Evaluation method for fretting wear of rail transit electric connector

Abstract

The invention relates to the technical field of fretting wear evaluation of electric connectors, and discloses a method for evaluating fretting wear of a rail transit electric connector, which comprises the following steps of: constructing an Elman model of fretting wear performance degradation of the electric connector based on an Elman neural network, then optimizing by adopting a squirrel search algorithm to obtain an SSA-Elman model of fretting wear performance degradation of the electric connector, and evaluating the fretting wear of the electric connector based on the SSA-Elman model of the fretting wear performance degradation of the electric connector. Then, a Tent chaotic mapping algorithm is introduced to initialize the position of the squirrel population, a Tent-SSA-Elman model is obtained, and finally, the Tent-SSA-Elman model is utilized to evaluate the fretting wear of the rail transit electric connector. Therefore, the evaluation efficiency and precision are better ensured, and the evaluation cost is controlled at the same time.

Description

Technical Field

[0001] The present invention relates to the technical field of wear assessment of electrical components, and particularly to an assessment method for fretting wear of rail transit electrical connectors. Background Art

[0002] As a key component of the vehicle electrical system, the rail transit electrical connector undertakes important functions such as signal transmission and power transmission. During the long-term operation of the train, the contact interface of the electrical connector is prone to fretting wear due to factors such as mechanical vibration, temperature change, and plugging and unplugging operations, that is, the wear phenomenon caused by the relative movement with a small amplitude between the contact surfaces. Fretting wear will significantly increase the contact resistance, cause signal distortion, abnormal temperature rise, and even connection failure, directly affecting the reliability and safety of the rail transit system.

[0003] However, the current assessment methods for fretting wear of rail transit electrical connectors mainly include the following three categories:

[0004] 1. Experimental simulation test method: This method simulates the actual working conditions of the electrical connector (such as amplitude 10 - 100μm, frequency 5 - 50Hz, load 10 - 100N) through a fretting wear testing machine, conducts an accelerated wear experiment under laboratory conditions, measures the change in contact resistance or observes the contact surface morphology (such as scanning electron microscope SEM analysis), and it is generally used for reciprocating friction tests in the ASTM D7904 standard or combined with an electrochemical workstation to synchronously monitor the dynamic fluctuation of the contact resistance. However, this method has a long experimental period (hundreds to thousands of hours), is difficult to meet the rapid assessment requirements, has a high test cost (requiring special equipment and consumables), and cannot achieve real-time monitoring under service conditions.

[0005] 2. Numerical simulation analysis method: This method is based on finite element analysis (FEA) or discrete element method (DEM), establishes a mechanical - wear coupling model for fretting contact, and predicts the wear depth and material loss rate. It generally uses the Archard wear model to calculate the wear amount of the contact surface, and combines multi - physical field simulation to analyze the influence of temperature and vibration on wear. However, this method requires relying on high - precision material parameters (such as friction coefficient, wear coefficient), it is difficult to accurately obtain actual working condition parameters, and the calculation complexity is high, making it difficult to achieve real - time simulation of the dynamic wear process.

[0006] 3. Resistance monitoring method: This method indirectly assesses the degree of fretting wear by online monitoring the change in the contact resistance of the electrical connector. It generally relies on high - precision resistance measurement based on the four - wire method (Kelvin detection), and uses a resistance mutation threshold (such as a 20% increase) to trigger a warning signal. However, in this method, the resistance change is interfered by multiple factors such as environmental temperature and humidity, and the formation of oxide films, resulting in a high false alarm rate, and it is impossible to locate the wear position and difficult to distinguish the contributions of mechanical wear and chemical corrosion.

[0007] Based on the above analysis, it can be seen that the existing evaluation methods are insufficient in accuracy and reliability, do not fully consider the synergistic effect of temperature, vibration and current under complex rail transit conditions, and have problems such as large deviations between the evaluation results and the actual wear mechanism. In view of the above problems, technicians in this field are in urgent need of a high-precision, low-cost fretting wear evaluation method. Summary of the invention

[0008] The purpose of the present invention is to solve the above problems and to design a method for evaluating the micro-motion wear of rail transit electrical connectors.

[0009] To achieve the above-mentioned purpose, the technical solution of the present invention is a method for evaluating the fretting wear of rail transit electrical connectors, the method comprising the following steps:

[0010] Step 1: construct an Elman model of fretting wear performance degradation of electrical connectors based on the Elman neural network;

[0011] Step 2: The squirrel search algorithm is used to optimize the Elman model of the fretting wear performance degradation of the electrical connector, and then the SSA-Elman model of the fretting wear performance degradation of the electrical connector is obtained;

[0012] Step 3, introduce the Tent chaotic mapping algorithm to initialize the position of the squirrel population and obtain the Tent-SSA-Elman model;

[0013] Step 4: Use the Tent-SSA-Elman model to evaluate the micro-motion wear of rail transit electrical connectors.

[0014] The wear characteristic value of the contact is used as the input of the Tent-SSA-Elman model, and the contact resistance value of the connector is used as the output of the Tent-SSA-Elman model.

[0015] The Elman neural network structure in step 1 includes: an input layer, a hidden layer, a receiving layer and an output layer, wherein the receiving layer is used for memory storage and feedback (ie, implementing a feedback mechanism).

[0016] The process of constructing the Elman model of the fretting wear performance degradation of the electrical connector in step 1 is as follows:

[0017] (1) Training samples are selected and the test data are divided according to the vibration direction. For each group of tests, all the monitored wear characteristic values ​​and contact resistance value test data of the three contact parts of a typical test piece are selected as sample data, that is, there are 9×3×20 groups of sample data for each vibration direction, totaling 540 groups, and 1080 groups of sample data for the two vibration directions. Since the contact resistance values ​​and wear characteristic values ​​of each group of test pieces are highly correlated, in order to make the constructed model more adaptable, this application disrupts the test groups and time series relationship of the sample data and establishes a sample database.

[0018] Before training the samples, this application normalizes the sample data:

[0019]

[0020] In the formula, x * is the normalized wear debris feature value; x is the wear debris feature value of the sample data; the application divides the normalized sample data into training samples and test samples in a ratio of 8:1;

[0021] (2) Determine the model structure. The nonlinear function of the Elman neural network is:

[0022] x(k)=f(ω1x c (k)+ω2u(k-1)+θ1) (2)

[0023] y(k)=g(ω3x(k)+θ2) (3)

[0024] x c (k) = x(k-1) (4)

[0025] Where x(k) is the output of the hidden layer at the kth cycle; f(*) is the transfer function of the hidden layer; ω1 and ω2 are the weights from the input layer to the hidden layer and from the receiving layer to the hidden layer, respectively; x c (k) is the output of the receiving layer in the kth cycle, corresponding to the output of the hidden layer in the k-1th cycle; u(k-1) is the input of the input layer; θ1 and θ2 are the thresholds of the hidden layer and the output layer respectively; ω3 is the weight from the hidden layer to the output layer; y(k) is the output of the output layer; g(*) is the transfer function of the output layer.

[0026] Among them, the input layer node and the output layer node are both 1 node, and the number of hidden layer nodes is calculated by formula (2) and the value range is 3-11; after simulation analysis, when the number of hidden layer nodes is 10, the performance of the Elman model is the best, so the number of hidden layer nodes is 10. The main function of the successor layer is to memorize the output value of the hidden layer at the previous time point, save the output value of the hidden layer, splice it with the input value of the hidden layer, and enter the hidden layer again. Therefore, the number of nodes in the successor layer should be consistent with the number of nodes in the hidden layer, and the number of nodes in the successor layer is 10. The transfer function of the hidden layer is the sigmoid function; the transfer function of the output layer is the purelin function.

[0027] The calculation formula for the number of hidden layer nodes is:

[0028]

[0029] In the formula, m is the number of nodes in the input layer; n is the number of nodes in the output layer; d is an integer between 1 and 10.

[0030] The number of nodes l in the hidden layer of the model constructed in this application ranges from 3 to 11. After multiple simulation comparisons, the model has the best simulation effect when l=11. In addition, this application selects the tansig function and the linear purelin function with smaller errors as the transfer function between the input layer and the hidden layer and the transfer function between the hidden layer and the output layer, respectively.

[0031] (3) Set the initial parameters, randomly assign weights and thresholds between the output layer and the hidden layer, and between the hidden layer and the output layer, set the maximum number of training times to 1000, the learning rate to 0.01, and the expected error to 10 -4 ;

[0032] (4) Data forward propagation: input a set of sample data (x, y), where x is the wear debris characteristic value and y is the contact resistance value;

[0033] (5) Calculate the input and output of the hidden layer and output layer:

[0034] The input and output of the jth node in the hidden layer are:

[0035]

[0036] In the formula, f(*) and are the transfer function and weights between the input layer and the hidden layer respectively; is the threshold of the hidden layer;

[0037] The input and output of the output layer are:

[0038]

[0039] Y = g(S - θ 2 ) (9)

[0040] where g(*) and are the transfer function and weight between the hidden layer and the output layer respectively; θ 2 is the threshold of the output layer;

[0041] (6) Backpropagation of error, calculate the node errors of the output layer and the hidden layer;

[0042] The node error of the output layer is:

[0043] δ 2 = Y(1 - Y)(y - Y) (10)

[0044] The error of the j-th node in the hidden layer is:

[0045]

[0046] (7) Update the weights and thresholds. The weight update is:

[0047]

[0048] where β is the learning rate;

[0049] The threshold update is:

[0050]

[0051] θ 2 (t + 1)= θ 2 (t)+ γδ 2 (15)

[0052] where γ is the learning rate;

[0053] (8) Training of the model, repeat steps (4)-(7), perform global error calculation, stop when the global error is less than the desired error or the number of training times has reached the maximum number of training times, and save the model;

[0054] (9) Load the test sample data, test the saved model, and obtain the prediction result.

[0055] In the squirrel search process of the squirrel search algorithm in the second step, there are the following 4 assumptions:

[0056] (1) There are n squirrels in the forest, one on each tree;

[0057] (2) Each squirrel searches for food on its own and optimally utilizes the existing food;

[0058] (3) There are only three types of trees in the forest: hickory trees (the best food), oak trees (average food), and common trees (no food).

[0059] (4) There are only four food-bearing trees in the forest, including three oak trees (sub-optimal solution) and one hickory tree (optimal solution).

[0060] The main steps for constructing the SSA-Elman model for the micro-motion wear performance degradation of the electrical connector in the second step are as follows:

[0061] (1) Initialize the population positions, set the number of squirrel populations Q = 50, the maximum number of evolutionary iterations t m = 500, and the probability P dp of the existence of predators = 0.1;

[0062] (2) Calculate the fitness F and perform ascending sorting;

[0063]

[0064] In the formula, y i and are respectively the measured value and the predicted value of the training sample of the contact resistance value; y j and are respectively the measured value and the predicted value of the test sample of the contact resistance value.

[0065] (3) Update the squirrel positions;

[0066] When the squirrel moves from an oak tree to a hickory tree, the position is updated as:

[0067]

[0068] In the formula, FS at and FS ht are respectively the positions of the squirrel at the oak tree and the hickory tree; t is the number of iterations; d g is the sliding distance of the squirrel; G c is the sliding constant, G c = 1.9; R1 is a random number within [0,1]; P dp is the probability of the existence of predators. If R > P dp , then the squirrel is safe, otherwise the squirrel is in danger and needs to move randomly;

[0069] When the squirrel moves from a common tree to an oak tree, the position is updated as:

[0070]

[0071] In the formula, FS nt is the position of the squirrel reaching the common tree; R2 is a random number with a value range within [0,1];

[0072] When the squirrel moves from an ordinary tree to a pecan tree, the position is updated as follows:

[0073]

[0074] In the formula, R3 is a random number, and its value range is within [0, 1];

[0075] (4) Seasonal monitoring and random relocation at the end of winter;

[0076] In different seasons, the foraging activity of squirrels is different. This algorithm uses seasonal monitoring to jump out of the local optimal solution;

[0077] Seasonal variable S c is:

[0078]

[0079] The seasonal change condition is:

[0080]

[0081] In the formula, t m is the maximum number of iterations;

[0082] If it indicates the end of winter; the position of the squirrel on the ordinary tree and without finding food is randomly updated as:

[0083]

[0084] In the formula, r a and r b are both random numbers that conform to the normal distribution within [0, 1]; β = 1.5; the calculation formula of σ is as follows:

[0085]

[0086] (5) When the squirrel is at the position of the pecan (i.e., when the global optimal solution is reached) or when the maximum number of iterations is reached, the algorithm is terminated; otherwise, repeat steps (2)-(4);

[0087] (6) Input the obtained optimal weight and threshold into the Elman model for model training.

[0088] The Tent-SSA-Elman model obtained by initializing the position of the squirrel population using the Tent chaos mapping algorithm in step three is:

[0089]

[0090] In the formula, δ = 0.5; X i is a randomly generated j-dimensional vector, and its value range is [0, 1].

[0091] Compared with the prior art, the present invention has the following beneficial effects:

[0092] 1. Based on the Elman model of fretting wear performance degradation, the Tent-SSA-Elman model is constructed in this application. Through the Tent-SSA-Elman model, the prediction result has higher accuracy, better fitting degree, smaller relative error of the prediction result, and has a more ideal prediction effect;

[0093] 2. The MAPE of the Tent-SSA-Elman model adopted in this application is only 2.58%. Compared with the Elman model and the SSA-Elman model, the MAPE of the Tent-SSA-Elman model decreases by 62.5% and 30.46% (longitudinal vibration samples), and the overall deviation degree between the predicted value and the measured value is smaller. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 is a flowchart of an evaluation method for fretting wear of a rail transit electrical connector according to the present invention;

[0095] Figure 2 is a schematic diagram of the structure of the Elman neural network according to the present invention;

[0096] Figure 3 is a flowchart of the construction of the Elman model for fretting wear performance degradation according to the present invention;

[0097] Figure 4 is a prediction result diagram of the Elman model for fretting wear performance degradation according to the present invention, where: (a) is longitudinal vibration, and (b) is lateral vibration;

[0098] Figure 5 is a flowchart of the construction of the SSA-Elman model for fretting wear performance degradation according to the present invention;

[0099] Figure 6 is a prediction result diagram of the SSA-Elman model for fretting wear performance degradation according to the present invention, where: (a) is longitudinal vibration; (b) is lateral vibration;

[0100] Figure 7 is a prediction result diagram of the Tent-SSA-Elman model for fretting wear performance degradation: (a) longitudinal vibration; (b) lateral vibration;

[0101] Figure 8 is a comparison diagram of the prediction results of the three fretting wear performance degradation models according to the present invention, where: (a) is longitudinal vibration; (b) is lateral vibration;

[0102] Figure 9It is the error analysis table of the fretting wear performance degradation Elman model described in the present invention;

[0103] Figure 10 It is the error analysis table of the fretting wear performance degradation SSA-Elman model described in the present invention;

[0104] Figure 11 It is the error analysis table of the fretting wear performance degradation Tent-SSA-Elman model described in the present invention;

[0105] Figure 12 It is the error comparison table of the three fretting wear performance degradation models described in the present invention. Specific embodiments

[0106] The present invention will be specifically described below in conjunction with the accompanying drawings, as Figures 1-12 shown;

[0107] The Elman neural network is a globally feed-forward and locally feedback-recursive neural network. Due to the existence of the feedback mechanism, this model can utilize the input information multiple times, so it can effectively handle non-linear dynamic problems. The Elman neural network structure includes an input layer, a hidden layer, a context layer, and an output layer. Compared with the BP neural network, the Elman neural network has an additional context layer for memory storage and feedback, and its structural schematic diagram is as Figure 2 shown.

[0108] The construction process of the fretting wear performance degradation Elman model of the electrical connector is as follows:

[0109] (1) Select training samples, divide the test data according to the vibration direction, and use all the monitored debris characteristic values and contact resistance values of 3 contacts of a typical test sample in each group of tests as sample data, that is, there are 540 groups of sample data of 9×3×20 for each vibration direction, and a total of 1080 groups of sample data for two vibration directions. Since the correlation between the contact resistance values and debris characteristic values of the test samples in each group is relatively strong, in order to make the constructed model have stronger adaptability, this application shuffles the test groups and time series relationships of the sample data to establish a sample database.

[0110] Before training the samples, this application performs normalization processing on the sample data:

[0111]

[0112] In the formula, x * is the debris characteristic value after normalization; x is the debris characteristic value of the sample data; this application divides the normalized sample data into training samples and test samples at a ratio of 8:1;

[0113] (2) Determine the model structure. The nonlinear function of the Elman neural network is as follows:

[0114] x(k) = f(ω1x c (k) + ω2u(k - 1) + θ1) (2)

[0115] y(k) = g(ω3x(k) + θ2) (3)

[0116] x c (k) = x(k - 1) (4)

[0117] In the formula, x(k) is the output of the hidden layer at the k-th cycle; f(*) is the transfer function of the hidden layer; ω1 and ω2 are the weights from the input layer to the hidden layer and from the connection layer to the hidden layer respectively; x c (k) is the output of the connection layer at the k-th cycle, corresponding to the output of the hidden layer at the (k - 1)-th cycle; u(k - 1) is the input of the input layer; θ1 and θ2 are the thresholds of the hidden layer and the output layer respectively; ω3 is the weight from the hidden layer to the output layer; y(k) is the output of the output layer; g(*) is the transfer function of the output layer.

[0118] Among them, both the input layer nodes and the output layer nodes are 1 node. The number of hidden layer nodes is calculated by formula (2) and the value range is 3 - 11; through simulation analysis, when the number of hidden layer nodes is 10, the Elman model performs the best. Therefore, the number of hidden layer nodes is taken as 10. The main function of the connection layer is to memorize the output value of the hidden layer at the previous time point, save the output value of the hidden layer, splice it with the input value of the hidden layer, and then enter it into the hidden layer again. Therefore, the number of connection layer nodes should be the same as the number of hidden layer nodes, and the number of connection layer nodes is taken as 10. The transfer function of the hidden layer is the sigmoid function; the transfer function of the output layer is the purelin function.

[0119] The calculation formula for the number of hidden layer nodes is:

[0120]

[0121] In the formula, m is the number of nodes in the input layer; n is the number of nodes in the output layer; d takes an integer between 1 and 10.

[0122] The number of hidden layer nodes l of the model constructed in this application has a value range of 3 - 11. After multiple simulation comparisons, when l = 11, the model has the best simulation effect. In addition, this application selects the tansig function with a smaller error and the linear purelin function as the transfer functions between the input layer and the hidden layer and between the hidden layer and the output layer respectively.

[0123] (3) Set the initial parameters, randomly assign the weights and thresholds between the output layer and the hidden layer, and between the hidden layer and the output layer. Set the maximum number of training times to 1000, the learning rate to 0.01, and the expected error to 10 -4 ;

[0124] (4) Forward propagation of data. Input a set of sample data (x, y), where x is the debris feature value and y is the contact resistance value;

[0125] (5) Calculate the inputs and outputs of the hidden layer and the output layer:

[0126] The input and output of the j-th node in the hidden layer are respectively:

[0127]

[0128] In the formula, f(*) and are respectively the transfer function and the weight between the input layer and the hidden layer; is the threshold of the hidden layer;

[0129] The input and output of the output layer are respectively:

[0130]

[0131] Y = g(S - θ 2 ) (9)

[0132] In the formula, g(*) and are respectively the transfer function and the weight between the hidden layer and the output layer; θ 2 is the threshold of the output layer;

[0133] (6) Backward propagation of error, calculate the node errors of the output layer and the hidden layer;

[0134] The node error of the output layer is:

[0135] δ 2 = Y(1 - Y)(y - Y) (10)

[0136] The error of the j-th node in the hidden layer is:

[0137]

[0138] (7) Update the weights and thresholds. The weights are updated as:

[0139]

[0140] In the formula, β is the learning rate;

[0141] The thresholds are updated as:

[0142]

[0143] θ 2 (t+1)=θ 2 (t)+γδ 2 (15)

[0144] In the formula, γ is the learning rate;

[0145] (8) Model training: repeat steps (4) to (7) to calculate the global error. When the global error is less than the expected error or the number of training times has reached the maximum number of training times, stop and save the model.

[0146] (9) Load test sample data, test the saved model, and obtain prediction results;

[0147] It should be noted that in step (2) of constructing the Elman model, the input layer node and the output layer node are both 1 node, and the number of hidden layer nodes calculated by formula (2) ranges from 3 to 11. According to simulation analysis, the Elman model performs best when the number of hidden layer nodes is 10, so the number of hidden layer nodes is 10. The main function of the receiving layer is to memorize the output value of the hidden layer at the previous time point, save the output value of the hidden layer, splice it with the input value of the hidden layer, and enter the hidden layer again. Therefore, the number of receiving layer nodes should be consistent with the number of hidden layer nodes. In this application, the number of receiving layer nodes is 10. The transfer function of the hidden layer is selected as the sigmoid function; the transfer function of the output layer is the purelin function.

[0148] The basic process of constructing the Elman model of the fretting wear performance degradation of electrical connectors in this application is as follows: Figure 3 The prediction results of the Elman model for the fretting wear performance degradation of electrical connectors are shown in Figure 4 The error analysis of the Elman model prediction results of the fretting wear performance degradation of electrical connectors is shown in Figure 9 shown.

[0149] Depend on Figure 4 and Figure 9 The analysis shows that the Elman model of fretting wear performance degradation of electrical connectors has higher prediction accuracy, better fitting effect and smaller prediction relative error for the sample data of longitudinal vibration.

[0150] However, the Elman model itself still has the disadvantages of slow convergence, easy to fall into local optimal solutions and difficult to find the global optimal solution. In order to improve the training speed of the Elman model and avoid falling into local optimal solutions, this application uses the squirrel search algorithm with strong optimization ability and fast convergence to optimize the Elman model.

[0151] The Squirrel Search Algorithm (SSA for short) is a nature-inspired optimization algorithm. Its global search process simulates the process of squirrels foraging among different trees and avoiding predators.

[0152] During the squirrel search process, there are the following four assumptions:

[0153] (1) There are n squirrels in the forest, one on each tree.

[0154] (2) Each squirrel searches for food on its own and optimally utilizes the existing food.

[0155] (3) There are only three types of trees in the forest: pecan trees (the best food), oak trees (average food), and ordinary trees (no food).

[0156] (4) There are only four food-bearing trees in the forest, including three oak trees (sub-optimal solutions) and one pecan tree (the optimal solution).

[0157] Based on these assumptions, the main steps for this application to construct the SSA-Elman model for the fretting wear performance degradation of electrical connectors are as follows:

[0158] (1) Initialize the population positions. Set the number of squirrels in the population Q = 50, the maximum number of evolutionary iterations t m = 500, and the probability P dp = 0.1 that a predator exists.

[0159] (2) Calculate the fitness F and sort it in ascending order.

[0160]

[0161] In the formula, y i and are respectively the measured value and the predicted value of the training sample of the contact resistance value; y j and are respectively the measured value and the predicted value of the test sample of the contact resistance value.

[0162] (3) Update the squirrel positions.

[0163] When the squirrel moves from an oak tree to a pecan tree, the position is updated as:

[0164]

[0165] In the formula, FS at and FS ht are respectively the positions of the squirrel on the oak tree and the pecan tree; t is the iteration number; d g is the sliding distance of the squirrel; G c is the sliding constant, G c= 1.9; R1 is a random number within [0, 1]; P dp is the probability of the presence of a predator. If R > P dp , then the squirrel is safe; otherwise, the squirrel is in danger and needs to move randomly.

[0166] When the squirrel moves from a common tree to an oak tree, the position is updated as follows:

[0167]

[0168] where FS nt is the position where the squirrel reaches the common tree; R2 is a random number with a value range within [0, 1].

[0169] When the squirrel moves from a common tree to a hickory tree, the position is updated as follows:

[0170]

[0171] where R3 is a random number with a value range within [0, 1].

[0172] (4) Seasonal monitoring and random relocation at the end of winter.

[0173] In different seasons, the foraging activity of squirrels is different. This algorithm uses seasonal monitoring to jump out of local optimal solutions.

[0174] The seasonal variable S c is as follows:

[0175]

[0176] The seasonal change condition is:

[0177]

[0178] where t m is the maximum number of iterations.

[0179] If it indicates the end of winter. The positions of squirrels on common trees that cannot find food are randomly updated as follows:

[0180]

[0181] where r a and r b are both random numbers that conform to the normal distribution within [0, 1]; β = 1.5; the calculation formula of σ is as follows:

[0182]

[0183] (5) When the squirrel is at the position of the pecan (i.e., when the global optimal solution is reached) or when the maximum number of iterations is reached, terminate the algorithm; otherwise, repeat steps (2)-(4).

[0184] (6) Input the obtained optimal weights and thresholds into the Elman model for model training.

[0185] The basic process of constructing the SSA-Elman model for the micro-motion wear performance degradation of the electrical connector is as Figure 5 shown. The prediction results obtained by the SSA-Elman model for the micro-motion wear performance degradation of the electrical connector are as Figure 6 shown. The error analysis of the prediction results of the SSA-Elman model for the micro-motion wear performance degradation of the electrical connector is as Figure 10 shown.

[0186] From Figure 6 and Figure 10 analysis, it can be seen that the MAE, RMSE, and MAPE of the SSA-Elman model for the micro-motion wear performance degradation of the electrical connector are all smaller than those of the Elman model for the micro-motion wear performance degradation. For the prediction of the longitudinal vibration sample data, the MAE of the SSA-Elman model is reduced by 0.0253 compared with the Elman model, a decrease of 483%; the RMSE is reduced by 0.0266, a decrease of 46.02%; the MAPE is reduced by 3.17%, a decrease of 46.08%. For the prediction of the transverse vibration sample data, the MAE of the SSA-Elman model is reduced by 0.0208, a decrease of 39.69%; the RMSE is reduced by 0.0201, a decrease of 36.02%; the MAPE is reduced by 2.97%, a decrease of 39.86%. In summary, the SSA-Elman model for the micro-motion wear performance degradation of the electrical connector has a better prediction effect than the Elman model, and the effect is more obvious for the longitudinal vibration sample data.

[0187] The Tent chaotic map has advantages such as uniform distribution and strong ergodicity. In this application, the Tent chaotic map algorithm is introduced to initialize the positions of the squirrel population to ensure the diversity of its population, which can improve the premature convergence of the algorithm and enhance its optimization efficiency.

[0188] The model for initializing the positions of the squirrel population by the Tent chaotic map is:

[0189]

[0190] In the formula, δ = 0.5; X i is a randomly generated j-dimensional vector, and its value range is [0, 1].

[0191] The prediction results obtained by the Tent-SSA-Elman model for the micro-motion wear performance degradation of the electrical connector are as Figure 7As shown, the error analysis is as Figure 11 shown.

[0192] From Figure 7 and Figure 11 analysis, it can be seen that the MAE, RMSE, and MAPE of the Tent-SSA-Elman model for the degradation of the fretting wear performance of the electrical connector are all smaller than those of the Elman model for the degradation of the fretting wear performance. For the prediction of the longitudinal vibration sample data, the MAE of the Tent-SSA-Elman model is reduced by 0.0344 compared with the Elman model, a decrease of 62.32%; the RMSE is reduced by 0.0342, a decrease of 59.17%; the MAPE is reduced by 4.3%, a decrease of 62.5%. For the prediction of the transverse vibration sample data, the MAE of the Tent-SSA-Elman model is reduced by 0.0302, a decrease of 57.63%; the RMSE is reduced by 0.0318, a decrease of 56.99%; the MAPE is reduced by 4.28%, a decrease of 57.45%. In summary, the Tent-SSA-Elman model for the degradation of the fretting wear performance of the electrical connector has higher prediction accuracy, better fitting effect, smaller prediction relative error, and better prediction effect than the Elman model.

[0193] The comparison chart of the prediction results of the Elman, SSA-Elman, and Tent-SSA-Elman models for the degradation of the fretting wear performance of the electrical connector is as Figure 8 shown, and the error analysis of the three fretting wear performance degradation models is as Figure 12 shown.

[0194] From Figure 8 and Figure 12Analysis shows that the MAE, RMSE, and MAPE of the Tent-SSA-Elman model for the fretting wear performance degradation of the electrical connector are the smallest among the three fretting wear performance degradation models. For the prediction of longitudinal vibration sample data, the MAE of the Tent-SSA-Elman model is reduced by 0.0091 compared with the SSA-Elman model, a decrease of 30.44%; the RMSE is reduced by 0.0076, a decrease of 24.36%; the MAPE is reduced by 1.13%, a decrease of 30.46%. For the prediction of transverse vibration sample data, the MAE of the Tent-SSA-Elman model is reduced by 0.0094, a decrease of 29.75%; the RMSE is reduced by 0.0117, a decrease of 32.77%; the MAPE is reduced by 1.31%, a decrease of 29.24%. The mean absolute percentage error of the Tent-SSA-Elman model is only 2.58%, indicating that the overall deviation between the predicted value and the measured value of this model is small. To sum up, the Tent-SSA-Elman model for the fretting wear performance degradation of the electrical connector is the most excellent among the three models in terms of prediction accuracy, fitting effect, prediction relative error, and prediction effect. Therefore, the Tent-SSA-Elman model for the fretting wear performance degradation of the electrical connector is selected as the final fretting wear performance degradation prediction model.

[0195] The above technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention. Some changes that may be made to some parts by those skilled in the art of the present technology all reflect the principles of the present invention and are within the protection scope of the present invention.

Claims

1. A method for evaluating fretting wear of rail transit electrical connectors, characterized in that: The method comprises the following steps: Step 1: construct an Elman model of fretting wear performance degradation of electrical connectors based on the Elman neural network; Step 2, using the squirrel search algorithm to optimize the Elman model of the fretting wear performance degradation of the electrical connector, and obtaining the SSA-Elman model of the fretting wear performance degradation of the electrical connector; Step 3, introduce the Tent chaotic mapping algorithm to initialize the position of the squirrel population and obtain the Tent-SSA-Elman model; Step 4: Use the Tent-SSA-Elman model to evaluate the micro-motion wear of rail transit electrical connectors.

2. The method for evaluating fretting wear of rail transit electrical connectors according to claim 1, characterized in that: The Elman neural network structure in step 1 includes: an input layer, a hidden layer, a receiving layer and an output layer.

3. The method for evaluating fretting wear of rail transit electrical connectors according to claim 2, characterized in that: The process of constructing the Elman model of the fretting wear performance degradation of the electrical connector in step 1 is: (1) Select training samples to establish a sample database and normalize the sample data: In the formula, x * is the normalized wear debris characteristic value; x is the wear debris characteristic value of the sample data; (2) Determine the model structure. The nonlinear function of the Elman neural network is: x(k)=f(ω1x c (k)+ω2u(k-1)+θ1)(2) y(k)=g(ω3x(k)+θ2)(3) x c (k)=x(k-1)(4) Where x(k) is the output of the hidden layer at the kth cycle; f(*) is the transfer function of the hidden layer; ω1 and ω2 are the weights from the input layer to the hidden layer and from the receiving layer to the hidden layer, respectively; x c (k) is the output of the receiving layer in the kth cycle, corresponding to the output of the hidden layer in the k-1th cycle; u(k-1) is the input of the input layer; θ1 and θ2 are the thresholds of the hidden layer and the output layer respectively; ω3 is the weight from the hidden layer to the output layer; y(k) is the output of the output layer; g(*) is the transfer function of the output layer; The calculation formula for the number of hidden layer nodes is: In the formula, m is the number of nodes in the input layer; n is the number of nodes in the output layer; d is an integer between 1 and 10; (3) Set the initial parameters, randomly assign weights and thresholds between the output layer and the hidden layer, and between the hidden layer and the output layer, set the maximum number of training times to 1000, the learning rate to 0.01, and the expected error to 10 -4 ; (4) Data forward propagation: input a set of sample data (x, y), where x is the wear debris characteristic value and y is the contact resistance value; (5) Calculate the input and output of the hidden layer and output layer: The input and output of the jth node in the hidden layer are: In the formula, S j is the input value of the jth node in the hidden layer, O j is the output value of the jth node in the hidden layer, f(*) and are the transfer function and weights between the input layer and the hidden layer respectively; is the threshold of the hidden layer; The input and output of the output layer are: Y=g(S-θ 2 )(9) In the formula, S is the input value of the output layer, Y is the output value of the output layer, g(*) and are the transfer function and weight between the hidden layer and the output layer respectively; θ 2 is the threshold of the output layer; (6) Error back propagation, calculating the node errors of the output layer and the hidden layer; The output layer node error is: δ 2 =Y(1-Y)(yY)(10) The error of the jth node in the hidden layer is: (7) Update weights and thresholds. The weights are updated as follows: Where β is the learning rate; The threshold is updated as: In the formula, γ is the learning rate; (8) Model training: repeat steps (4) to (7) to calculate the global error. When the global error is less than the expected error or the number of training times has reached the maximum number of training times, stop and save the model. (9) Load the test sample data, test the saved model, and obtain the prediction results.

4. The method for evaluating fretting wear of rail transit electrical connectors according to claim 1, characterized in that: The main steps of constructing the SSA-Elman model of the fretting wear performance degradation of the electrical connector in step 2 are as follows: (1) Initialize the population position, set the squirrel population number Q = 50, and the maximum number of evolutionary iterations t m =500, the probability of the predator existing is P dp =0.1; (2) Calculate the fitness F and sort it in ascending order; In the formula, yi and are the measured value and predicted value of the contact resistance training sample respectively; y j and They are the measured value and predicted value of the contact resistance test sample respectively; (3) Update the squirrel's position; When the squirrel moves from the oak tree to the hickory tree, the position is updated to: Where, FS at and FS ht are the positions of the squirrel in the oak tree and the hickory tree respectively; t is the number of iterations; d g is the sliding distance of the squirrel; G c is the glide constant, G c =1.9; R1 is a random number in [0,1]; P dp is the probability of the predator existing, if R>P dp , then the squirrel is safe, otherwise the squirrel is in danger and needs to wander randomly; When the squirrel moves from the normal tree to the oak tree, the position is updated as follows: Where, FS nt is the position of the squirrel when it reaches the ordinary tree; R2 is a random number in the range of [0,1]; When the squirrel moves from the normal tree to the hickory tree, the position is updated to: Where R3 is a random number, and its value range is [0,1]; (4) seasonal monitoring and random relocation in late winter; Use seasonal monitoring to escape from local optimal solutions; Then the seasonal variable S c for: The seasonal change conditions are: Where, t m is the maximum number of iterations; if Indicates the end of winter; the positions of squirrels on ordinary trees that cannot find food are randomly updated to: In the formula, r a and r b All are random numbers that conform to the normal distribution within [0,1]; β = 1.5; the calculation formula of σ is as follows: (5) When the squirrel is at the pecan position or the maximum number of iterations is reached, terminate the algorithm, otherwise repeat steps (2)-(4); (6) The obtained optimal weights and thresholds are input into the Elman model for model training.

5. The method for evaluating fretting wear of rail transit electrical connectors according to claim 1, characterized in that: The Tent-SSA-Elman model obtained by initializing the position of the squirrel population using the Tent chaotic mapping algorithm in step 3 is: Where, δ = 0.5; X i is a randomly generated j-dimensional vector with a value range of [0,1].

Images