradient descent) is used to find the value that minimizes the right-hand side. .

[0153] Convergence criterion: Iterate until the value function converges, i.e., the change between two iterations is less than a preset threshold. : .

[0154] Generate final weights: After the iteration converges, the optimal and stable combination coefficients are obtained. Using this optimal coefficient, calculate the final combined weight vector:

[0155]

[0156] The vector This refers to the final output, dynamically optimized combined weights.

[0157] See also Figure 3 This is a visualization of the comprehensive weights of the indicators in steps S3 and S4 above, used to show the subjective, objective, and combined weights of each dynamic indicator.

[0158] S5: Combining historical and simulation data, the probability density function of each key dynamic performance index is fitted using a kernel density estimation algorithm, and the membership function of each index under multiple preset health levels is determined using a fuzzy C-means clustering algorithm, thereby constructing a multi-dimensional state space partitioning model.

[0159] This step aims to construct a clear quantitative boundary for each predefined health level (excellent, good, average, poor, and very poor, a total of five levels). By comprehensively applying statistical and fuzzy mathematics theories, and processing the simulation data generated in step S1 and the actual historical data collected, the probability density function and membership function of each key dynamic indicator selected in step S2 under different health levels are established, thereby constructing a multi-dimensional state space partitioning model.

[0160] I. Construction of the analysis dataset

[0161] Integrate the dynamic simulation response matrix from step S1 Extract the columns corresponding to the n key dynamic indicators determined in step S2, and merge these data. For the nth... j Key Indicators All available data points constitute a one-dimensional dataset. ,in This represents the total number of data samples for this indicator.

[0162] II. To accurately describe the distribution characteristics of each indicator across the entire health status range, without being limited by any parameterized assumptions, an improved adaptive kernel density estimation algorithm (Adaptive KDE) is used for each indicator. j Fit its probability density function (PDF).

[0163] 1. For the dataset Its standard kernel density estimate at any point x Defined as:

[0164]

[0165] In the formula, For kernel functions (Gaussian kernel function). Bandwidth is a key parameter for controlling the smoothness of the surface.

[0166] 2. Adaptive bandwidth adjustment:

[0167] To address the issue of overly smooth data flow in sparse regions and insufficient smoothness in dense regions with fixed bandwidth, adaptive bandwidth is employed. An initial bandwidth is used. A preliminary estimate of the "guided" density was calculated. Then, for each data point Calculate a local bandwidth factor :

[0168]

[0169] In the formula, It is all Geometric mean of the values It is the sensitivity parameter (take) ).

[0170] Finally, the first j Adaptive kernel density estimation of each index It is given by the following formula:

[0171] .

[0172] III. Determination of State Level Membership Functions Based on Fuzzy C-Means (FCM) Clustering

[0173] To address the ambiguity and transitional nature of the boundaries between health levels, a fuzzy C-means (FCM) clustering algorithm is employed to cluster each indicator. Classified to Among the predefined health levels.

[0174] 1. Cluster center initialization based on expert knowledge:

[0175] To ensure the rapid convergence of the FCM algorithm and the physical meaning of the results, its initial cluster centers are determined based on the boundary value table of health levels for each indicator. For the th... j The initial cluster centers of the five health grades (AE) of the indicators. The calculation is as follows:

[0176] For intervals with clearly defined upper and lower bounds (e.g., level B of T11: 0.26–0.43), the initial center is taken as the midpoint; for intervals with only one-sided boundaries, a typical value close to the boundary is selected as the initial center based on engineering experience. Using this method, a high-quality initial cluster center is determined for each of the five evaluation levels (A, B, C, D, E) for all 12 indicators, forming a set of initial center vectors. .

[0177] 2. FCM Objective Function and Iterative Solution

[0178] by Starting from this point, the following objective function is minimized through iterative optimization.

[0179] (1) FCM objective function:

[0180] The FCM algorithm minimizes the following objective function. To achieve clustering:

[0181]

[0182] In the formula:

[0183] : The number of predefined health grades (cluster centers).

[0184] Data points Belonging to the Membership degree of each health level.

[0185] :No. j The first indicator Each health level has a cluster center, representing a typical value for that level.

[0186] :Fuzziness index, which controls the degree of fuzziness in the clustering results.

[0187] Membership degree must satisfy the following constraints: .

[0188] (2) Iterative solution:

[0189] Iteratively update cluster centers and membership matrix Please provide a solution:

[0190] Update cluster centers:

[0191]

[0192] Update membership:

[0193] .

[0194] 3. Generate membership functions:

[0195] After the algorithm converges, each index is obtained. j Stable cluster centers corresponding to five health levels Therefore, for any new index value x, its relation to the first... Membership functions for each health level :

[0196] .

[0197] IV. Formation of the Multidimensional State Space Partition Model

[0198] Based on the above results, the final state-space partitioning model consists of two parts:

[0199] Probability density function set: For each of the n key indicators, there is a set (5 in total) of probability density functions associated with it that describe its distribution characteristics at each health level. (Its integrals in different intervals can be regarded as probabilities of different levels).

[0200] Membership function set: For each of the n key indicators, there is a set (5 in total) of membership functions. It is used to calculate the degree to which any indicator value belongs to one of the five levels: "excellent", "good", "medium", "poor" and "inferior".

[0201] S6: Obtain the actual geometric parameters of the wheelset to be evaluated, and use the dynamic simulation platform to calculate the predicted values ​​of various key dynamic performance indicators of the wheelset to be evaluated under preset operating conditions.

[0202] This step marks the formal starting point of the evaluation process, transforming the current physical state of the wheelset to be evaluated into a set of quantifiable dynamic performance predictions. Using the simulation system built and validated in step S1, the dynamic response of the actually measured wheelset geometry parameters under typical operating scenarios is accurately calculated.

[0203] I. Acquisition and Vectorization of Geometric Parameters of the Wheelset to be Evaluated

[0204] Using high-precision laser profilometry equipment, a batch (total) of individuals requiring health status assessment were analyzed. Geometric parameters were measured for each of the wheelsets. For the first wheelset... One pair of wheels to be evaluated ( ), to obtain the latest measured values ​​of all 6 key geometric parameters.

[0205] These measurements are constructed in a way that is completely consistent with the structure of the steps. 3D geometric parameter vector :

[0206]

[0207] In the formula, It is the first The first round to be evaluated j The actual measured values ​​of each geometric parameter. All The parameter vectors of the wheelsets to be evaluated together constitute the input sample set for this evaluation.

[0208] II. Definition of Typical Operating Conditions

[0209] To ensure the consistency and comparability of the evaluation results, one or more standardized typical operating conditions need to be defined in advance. An operating condition is defined by a series of parameters and can be represented as a condition parameter vector. .

[0210] III. Batch Simulation Calculation of Key Dynamic Performance Indicators

[0211] The geometric parameter vector of each wheelset to be evaluated Compared with the preset typical working condition vector As input, the high-fidelity train-track coupled dynamics simulation model established in step S1 is substituted into the model. middle.

[0212] For the For each wheelset to be evaluated, the simulation model will calculate its complete dynamic response vector. Subsequently, from this complete response vector, based on the key index system determined in step S2, the predicted values ​​of 12 key dynamic performance indicators are extracted to form the predicted index vector for the wheelset. :

[0213]

[0214] In the formula, This represents a selection operator used to extract the items selected as key indicators in step S2 from the full simulation results. The structure is as follows:

[0215]

[0216] For all Repeat this process for each wheel pair to be evaluated, perform batch simulation calculations, and finally generate a... Dimensional dynamic index prediction matrix :

[0217]

[0218] This matrix contains the core dynamic behavior of each wheelset under typical operating conditions (e.g., Figure 4 As shown, the radar chart predicting the dynamic performance of the wheelset to be evaluated is used to visualize the predicted values ​​of various key dynamic indicators of one or more wheelsets to be evaluated and compare them with the performance partition thresholds. It provides the input set for subsequent standardization processing (step S7) and final health status rating (step S8).

[0219] S7: Standardize the predicted values ​​and input the standardized data into the multidimensional state space partitioning model to calculate the membership value and probability density value of each indicator for each health level.

[0220] I. The Z-score standardization method is adopted. The standardization benchmarks (mean and standard deviation) are derived from the full dataset containing simulation and historical data used for modeling in step S5.

[0221] 1. Calculate the reference statistic:

[0222] For the n=12 key dynamic indicators, the first one j Individual indicators ( ), from its full dataset Calculate the mean and standard deviation :

[0223]

[0224] .

[0225] 2. Implement standardization:

[0226] For the prediction matrix Each element in (i.e., the first) The first round to be evaluated Apply Z-score transformation to the predicted values ​​of each indicator:

[0227]

[0228] After processing, we obtain a Dimensional standardized dynamic index prediction matrix :

[0229] .

[0230] II. Calculation of Probability Density and Membership Features

[0231] The standardized predicted values ​​are then input one by one into the multidimensional state space partitioning model constructed in step S5 to extract its deep features. (Note: To ensure consistency, the KDE and FCM models in step S5 should also be based on the standardized full dataset.) (Construction)

[0232] 1. Probability density value calculation: Standardized predicted values Input its corresponding adaptive kernel density estimation function established in step S5 Calculate the probability density of the occurrence of this index value: All the calculation results constitute a P×12-dimensional probability density matrix. , of which elements This reflects the universality or anomaly of the observed values ​​of the indicators.

[0233] 2. Membership matrix generation: This involves generating standardized predicted values. Input its corresponding value generated by the FCM algorithm Membership functions In the middle. For the first The first round to be evaluated j One indicator, which corresponds to five health levels ( The membership vectors (where 1, ..., 5 correspond to Excellent, Good, Average, Poor, and Inferior, respectively) are:

[0234]

[0235] For each round pair to be evaluated The membership vectors of all 12 key indicators can be combined into a 12×5 dimensional membership matrix. It fully describes the first The fuzzy distribution of various dynamic performance characteristics of each wheelset across five health levels.

[0236] .

[0237] S8: By integrating the final combined weights, membership values, and probability density values, and using a decision fusion algorithm based on evidence theory, the comprehensive health index of the wheel pair to be evaluated is calculated, and its health status classification result is output according to a preset threshold.

[0238] I. Defining the Identification Framework: Identification Framework A set of five predefined health levels:

[0239]

[0240] The 12 key dynamic indicators were considered as 12 independent sources of evidence. For the first... The first round to be evaluatedj The first indicator, input the first... j The final combined weight of each indicator (Vector from step S4) Simultaneously enter the first... The first pair of wheels, the first j The first indicator for the first Membership value of each health level (Matrix from step S7) ).

[0241] Furthermore, regarding the first The first round to be evaluated j A dynamic index, its BPA function The specific generation process is as follows:

[0242] Basic probability allocation for a single health level: ;

[0243] Basic probability assignment of uncertainty: ;

[0244] This leads to the basic probability allocation function (BPA) for each piece of evidence.

[0245] II. By using the Dempster combination rule, the BPA functions of the 12 independent evidence bodies are iteratively merged to build consensus, reduce conflicts, and ultimately form a general evidence judgment.

[0246] Any two independent BPA functions and The combination of is denoted as For any nonempty subset The calculation formula is:

[0247]

[0248] Among them, the conflict coefficient Used to measure the degree of conflict between two sources of evidence.

[0249] For the The fusion process of the 12 indicators of evidence for each pair of pairs to be evaluated is sequential:

[0250] (1) Initial fusion results:

[0251] (2) Iterative fusion: ,in

[0252] (3) Final fusion result:

[0253] After 11 combination operations, the final result is Information from all 12 dynamic indicators was integrated as the basis for the final decision.

[0254] III. Calculation of Comprehensive Health Index (CHI)

[0255] In order to transform the merged BPA function (which still includes beliefs about each level) into a single, intuitive, and comparable continuous score, this invention defines a comprehensive health index.

[0256] Define the utility value of health levels: Assign specific utility scores representing the degree of health to the five health levels H1 to H5, forming a utility vector. This vector specifies that level A corresponds to 100 points, level B corresponds to 80 points, and so on, with level E corresponding to 20 points.

[0257] CHI is defined as the mathematical expectation of the utility value for each health class. Its calculation is based on the fused BPA function. The beliefs assigned to specific levels are normalized (using the Pignistic probability transformation in DS theory) to eliminate global uncertainty terms. Interference:

[0258]

[0259] It is the first The final quantitative score of the overall health status of each round of evaluation is in the range of [20, 100]. The higher the score, the better the health status.

[0260] IV. By mapping continuous CHI scores to discrete health levels, a final, easy-to-understand assessment conclusion is output.

[0261] The threshold for determining the level is defined as follows:

[0262] Boundary A: T1=90

[0263] B boundary: T2=70

[0264] Boundary C: T3=50

[0265] Boundary points D and E: T4 = 30

[0266] Execute the judgment and output: Based on the comparison between the calculated CHIs and the above threshold, the judgment logic is as follows:

[0267] If CHIs > 90, the final status level is determined to be H1(A).

[0268] If 70 < CHIs ≤ 90, the final state level is determined as H2(B).

[0269] If 50 < CHIs ≤ 70, the final state level is determined as H3(C).

[0270] If 0 < CHIs ≤ 50, the final state level is determined as H4(D).

[0271] If CHIs ≤ 30, the final state level is determined as H5(Poor E).

[0272] So far, the system has completed the evaluation of a single wheel set, and output its unique comprehensive health index (CHI) and the corresponding clear health status classification (A - E), providing a direct, quantitative and highly reliable scientific basis for the user's management decisions such as maintenance and replacement. Please refer to Figure 5 , which is the belief assignment evolution diagram in the evidence fusion process, used to show the dynamic change process of the system's belief and uncertainty about each health level when multiple index evidences are fused one by one under the D - S evidence theory.

[0273] Embodiment 2

[0274] Please refer to Figure 6 , which shows the structural schematic diagram of a wheel set multi - parameter comprehensive state evaluation system driven by both data and knowledge proposed in the second embodiment of this application. The system includes the following key modules:

[0275] Data acquisition and key index screening module 100, configured to generate parameter combination samples based on a preset multi - dimensional geometric parameter space of the wheel set through sampling methods, and calculate the dynamic performance indicators corresponding to the parameter combination samples using a rigid - flexible coupling dynamics simulation platform to obtain an input - output data set; conduct global sensitivity analysis based on the input - output data set; and screen out a subset of key dynamic performance indicators from the dynamic performance indicators according to the results of the global sensitivity analysis

[0276] Index weight determination and dynamic optimization module 200, configured to determine the subjective weight and objective weight of each key dynamic performance indicator based on the subset of key dynamic performance indicators, respectively using the analytic hierarchy process and the entropy weight method; construct a weight dynamic optimization model with the goal of minimizing the weighted deviation between the subjective weight and the objective weight, and introduce a dynamic adjustment mechanism based on the Bellman equation to fuse and iteratively optimize the subjective weight and the objective weight to generate the final combined weight

[0277] The state space partitioning model construction module 300 is configured to combine historical data and simulation data, use kernel density estimation algorithm to fit the probability density function of each key dynamic performance index, and use fuzzy C-means clustering algorithm to determine its membership function under multiple preset health levels, thereby constructing a multidimensional state space partitioning model.

[0278] The real-time state assessment and parameter calculation module 400 is configured to obtain the actual geometric parameters of the wheelset to be assessed, and use the dynamic simulation platform to calculate the predicted values ​​of various key dynamic performance indicators corresponding to the wheelset to be assessed under preset operating conditions; the predicted values ​​are standardized, and the standardized data is input into the multi-dimensional state space partitioning model to calculate the membership degree value and probability density value of each indicator for each health level.

[0279] The decision fusion and health status output module 500 is configured to fuse the final combined weights, membership values ​​and probability density values, use the decision fusion algorithm based on evidence theory to calculate the comprehensive health index of the wheel pair to be evaluated, and output its health status classification result according to a preset threshold.

[0280] The data- and knowledge-driven multi-parameter integrated state assessment system for wheelsets, as described in this application embodiment, can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), etc. This application embodiment does not impose specific limitations.

[0281] The data- and knowledge-driven multi-parameter integrated state evaluation system for wheelsets in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0282] This application provides a data- and knowledge-driven multi-parameter integrated condition assessment system for wheelsets, which can achieve... Figure 1The various processes of the data- and knowledge-driven multi-parameter integrated state evaluation method for wheelsets in the method embodiment are not described in detail here to avoid repetition.

[0283] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the various processes of the above-described embodiment of a data and knowledge-driven multi-parameter comprehensive state evaluation method for wheelsets, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0284] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described embodiment of a data and knowledge-driven multi-parameter integrated state evaluation method for wheelsets, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0285] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0286] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0287] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0288] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A data- and knowledge-driven multi-parameter comprehensive state assessment method for wheelsets, characterized in that, Includes the following steps: S1: Based on the preset multidimensional geometric parameter space of wheelsets, parameter combination samples are generated through sampling methods, and the dynamic performance indicators corresponding to the parameter combination samples are calculated using a rigid-flexible coupling dynamic simulation platform to obtain an input-output dataset; global sensitivity analysis is performed based on the input-output dataset. S2: Based on the results of the global sensitivity analysis, a subset of key dynamic performance indicators is selected from the aforementioned dynamic performance indicators; S3: Based on the aforementioned subset of key dynamic performance indicators, the subjective and objective weights of each key dynamic performance indicator are determined using the analytic hierarchy process (AHP) and the entropy weight method, respectively. S4: Construct a dynamic weight optimization model with the goal of minimizing the weighted deviation between subjective and objective weights, and introduce a dynamic adjustment mechanism based on the Bellman equation to fuse and iteratively optimize the subjective and objective weights to generate the final combined weights. S5: Combining historical data and simulation data, the probability density function of each key dynamic performance index is fitted by the kernel density estimation algorithm, and the membership function of each index under multiple preset health levels is determined by the fuzzy C-means clustering algorithm, thereby constructing a multi-dimensional state space partitioning model. S6: Obtain the actual geometric parameters of the wheelset to be evaluated, and use the dynamic simulation platform to calculate the predicted values ​​of various key dynamic performance indicators of the wheelset to be evaluated under preset operating conditions. S7: Standardize the predicted values ​​and input the standardized data into the multidimensional state space partitioning model to calculate the membership value and probability density value of each indicator for each health level. S8: By integrating the final combined weights, membership values, and probability density values, and using a decision fusion algorithm based on evidence theory, the comprehensive health index of the wheel pair to be evaluated is calculated, and its health status classification result is output according to a preset threshold.

2. The method according to claim 1, characterized in that, In step S1: The sampling method is Latin hypercube sampling, which generates uniformly distributed parameter combination samples within the multidimensional geometric parameter space of the wheelset. The global sensitivity analysis is a global sensitivity analysis method based on variance decomposition, used to quantitatively calculate the first-order sensitivity index and full-order sensitivity index of each geometric parameter to each dynamic performance index, and to screen according to these indices in step S2.

3. The method according to claim 1, characterized in that, In step S4, the construction of the weight dynamic optimization model specifically includes: With the objective of minimizing the sum of squared weighted deviations between subjective and objective weights, a set of initial combination coefficients is obtained by using the Lagrange multiplier method. A dynamic programming model based on the Bellman equation is constructed, and the combination coefficients are iteratively optimized through value iteration until convergence, thereby generating the final combination weights.

4. The method according to claim 1, characterized in that, In step S5: The kernel density estimation algorithm is an adaptive kernel density estimation algorithm, which adjusts the smoothing bandwidth by calculating a local bandwidth factor for each data point. The initial cluster centers of the fuzzy C-means clustering algorithm are determined based on the predefined health level boundary values ​​of each key dynamic performance index to ensure the physical meaning and convergence speed of the clustering results.

5. The method according to claim 1, characterized in that, In step S7: The standardization process employs the Z-score standardization method, and the mean and standard deviation of the Z-score are derived from the full historical data and simulation data used in step S5 to construct the multidimensional state space partitioning model. The standardized data is input into the membership function and probability density function to calculate the membership degree value and probability density value.

6. The method according to claim 1, characterized in that, In step S8: The decision fusion algorithm of the evidence theory is DS evidence theory, which specifically includes: treating each key dynamic performance index as an independent source of evidence and constructing a basic probability allocation function for each source of evidence; For any health level, its basic probability allocation value is determined by the final combined weight of the indicator and its membership value to that health level.

7. The method according to claim 6, characterized in that, Step S8 further includes: The Dempster combination rule is used to iteratively fuse the basic probability assignment functions of all evidence sources to obtain the final fused evidence. Furthermore, a utility score is preset for each health level, and the comprehensive health index is obtained by calculating the weighted expected value of the beliefs of each health level in the final fused evidence.

8. A data- and knowledge-driven multi-parameter integrated state assessment system for wheelsets, characterized in that, include: The data acquisition and key indicator screening module is configured to generate parameter combination samples based on a preset multidimensional geometric parameter space of wheelsets through a sampling method, and calculate the dynamic performance indicators corresponding to the parameter combination samples using a rigid-flexible coupling dynamic simulation platform to obtain an input-output dataset. Perform global sensitivity analysis based on the input-output dataset; Based on the results of the global sensitivity analysis, a subset of key dynamic performance indicators was selected from the aforementioned dynamic performance indicators; The indicator weight determination and dynamic optimization module is configured to determine the subjective and objective weights of each key dynamic performance indicator based on the subset of key dynamic performance indicators, using the analytic hierarchy process (AHP) and entropy weight method respectively; construct a weight dynamic optimization model with the goal of minimizing the weighted deviation between subjective and objective weights, and introduce a dynamic adjustment mechanism based on the Bellman equation to fuse and iteratively optimize the subjective and objective weights to generate the final combined weights. The state space partitioning model construction module is configured to combine historical data and simulation data, use kernel density estimation algorithm to fit the probability density function of each key dynamic performance index, and use fuzzy C-means clustering algorithm to determine its membership function under multiple preset health levels, thereby constructing a multidimensional state space partitioning model. The real-time status assessment and parameter calculation module is configured to obtain the actual geometric parameters of the wheelset to be evaluated, and use the dynamic simulation platform to calculate the predicted values ​​of various key dynamic performance indicators of the wheelset to be evaluated under preset operating conditions. The predicted values ​​are standardized, and the standardized data is input into the multidimensional state space partitioning model to calculate the membership degree and probability density value of each indicator for each health level. The decision fusion and health status output module is configured to fuse the final combined weights, membership values, and probability density values, use an evidence-based decision fusion algorithm to calculate the comprehensive health index of the wheel pair to be evaluated, and output its health status classification result according to a preset threshold.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of the data and knowledge dual-driven multi-parameter integrated state evaluation method for wheelsets as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The program or instructions are stored on the readable storage medium, and when the program or instructions are executed by the processor, they implement the steps of the data and knowledge dual-driven multi-parameter integrated state evaluation method for wheelsets as described in any one of claims 1-7.

Images