A method and system for predicting the friction and wear characteristics of a pantograph carbon strip

Abstract

The application provides a pantograph carbon slide plate friction and wear characteristic prediction method and system, relates to the railway maintenance technical field, and comprises the following steps: obtaining a pantograph carbon slide plate friction and wear data signal set; constructing a pantograph carbon slide plate friction and wear physical model; sending the data signal set to the pantograph carbon slide plate friction and wear physical model for processing to obtain a pantograph carbon slide plate friction and wear characteristic data set; performing normalization processing on the pantograph carbon slide plate friction and wear characteristic data set, and establishing a deep neural network model embedded with a physical model based on the normalized data set; optimizing the deep neural network model, establishing an iterative optimization framework with physical interpretation, and obtaining a prediction model for real-time prediction of pantograph carbon slide plate friction and wear characteristics. The application provides an efficient and reliable solution for real-time monitoring and prediction of pantograph carbon slide plate friction and wear characteristics.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of railway maintenance, in particular to a pantograph carbon strip friction and wear characteristic prediction method and system. BACKGROUND

[0002] As a core component of the pantograph of electrified railway, the carbon strip realizes power transmission by sliding contact with the overhead contact line, and its friction and wear characteristics directly affect the safety and stability of vehicle operation. During the long-term service of high-speed trains, the surface of the carbon strip gradually wears out or even cracks due to the combined effects of mechanical friction, arc erosion and various environmental factors, which may cause poor contact between the pantograph and the overhead contact line, leading to power interruption or equipment damage. Therefore, accurately predicting the friction and wear characteristics of the carbon strip is of great significance for formulating accurate maintenance strategies, prolonging the service life of components and ensuring the safe operation of high-speed trains.

[0003] Current methods for predicting the friction and wear characteristics of carbon strips mainly fall into two categories: one is the physical model-based method: by establishing tribology equations, material mechanics equations or thermodynamic models, combined with working condition parameters for prediction. This method has clear physical interpretability, but it is difficult to fully reflect the actual complex working conditions, and the simplifying assumptions in the model may lead to deviations between the predicted results and the actual situation. The second is the data-driven method: using machine learning algorithms to train historical monitoring data to establish the mapping relationship between input features and friction and wear characteristics. However, this method is highly dependent on data quality and quantity, and its generalization ability decreases significantly when the data is sparse or noisy, and it lacks constraints on physical laws, which may lead to discrepancies between the predicted results and the true physical mechanism.

[0004] Therefore, there is a need for a pantograph carbon strip friction and wear characteristic prediction method and system to solve the above problems. SUMMARY

[0005] The present application aims to provide a pantograph carbon strip friction and wear characteristic prediction method and system to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present application is as follows:

[0006] In a first aspect, the present application provides a pantograph carbon strip friction and wear characteristic prediction method, comprising:

[0007] obtaining a set of pantograph carbon strip friction and wear data signals;

[0008] based on the set of pantograph carbon strip friction and wear data signals, constructing a pantograph carbon strip friction and wear physical model;

[0009] sending the data signal set to the pantograph carbon strip friction and wear physical model for processing to obtain a set of pantograph carbon strip friction and wear characteristic data;

[0010] normalizing the pantograph carbon strip friction and wear characteristic dataset, and establishing a deep neural network model embedded with a physical model based on the normalized dataset, wherein the model comprises a dual loss function of a physical information constraint term and a data fitting term;

[0011] optimizing the deep neural network model, and establishing an iterative optimization framework with physical interpretability to obtain a prediction model for real-time prediction of the pantograph carbon strip friction and wear characteristics.

[0012] In a second aspect, the application further provides a pantograph carbon strip friction and wear characteristic prediction system, comprising:

[0013] an acquisition unit configured to acquire a pantograph carbon strip friction and wear data signal set;

[0014] a construction unit configured to construct a pantograph carbon strip friction and wear physical model based on the pantograph carbon strip friction and wear data signal set;

[0015] a processing unit configured to send the data signal set to the pantograph carbon strip friction and wear physical model for processing to obtain a pantograph carbon strip friction and wear characteristic dataset;

[0016] a calculation unit configured to normalize the pantograph carbon strip friction and wear characteristic dataset, and establish a deep neural network model embedded with a physical model based on the normalized dataset, wherein the model comprises a dual loss function of a physical information constraint term and a data fitting term;

[0017] an optimization unit configured to optimize the deep neural network model, and establish an iterative optimization framework with physical interpretability to obtain a prediction model for real-time prediction of the pantograph carbon strip friction and wear characteristics.

[0018] The application has the following advantages:

[0019] The application is based on a pantograph carbon strip friction and wear physical model, develops a machine learning architecture with physical information fusion, and proposes a physical information constraint strategy to optimize the hyperparameters of the prediction model. The optimization process is accelerated, the model interpretability and prediction accuracy are improved, and an efficient and reliable solution is provided for real-time monitoring and prediction of the friction and wear characteristics of the pantograph carbon strip of a high-speed train. The application improves the prediction accuracy while considering physical interpretability, computational efficiency and engineering practicability, provides reliable technical support for intelligent operation and maintenance of the pantograph carbon strip of a high-speed railway, and is especially suitable for online monitoring of friction and wear characteristics under complex dynamic contact conditions.

[0020] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or can be learned by practice of the application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 Flow chart of the method for predicting the friction and wear characteristics of the carbon slide plate of the pantograph described in the embodiments of the present application;

[0023] Figure 2 Structure schematic diagram of the system for predicting the friction and wear characteristics of the carbon slide plate of the pantograph described in the embodiments of the present application.

[0024] In the drawings: 701, acquisition unit; 702, construction unit; 703, processing unit; 704, calculation unit; 705, optimization unit. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art without creative labor on the basis of the embodiments in the present application belong to the scope of protection of the present application.

[0026] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0027] Embodiment 1:

[0028] This embodiment provides a method for predicting the friction and wear characteristics of a pantograph carbon sliding plate.

[0029] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0030] Step S1: Acquire the data signal set of friction and wear of the pantograph carbon sliding plate;

[0031] Understandably, in this step, the operating data of the pantograph carbon sliding contactor is collected in real time through a sensor network installed on the railway line. The data signal set includes multi-source parameters such as current (I), sliding speed (v), contact pressure (p), voltage (U), and historical wear (h). The sampling frequency is 100Hz, and continuous monitoring is conducted for 12 hours, collecting approximately 4.32 million data points to accurately characterize the performance changes of the carbon sliding contactor under different operating conditions.

[0032] Step S2: Based on the friction and wear data signal set of the pantograph carbon sliding plate, construct a physical model of the friction and wear of the pantograph carbon sliding plate;

[0033] It is understandable that this step is based on the friction and wear data signal set to construct a physical model of the friction and wear of the pantograph carbon sliding plate, which includes mechanical wear and arc wear terms. In this step, step S2 includes steps S21 and S22.

[0034] Step S21: Construct the mechanical wear term of the physical model based on the preset calculation formula for calculating mechanical wear;

[0035] It is understood that the mechanical wear amount is calculated using the following formula:

[0036] ;

[0037] in, This refers to mechanical wear. Frictional heat represents the energy source for melting wear, which is... ; For adhesive wear; The coefficient of friction; For contact pressure; The sliding speed; Runtime; The heat of fusion of the skateboard material; The wear coefficient; This is the sliding distance; The hardness of the skateboard material.

[0038] Step S22: Construct the arc wear term of the physical model based on the preset calculation formula for calculating arc wear.

[0039] The amount of arc wear is calculated using the following formula:

[0040] ;

[0041] in, This refers to arc wear. It is the energy of the electric arc; and These represent the voltage and current during the generation of the electric arc, and how they change over time. Sublimation heat for skateboard materials; This is a correction factor; This represents the duration of the electric arc.

[0042] Step S3: Send the data signal set to the pantograph carbon sliding plate friction and wear physical model for processing to obtain the pantograph carbon sliding plate friction and wear characteristic dataset;

[0043] Understandably, this step utilizes the constructed physical model to enhance and expand the data signal set within the physical constraint space using a Gaussian noise enhancement method, generating a characteristic dataset reflecting the friction and wear patterns. In this step, step S3 includes steps S31 and S32.

[0044] Step S31: Based on the physical constraint space defined by the constructed physical model, Gaussian noise is introduced into the physical constraint space to expand the data signal set, thereby obtaining the expanded data signal set;

[0045] The probability density function of Gaussian noise is shown below:

[0046] ;

[0047] in, It is Gaussian noise. For the original data points, The mean, To determine the standard deviation, in this embodiment, Gaussian noise with a standard deviation σ = 0.05 is added to the current signal, Gaussian noise with a standard deviation σ = 0.03 is added to the sliding speed, and Gaussian noise with a standard deviation σ = 0.04 is added to the contact pressure. By adaptively adjusting the σ value, it is ensured that the enhanced data does not exceed the physical constraint space, for example, the friction coefficient μ does not exceed the theoretical upper limit of 0.35, and the wear amount does not become negative.

[0048] Step S32: Merge the expanded data signal set with the original data signal set, and combine it with the parameters of the physical model to extract feature variables that reflect the friction and wear law, and generate a uniformly distributed feature dataset.

[0049] It can be understood that this step expands the original data within the constraint space defined by the physical model (such as the friction coefficient range, contact pressure threshold) using a dynamic Gaussian noise injection strategy: the noise amplitude is adaptively adjusted according to the measured data variance and the allowed deviation of physical parameters (such as the friction coefficient fluctuation range ±0.05), which not only simulates the dynamic disturbance in actual working conditions (such as the instantaneous impact caused by the unevenness of the catenary), but also avoids generating data that violates physical laws (such as negative wear rate). In the data fusion stage, by weighted mixing of the original data and the enhanced data (weight ratio 7:3), the total amount of data is expanded from 4.32 million to 7 million while preserving the distribution characteristics of the real data. In the feature extraction stage, combined with the material hardness parameters of the Archard wear model and the Hertz contact theory, a multi-dimensional feature space is constructed: ① physical derived features (such as equivalent wear depth calculated based on contact patch area); ② time-frequency domain composite features (such as wavelet packet energy entropy of vibration signal); ③ working condition related features (such as the interaction term of sliding speed-contact pressure). Through data enhancement technology, the original data set is expanded from 4.32 million data points to about 7 million data points, effectively improving the balance of data distribution and the robustness of model training.

[0050] Step S4, normalizing the pantograph carbon slide plate friction and wear characteristic data set, and establishing a deep neural network model embedded with a physical model based on the normalized data set, the model including a dual loss function of physical information constraint term and data fitting term;

[0051] It can be understood that in this step, a linear normalization method is used to unify the dimension of multi-source heterogeneous data, and each feature value is mapped to the [0, 1] interval, which not only eliminates the risk of gradient explosion caused by the difference in sensor magnitudes (such as the weight suppression of high-amplitude vibration signals on low-magnitude electrical noise features), but also makes the step size of each parameter update in the backpropagation process tend to be balanced. Step S4 includes step S41 and step S42.

[0052] Step S41, normalizing the pantograph carbon slide plate friction and wear characteristic data set based on the linear normalization method, and dividing the normalized data set into training set, validation set and test set in the ratio of 8:1:1;

[0053] It can be understood that in this step, the characteristic data set is linearly normalized to the [0, 1] interval, and the normalization formula is as follows:

[0054] ;

[0055] Where, is the original feature value, and These are the minimum and maximum values ​​of the feature, respectively. This is the normalized value.

[0056] In this step, frictional heat is... Arc energy Normalize the features to eliminate the differences in the units of measurement between the features, and divide the dataset into training set, validation set and test set in a ratio of 8:1:1.

[0057] Step S42: Build a deep neural network model based on the training set, validation set, and test set, and integrate the physical model to construct a dual loss function that includes physical information constraint terms and data fitting terms.

[0058] It is understandable that in this step, the dual loss function is defined as follows:

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] in, These are neural network parameters. Input parameters (signal set data). For the data fitting term, where This represents the actual wear and tear. To predict the wear amount output by the model ( ), The number of samples; For physical information constraints, and To calculate the mechanical wear and arc wear, and This is the corresponding value output by the prediction model; , These are weighting coefficients used to balance data fitting and physical constraints.

[0066] Step S5: Optimize the deep neural network model and establish an iterative optimization framework with physical interpretation to obtain a prediction model for real-time prediction of the friction and wear characteristics of the pantograph carbon sliding plate.

[0067] It can be understood that in the present step, step S5 includes step S51 and step S52.

[0068] Step S51, taking the double loss function as the optimization target, step-by-step optimization of feature extraction and time series prediction ability through a multi-stage hybrid training mechanism, and determination of the model parameter feasible region combined with hyperparameter sensitivity analysis.

[0069] It can be understood that the present step takes the double loss function as the optimization target, step-by-step optimization of feature extraction and time series prediction ability through a multi-stage hybrid training mechanism, and determination of the model parameter feasible region combined with hyperparameter sensitivity analysis.

[0070] In the present embodiment, the physical information fusion machine learning prediction model is decomposed into a feature extraction module and a time series prediction module, and a convolutional neural network (CNN) and a long short-term memory network (LSTM) are respectively used. The CNN module contains 3 convolutional layers, respectively using 32, 64 and 128 convolutional kernels, with a convolutional kernel size of 3x3 and a step size of 1, followed by a max pooling layer and a batch normalization layer after each layer; the LSTM module contains 2 layers of bidirectional LSTM layers, each with 128 hidden units, followed by a fully connected layer to output the prediction result.

[0071] The training adopts a three-stage strategy:

[0072] In the first stage, only the data fitting term is used to train the CNN module, the learning rate is set to 0.001, the batch size is 64, and the training is performed for 50 rounds;

[0073] In the second stage, the output of the CNN module trained in the first stage is used as the input, and only the data fitting term is used to train the LSTM module, the learning rate is set to 0.0005, the batch size is 32, and the training is performed for 30 rounds;

[0074] In the third stage, the CNN-LSTM model is trained jointly, the complete double loss function is used, the learning rate is set to 0.0002, the batch size is 32, and the training is performed for 100 rounds.

[0075] Through hyperparameter sensitivity analysis, the parameter feasible region of the CNN-LSTM model is determined: the number of convolutional kernels ranges from 16 to 256, the number of LSTM hidden units ranges from 64 to 256, the learning rate ranges from 0.0001 to 0.01, and the β weight coefficient ranges from 0.2 to 0.8.

[0076] Step S52, within the model parameter feasible region, a sparrow search algorithm guided by physical information is used to realize adaptive optimization of hyperparameters, and an optimized neural network model is obtained.

[0077] It can be understood that in the present step, a sparrow search algorithm guided by physical information is used to realize adaptive optimization of hyperparameters within the parameter feasible region.

[0078] The embodiment adopts a physical information constrained sparrow search algorithm (PISSA), specifically including:

[0079] The parameters (such as friction coefficient, contact pressure, sliding speed, etc.) in the physical model constructed in step S2 are taken as constraint conditions to narrow the search space of hyperparameters; the population size is initialized to 30, and the maximum number of iterations is 100. Each sparrow represents a combination of hyperparameters, such as the number of convolution kernels, the number of LSTM hidden units, the learning rate, and the β weight coefficient, etc. The population matrix is represented as:

[0080] ;

[0081] wherein, is the number of sparrows, is the dimension of hyperparameters.

[0082] The fitness value of each sparrow is calculated according to the double loss function of step S5, and the fitness function is defined as:

[0083] ;

[0084] wherein, is the number of test set samples, is the model prediction value, is the actual value. Through the iterative updating mechanism, combined with the physical information constraint, the best hyperparameter combination is searched.

[0085] Through the iterative updating mechanism of the sparrow search algorithm, combined with the physical information constraint, the best hyperparameter combination is searched, and the position updating formula of the producer (explorer) is:

[0086] ;

[0087] wherein, is the current iteration number; is the maximum iteration number; is a random number in the range of (0, 1]; is a random number subject to normal distribution; is a vector with all elements being 1; is an alarm, ; is a safety threshold, .

[0088] The position updating formula of the forager (follower) is:

[0089] ;

[0090] wherein, is the optimal position of the current producer; the global worst position; a random vector, ; and are defined as above.

[0091] Through the PISSA algorithm, the optimal super parameter combination is obtained by iterative search: the number of convolution kernels [32, 64, 128], the number of LSTM hidden units 128, the learning rate 0.0003, and the beta weight coefficient 0.6.

[0092] It can be understood that step S5 further includes step S53, step S54, step S55 and step S56.

[0093] Step S53, initialize the parameters of the optimized neural network model, and forward propagation to calculate the predicted wear amount and the intermediate variables of the physical model;

[0094] It can be understood that this step significantly improves the initial convergence speed of the prediction model and the physical interpretability through the physically guided model parameter initialization and the multi-level feature fusion mechanism. After initializing the parameters of the prediction model, the forward propagation calculates the predicted wear amount and the intermediate variables of the physical model and ;

[0095] Step S54, calculate the double loss function based on the predicted wear amount and the intermediate variables of the physical model, and use it as the updated parameter for back propagation update;

[0096] It can be understood that this step calculates the double loss function as the back propagation update model parameter, and the formula of the back propagation update is as follows:

[0097] ;

[0098] wherein, is the model parameter of the current iteration step; is the updated model parameter; is the learning rate (0.0003 in the optimal parameter combination); is the loss function to the gradient of the parameter .

[0099] Step S55, re-optimize the model parameters based on the physical constraints to ensure that the intermediate variables of the physical model meet the physical boundary conditions and the preset conditions;

[0100] It can be understood that this step re-optimizes the model parameters based on the physical constraints to ensure that and meet the physical boundary conditions.

[0101] Step S56, repeat all the above steps except the initialization process until the model meets the preset convergence condition, and obtain the prediction model for real-time prediction of the friction and wear characteristics of the pantograph carbon strip.

[0102] It can be understood that steps S53-S55 are repeated until the model converges, and the convergence condition is that the validation set loss changes less than 0.001 for 5 consecutive rounds.

[0103] The optimized prediction model achieves high prediction accuracy on the test set, with an average relative error of 3.6%, a root mean square error of 0.021 mm, and an R² value of 0.968, which is significantly better than the prediction results of traditional physical models and pure data-driven models.

[0104] Embodiment 2:

[0105] As shown in Figure 2 The embodiment provides a pantograph carbon strip friction and wear characteristic prediction system, see Figure 2 The system comprises an acquisition unit 701, a construction unit 702, a processing unit 703, a calculation unit 704 and an optimization unit 705.

[0106] The acquisition unit 701 is configured to acquire a pantograph carbon strip friction and wear data signal set.

[0107] The construction unit 702 is configured to construct a pantograph carbon strip friction and wear physical model based on the pantograph carbon strip friction and wear data signal set.

[0108] The processing unit 703 is configured to send the data signal set to the pantograph carbon strip friction and wear physical model for processing to obtain a pantograph carbon strip friction and wear characteristic data set.

[0109] The calculation unit 704 is configured to normalize the pantograph carbon strip friction and wear characteristic data set, and establish a deep neural network model embedded with a physical model based on the normalized data set, wherein the model comprises a dual loss function of a physical information constraint term and a data fitting term.

[0110] The optimization unit 705 is configured to optimize the deep neural network model and establish an iterative optimization framework with physical interpretation to obtain a prediction model for real-time prediction of the friction and wear characteristics of the pantograph carbon strip.

[0111] It should be noted that, as for the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment related to the method, and will not be described in detail here.

[0112] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0113] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0113] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0113] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0113] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0113] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in

Claims

1. A method of predicting friction and wear properties of a carbon slide plate of a pantograph, characterized by, The method comprises the following steps: obtaining a set of pantograph carbon strip friction and wear data signals; constructing a pantograph carbon strip friction and wear physical model based on the set of pantograph carbon strip friction and wear data signals; sending the set of data signals to the pantograph carbon strip friction and wear physical model for processing to obtain a set of pantograph carbon strip friction and wear characteristic data; normalizing the set of pantograph carbon strip friction and wear characteristic data and establishing a deep neural network model embedded with the physical model based on the normalized data set, wherein the model comprises a dual loss function of physical information constraint term and data fitting term; optimizing the deep neural network model and establishing an iterative optimization framework with physical interpretation to obtain a prediction model for real-time prediction of pantograph carbon strip friction and wear characteristics; wherein the pantograph carbon strip friction and wear physical model comprises mechanical wear and arc wear terms; the mechanical wear amount is calculated by the following relationship: ； wherein is the mechanical wear; is the friction heat, indicating the energy source for the melting wear, the melting wear being ; is the adhesive wear; is the friction coefficient; is the contact pressure; is the sliding speed; is the running time; is the heat of fusion of the slide plate material; is the wear coefficient; is the sliding distance; is the hardness of the slide plate material; the arc wear amount is calculated by the following relationship: ； wherein, is the amount of arc wear; is the arc energy; and are the voltage and current, respectively, as a function of time, when the arc is occurring; is the sublimation heat of the slide material; is a correction factor; is the arc duration; wherein the set of pantograph carbon strip friction and wear characteristic data is normalized and a deep neural network model embedded with the physical model is established based on the normalized data set, comprising: normalizing the set of pantograph carbon strip friction and wear characteristic data based on a linear normalization method, and dividing the normalized data set into a training set, a validation set and a test set in a ratio of 8:1:1; building a deep neural network model based on the training set, the validation set and the test set, and fusing the physical model to construct a dual loss function comprising a physical information constraint term and a data fitting term; wherein the deep neural network model is optimized, comprising: taking the dual loss function as the optimization target, optimizing the feature extraction and time series prediction ability step by step through a multi-stage hybrid training mechanism, and determining the model parameter feasible region combined with the hyperparameter sensitivity analysis; in the model parameter feasible region, a sparrow search algorithm guided by physical information is used to realize adaptive optimization of hyperparameters to obtain an optimized neural network model; wherein the dual loss function relationship is defined as: ； wherein, are neural network parameters, are input parameters (signal set data), is a data fitting term, wherein is the true wear amount, is the wear amount predicted by the model, , is the number of samples; is a physical information constraint term, and are the calculated mechanical wear amount and arc wear amount, and are the corresponding values predicted by the model; is a weight coefficient for balancing the data fitting and the physical constraint.

2. The method of claim 1, wherein the friction and wear properties of the pantograph carbon strip are predicted by the steps of: based on the set of pantograph carbon strip friction and wear data signals, the pantograph carbon strip friction and wear physical model is constructed, comprising: ​ constructing the mechanical wear term of the physical model based on the pre-designed calculation formula of mechanical wear amount; constructing the arc wear term of the physical model based on the pre-designed calculation formula of arc wear amount.

3. The method of claim 1, wherein the friction and wear properties of the pantograph carbon strip are predicted by the steps of: the data signal set is sent to the pantograph carbon strip friction and wear physical model for processing, comprising: ​ based on the physical constraint space defined by the constructed physical model, and introducing Gaussian noise in the physical constraint space to expand the data signal set to obtain an expanded data signal set; combining the expanded data signal set with the original data signal set, and combining the parameters of the physical model to extract characteristic variables reflecting the friction and wear law to generate an evenly distributed characteristic data set.

4. A system for predicting the friction and wear properties of a pantograph carbon strip, characterized by, The method comprises the following steps: an obtaining unit for obtaining a set of pantograph carbon strip friction and wear data signals; a constructing unit for constructing a pantograph carbon strip friction and wear physical model based on the set of pantograph carbon strip friction and wear data signals; The processing unit is configured to send the data signal set to a pantograph carbon strip friction and wear physical model for processing to obtain a pantograph carbon strip friction and wear characteristic data set; The computing unit is configured to normalize the pantograph carbon strip friction and wear characteristic data set, and establish a deep neural network model embedded with a physical model based on the normalized data set, wherein the model comprises a dual loss function of a physical information constraint term and a data fitting term; The optimization unit is configured to optimize the deep neural network model, and establish an iterative optimization framework with physical interpretation to obtain a prediction model for real-time prediction of pantograph carbon strip friction and wear characteristics; The pantograph carbon strip friction and wear physical model comprises a mechanical wear term and an arc wear term; The mechanical wear amount is calculated by the following relationship: ； wherein is the mechanical wear amount; is the friction heat, indicating the energy source for the melting wear, the melting wear being ; is the adhesive wear; is the friction coefficient; is the contact pressure; is the sliding speed; is the running time; is the heat of fusion of the slide plate material; is the wear coefficient; is the sliding distance; is the hardness of the slide plate material; The arc wear amount is calculated by the following relationship: ； wherein, is the amount of arc wear; is the arc energy; and are the voltage and current, respectively, as a function of time, when the arc is occurring; is the sublimation heat of the slide material; is a correction factor; is the arc duration; The computing unit comprises: The first computing subunit is configured to normalize the pantograph carbon strip friction and wear characteristic data set based on a linear normalization method, and divide the normalized data set into a training set, a validation set and a test set according to a ratio of 8:1:1; The second computing subunit is configured to build a deep neural network model based on the training set, the validation set and the test set, and fuse the physical model to construct a dual loss function comprising a physical information constraint term and a data fitting term; The optimization unit comprises: The first optimization subunit is configured to take the dual loss function as an optimization target, and optimize feature extraction and time series prediction ability in steps through a multi-stage hybrid training mechanism, and determine a model parameter feasible region in combination with hyperparameter sensitivity analysis; The second optimization subunit is configured to realize hyperparameter adaptive optimization in the model parameter feasible region by using a sparrow search algorithm guided by physical information to obtain an optimized neural network model; The dual loss function relationship is defined as: ； wherein, are neural network parameters, are input parameters (signal set data), are data fitting terms, wherein is the true wear amount, is the wear amount predicted by the model, ), is the number of samples; are physical information constraints, and are the calculated mechanical wear amount and arc wear amount, and are the corresponding values predicted by the model; are weight coefficients for balancing the data fitting and the physical constraints.

5. The pantograph carbon strip friction and wear characteristics prediction system of claim 4, wherein, The construction unit comprises: The first construction subunit is configured to construct the mechanical wear term of the physical model based on a pre-designed calculation formula of the mechanical wear amount; The second construction subunit is configured to construct the arc wear term of the physical model based on a pre-designed calculation formula of the arc wear amount.

6. The pantograph carbon strip friction and wear characteristics prediction system of claim 4, wherein, The processing unit comprises: The first processing subunit is configured to expand the data signal set in a physical constraint space defined based on the constructed physical model, and introduce Gaussian noise into the physical constraint space to obtain an expanded data signal set; The second processing subunit is configured to fuse the expanded data signal set with the original data signal set, and extract characteristic variables reflecting friction and wear laws in combination with parameters of the physical model to generate an evenly distributed characteristic data set.

Images