An analytical algorithm for predicting the remaining life of a wheel

Abstract

The present invention discloses an analysis algorithm for predicting the remaining life of a wheel, including: firstly, establishing a prediction model for the wheel diameter and the wear of the wheel flange thickness; secondly, establishing a wheel set turning model; then, establishing a train set rating model; and finally, based on the wheel wear and turning models, establishing a remaining life prediction model. The present invention belongs to the technical field of wheel life prediction algorithms, and specifically is an algorithm for predicting the remaining life of train wheels based on multi-source data. By integrating machine tool detection, turning, and condition monitoring data, a wheel wear and turning model is established, so as to accurately estimate the remaining life of the wheel and the evolution of parameters, in order to reasonably arrange turning operations and reduce the maintenance cost of the train set for the analysis algorithm of predicting the remaining life of the wheel.

Description

Technical Field

[0001] The present invention belongs to the technical field of wheel life prediction algorithms, and specifically refers to an analysis algorithm for predicting the remaining life of wheels. Background Art

[0002] As an important component of railway vehicles, the health status of wheels will directly affect the operation quality and driving safety of vehicles. Predicting the wear and remaining life of wheels can ensure the safe operation of trains, arrange the wheel turning plan in advance and make accurate wheel turning decisions, extend the service life of wheel sets and thus reduce costs.

[0003] At present, the methods for predicting the remaining life of wheel sets can be mainly divided into two points: one is the method based on physical models, and the other is the method based on mathematical statistics. Method 1, the method based on physical models, according to the dynamic model and related physical quantities, uses simulation software to establish a model, simulates the real operation conditions of the wheel set to obtain wear and life data, and compares with the actual data to verify its accuracy. Method 2, in the method for predicting the remaining life based on mathematical statistics, according to the characteristics of the degradation trajectory of wheel performance parameters, it can be assumed that the degradation path satisfies a certain regression equation or stochastic process, and then a mathematical model representing the degradation process is constructed to obtain the wear and turning rules of the wheels, so as to conduct the prediction and analysis of the remaining life.

[0004] However, since the degradation of wheel performance and geometric parameters is an extremely complex process, it is difficult to establish an accurate physical model to reflect the real working conditions of the wheels. With the continuous improvement and perfection of the means of wheel management and monitoring, the state maintenance data during the operation of wheels is continuously enriched, making the data-based method become a research hotspot. Summary of the Invention

[0005] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an algorithm for predicting the remaining life of train wheels based on multi-source data. By integrating machine tool detection, wheel turning, and condition monitoring data, a wheel wear and turning model is established, so as to accurately estimate the remaining life of the wheels and the evolution of parameters, in order to reasonably arrange wheel turning operations and reduce the maintenance cost of the train set.

[0006] The technical solution adopted by the present invention is as follows: The analysis algorithm for predicting the remaining life of wheels in this solution includes four parts:

[0007] Content 1: First, establish a prediction model for wheel diameter and wheel flange thickness wear, and the steps are as follows:

[0008] S11: Extract machine tool wheel turning data; calculate parameters such as the wear amount of wheel diameter and wheel flange thickness during two adjacent wheel turnings of the wheel;

[0009] S12: Establish a BP neural network model for training;

[0010] S13. Optimize the neural network parameters using the genetic algorithm to obtain the optimal model;

[0011] S14. Save the trained model for future reference.

[0012] Content Two: Establish a wheel profile turning model, and the steps are as follows:

[0013] S21. Extract the train turning data; calculate the wheel diameters before and after turning and the change amount of the flange thickness;

[0014] S22. Analyze the turning data to establish the relationship between the turning amount and the flange thickness compensation;

[0015] S23. Analyze the selection of the flange thickness template and establish a flange thickness template selection model;

[0016] S24. Establish a wheel life prediction turning model.

[0017] Content Three: Establish a car body rating model, and the steps are as follows:

[0018] S31. Determine the wheel polygon rating standard according to the wheel polygon data;

[0019] S32. Determine the wheel equivalent taper rating standard according to the wheel equivalent taper data;

[0020] S33. Determine the head car defect rating standard according to the number of head car defects;

[0021] S34. Conduct a comprehensive rating of the train based on the above three rating standards and determine the next turning cycle.

[0022] Content Four: Based on the wheel wear and turning model, establish a remaining life prediction model, and the algorithm steps are as follows:

[0023] S41. Search for parameters such as the date of the last wheel turning, the running mileage, the geometric dimensions after turning, and the status of the wheel;

[0024] S42. Determine the cycle mileage, and judge whether the wheel reaches the limit within the cycle. If it reaches the limit, enter S35; otherwise, enter S33;

[0025] S43. Calculate the wheel diameters and flange thickness values before and after the next turning, and judge whether it exceeds the limit after turning. If it exceeds the limit, it is considered that the operation repair reaches the limit, and enter S45; otherwise, enter S42 for iterative loop until it reaches the limit;

[0026] S44. When the wheel reaches the limit, output data such as the time when it reaches the limit, the running mileage, the parameter values of each item, and the predicted turning situation.

[0027] An analysis algorithm for predicting the remaining life of a wheel in this solution. The beneficial effects achieved by the present invention using the above solution are as follows:

[0028] 1. Based on the wheel turning data, a BP neural network model optimized by a genetic algorithm is established to accurately predict the wheel diameter and the wear of the wheel flange thickness; based on the historical turning data, a method for establishing a wheel turning model is successfully established, including the selection of the turning wheel flange thickness template, the determination of the wheel flange thickness compensation ratio, and the determination of the wheel diameter turning amount; based on the wheel polygon, equivalent taper, and head car defects, a comprehensive rating method for the car body is successfully established to guide wheel turning and for predicting the remaining life.

[0029] 2. Based on data such as wheel turning, LY detection, equivalent taper, polygon, and head car defects, the advantages of each data are fully utilized, and the value of each data is fully explored; a wheel tread wear model and a wheel turning model are established using the turning data, a wheel state rating and a turning cycle mileage determination model are established using the equivalent taper, polygon, and head car defect data, and each model is integrated and a wheel remaining life prediction model is established with the assistance of LY detection data. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is the wheel life prediction flow chart of this solution;

[0031] Figure 2 is the BP neural network structure diagram in this solution;

[0032] Figure 3 is the schematic diagram of the fitness curve in this solution;

[0033] Figure 4 is the evolution law diagram of the wheel diameter and wheel flange thickness of the head and tail cars in this embodiment;

[0034] Figure 5 is the evolution law diagram of the wheel diameter and wheel flange thickness of the intermediate motor cars in this embodiment;

[0035] Figure 6 is the evolution law diagram of the wheel diameter and wheel flange thickness of the intermediate trailers in this embodiment.

[0036] Among them, Figure 4 、 Figure 5 、 Figure 6 in the upper broken line is the schematic diagram of the relationship between the running mileage and the wheel diameter value, and the lower broken line is the schematic diagram of the relationship between the running mileage and the wheel flange thickness.

[0037] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0039] As Figures 1-6 shown, an analysis algorithm for predicting the remaining life of a wheel of the present invention is specifically implemented as follows:

[0040] Referring to the attached Figure 1 drawings, this method includes four parts. First, a wheel tread wear prediction model is established based on the wheel turning data; second, a wheel turning model is established based on the turning data; third, the wheel operation and repair cycle is determined according to the wheel condition detection data; finally, a wheel remaining life prediction model is established based on the wheel tread wear model and the turning model. The specific algorithm is as follows:

[0041] S1: Model the wheel tread wear, and the specific steps are as follows:

[0042] S11: Input the vehicle type to be modeled to obtain the corresponding maintenance data, advanced maintenance plan, and resume. If the number of wheels turned in the same train set within 7 days is greater than 50%, it is considered that a full train turning has occurred; subtract the wheel diameter and flange thickness values before the next full train turning from the wheel diameter and flange thickness values after the last full train turning of the same wheel to obtain the wear amounts of the wheel diameter and flange thickness during this operation cycle. Corresponding the parameters such as the wheel diameter value after the last turning, the flange thickness value after the last turning, the equivalent taper, the coaxial wheel diameter difference, the coaxial flange thickness difference, the operating mileage, the wheel diameter wear amount, and the flange thickness wear amount one by one, and separating the data of the head and tail cars, the intermediate motor cars, and the intermediate trailers to establish wear models respectively.

[0043] S12: Establish a BP neural network model for training; commonly used is a three-layer neural network, that is, an input layer, a hidden layer, and an output layer (as Figure 2 shown). Determine the wheel diameter value after the last turning, the flange thickness value after the last turning, the coaxial wheel diameter difference, the coaxial flange thickness difference, and the operating mileage as input parameters, and the wheel diameter wear amount and the flange thickness wear amount as output parameters. Let the input be X = [X1,..., Xn], the output be Y = [Y1,..., Yl], and the number of hidden layers be m. The model training is as follows:

[0044] Randomly set the connection weights W, the hidden layer and output layer thresholds γ j , θ k , and then perform the output calculation of the hidden layer, as shown in Equation (1):

[0045]

[0046] In formula (1): f is the activation function of the hidden layer; α j is the input of the j-th hidden layer; W ij is the connection weight between the input layer and the hidden layer; m is the number of nodes in the hidden layer.

[0047] Calculation of the input of the output layer:

[0048]

[0049] In formula (2): g is the activation function of the output layer; β k is the input of the k-th output layer; W jk is the connection weight between the hidden layer and the output layer; l is the number of nodes in the output layer.

[0050] Based on the network predicted output Y and the desired output O, calculate the mean square error:

[0051]

[0052] The updated estimation formula for any parameter v is:

[0053] v = v + Δv Formula (4)

[0054] The BP algorithm is based on the gradient descent strategy, and adjusts the weights and thresholds in the negative gradient direction of the objective. If the iteration ends, the relevant results are output. If the iteration algorithm does not end, it returns to repeat the training process.

[0055] S13: Optimize the neural network parameters using the genetic algorithm to obtain the optimal model.

[0056] Determine the number of each weight and threshold, and encode each individual; the fitness calculation method is as shown in formula (5):

[0057]

[0058] In formula (5): n is the amount of prediction data; O i is the actual value; Y i is the predicted value.

[0059] Calculate the fitness value of each individual in the initial population according to the formula. Determine that the probability of randomly selecting an individual is inversely proportional to its fitness function value, and the roulette wheel method is used for the genetic algorithm selection; crossover is to inherit the excellent gene combinations of the previous generation to the next generation. Randomly select a gene position of two individuals as the crossover position to form a new excellent individual.

[0060] When the number of iterations of the genetic algorithm and the prediction result meet the expected error value, output the optimal individual. Assign the initial weights and thresholds optimized by the genetic algorithm to the neural network for training and prediction.

[0061] S14: Train and save separately according to vehicle type and car body type (head and tail cars, intermediate motor cars, intermediate trailers) for subsequent use in life prediction queries.

[0062] S2: Establish a wheel turning model based on turning data. The specific steps are as follows:

[0063] S21. Input the vehicle type to be modeled to obtain the corresponding turning data. If the number of turned wheels in the same car group within 7 days is greater than 50%, it is considered that a full - train turn has occurred. Subtract the wheel diameter and flange thickness values after turning from the wheel diameter and flange thickness values before turning in the same full - train turn to obtain the turning amount (R di ) and the flange thickness change amount (R ti ) of the turned wheels in this turn.

[0064] S22. Select the turning data with the flange thickness change amount R ti ≤ - 0.2. Starting from the flange thickness change amount = - 0.2, divide the interval to the left at intervals of 0.1. The data volume of each interval is denoted as N i , [- 0.3, - 0.2], [- 0.4, - 0.3], [- 0.5, - 0.4]... and so on until the nth interval, such that N n >100 and N n+1 <100.

[0065] S23. For each flange thickness change amount interval divided, group the data of each interval together and calculate the mean μ i and the standard deviation σ i ; conduct a second - round screening on the divided intervals, and save the data with the turning amount less than μ i +3*σ i in each interval. At this time, the data volume of each interval is re - denoted as N i , the total data volume is N, and recalculate the mean μ i and the standard deviation σ i of the remaining data; define the weight ω i of each interval as ω i =N i / N, and the flange thickness change amount of each interval is denoted as the intermediate value t

[0066] S24. Let μ = a + b*t. Calculate the values of a and b according to formula (6).

[0067]

[0068] Take the partial derivatives of J(a, b) with respect to a and b respectively, and the (a, b) that makes these two partial derivatives equal to 0 at the same time is the optimal solution. Denote |b| at this time as the flange thickness compensation ratio corresponding to the vehicle type. Let the minimum turning amount of the machine tool be Rdmin , denote R t0 =(R dmin -a) / b as the reference for flange thickness compensation. During wheel turning, when the change in flange thickness R t ≥R t0 , the turning amount is taken as R dmin ; when R t <R t0 , the turning amount is taken as |b|*(R t0 -R t )+R dmin . The minimum turning amount is taken for the left and right wheels on the same axle.

[0069] S25. Select the minimum flange thickness before turning and the corresponding flange thickness template for the same axle of the corresponding vehicle type. If the flange thickness application range is 26 - 34 mm, the minimum template is 28;

[0070] When the minimum flange thickness before turning of the same axle is less than 28 mm, the turning template is 28 mm; when the minimum flange thickness before turning of the same axle is greater than 33.5 mm, the turning template is 33.5 mm;

[0071] Starting from 28 mm, divide an interval every 0.5 mm, i.e., [28, 28.5), [28.5, 29), ……, [33, 33.5). The corresponding templates are 28.5 mm, 29 mm, ……, 33.5 mm. The probability of using the corresponding template for the corresponding interval is denoted as p i , and the sum of p i is denoted as P1. Move the interval division up by 0.1 mm, i.e., [28.1, 28.6), [28.6, 29.1), ……, [33.1, 33.6), and calculate the corresponding P2 at this time. And so on, calculate P3, P4 until P 10 . Take P = max{P1, P2, ……, P 10}}, and the interval corresponding to P is the selection logic of the flange thickness template. For those less than the lower limit of the interval of the minimum flange thickness before turning of the same axle, the flange thickness template for the same axle is taken as 28 mm.

[0072] S26. Integrate the selection logic of the turning flange thickness template and the turning amount calculation formula to establish a wheel turning model for subsequent reference.

[0073] S3: According to the wheel condition monitoring data, establish a wheel cycle determination model. The specific algorithm is as follows:

[0074] S31. Determine the wheel polygon rating standard according to the wheel polygon data. The wheel polygon rating standard is shown in Table 1 below:

[0075] Table 1 Wheel polygon rating standard

[0076]

[0077] S32. Determine the wheel equivalent taper rating standard based on the wheel equivalent taper data; calculate the average value of the top 10% of the equivalent taper before the last full train turning, denoted as a1; calculate the average value of the top 10% of the equivalent taper before the penultimate full train turning, denoted as a2;

[0078] If both a1 and a2 are empty, the default rating is EA; (only when it appears after advanced maintenance and before full train turning)

[0079] If one of a1 and a2 is empty, then a1 = a2;

[0080] If 0.6 * (a1 + 0.05 - elim) + 0.4 * (a2 + 0.05 - elim) ≤ 0, it is EA;

[0081] If 0.6 * (a1 + 0.05 - elim) + 0.4 * (a2 + 0.05 - elim) > 0, it is EB. Where elim is the equivalent taper limit value of this train set.

[0082] S33. The train set comprehensive rating standard is shown in Table 2 below:

[0083] Table 2 Train set comprehensive rating standard

[0084]

[0085] S34. Based on the wheel comprehensive rating and the number of head car defects, comprehensively determine the turning cycle. Taking the turning cycle of 220,000 - 300,000 km as an example:

[0086] Table 3 Comprehensive determination table of turning cycle

[0087]

[0088]

[0089] S35. When predicting the remaining life, the mileage of the first turning cycle is determined according to the train set rating. For each subsequent cycle, the train set rating is upgraded by one level until the mileage limit is reached. The running mileage of the first cycle after advanced maintenance takes the mileage limit.

[0090] S4: Based on the wheel tread wear prediction model, wheel turning model, and train set rating model established above, the remaining life prediction of the wheel can be realized. The specific algorithm is as follows:

[0091] S41. By sub - depot and vehicle type, calculate the average turning cycle `m` and average daily running mileage `s` of this vehicle type according to the historical turning data;

[0092] S42. Input the wheel information to be predicted, including the car number, carriage, axle position, wheel position, etc. If the number of wheels turned in the same car group within 7 days is greater than 50%, it is considered that a full train wheel turning has occurred. Search for parameters such as the date of the last full train wheel turning, running mileage, geometric dimensions after turning, and status of the wheels.

[0093] S43. If the last turning is a high-level repair, take the average wheel diameter value D1 and wheel flange thickness value T1 of the last three LY test data after the high-level repair as the calculation basis points; if the last turning is an operation repair, take the wheel diameter value D1 and wheel flange thickness value T1 after turning as the calculation basis points.

[0094] S43. Calculate the wheel diameter wear amount w di and wheel flange thickness wear amount w ti in this cycle according to the wheel wear model; D i-wd is the wheel diameter value before the next turning, and T i-wt is the wheel flange thickness value before the next turning; judge whether it reaches the limit. If it reaches the limit, enter S35; if it does not reach the limit, enter S34. (i is the cycle number, starting from 1)

[0095] S44. According to the established wheel turning model, input the wheel diameter and wheel flange thickness values before turning, judge the wheel flange thickness turning template and the turning amount of wheel diameter and wheel flange thickness, and calculate the wheel diameter value D i and T i after turning; judge whether the wheel diameter value after turning reaches the limit. If it reaches the limit, it is the limit of wheel operation repair and enter S46; if it does not reach the limit, i = i + 1 and enter S33 to start iterative loop until it reaches the limit.

[0096] S45. When the wheel reaches the limit during this operation period, the mileage s of this cycle reaching the limit is s = m*(D i -D lim ) / w di . The date d of reaching the limit is d = d0 + s*`s. Where d0 is the date after the last turning.

[0097] S46. When the wheel reaches the limit, output data such as the time of reaching the limit, cumulative running mileage, various parameter values, and predicted turning situation.

[0098] The embodiments of this solution are as follows:

[0099] Analyze the data of a certain type of vehicle in a certain EMU depot.

[0100] First, establish a wheel tread wear prediction model based on this vehicle type. The input parameters are the wheel diameter value after the last turning, the wheel flange thickness value after the last turning, running mileage, coaxial wheel diameter difference, and coaxial wheel flange thickness difference. There are 12 hidden layers, and the output parameters are wheel diameter wear and wheel flange thickness wear. Establish a BP neural network model and optimize it with a genetic algorithm.

[0101] The results are as Figure 2As shown, after iterative optimization, the model fitness reached below 0.149, proving that the genetic algorithm optimization achieved good results and the final accuracy of the model met the requirements.

[0102] Based on the turning data, a turning model for this vehicle type was established, and the parameters are shown in Table 4 below:

[0103] Table 4 Parameters of the turning model for the vehicle type

[0104] Benchmark for flange thickness compensation 0.2 Ratio of flange thickness compensation 5.5 Minimum amount of flange thickness compensation 0.4 Maximum amount of flange thickness compensation 20

[0105] The minimum flange thickness template for this round is 27.5, and the advanced repair flange thickness template is 28. The selection of the decision flange thickness template is shown in Table 5 below:

[0106] Table 5 Selection table of flange thickness template

[0107]

[0108]

[0109] The equivalent taper limit for this vehicle type is 0.3. Calculate the average value of the top 10% of the equivalent taper before the last full train turning, denoted as a1; calculate the average value of the top 10% of the equivalent taper before the penultimate full train turning, denoted as a2;

[0110] If both a1 and a2 are empty, the default rating is EA; (only when it appears after advanced repair and before full train turning) if one of a1 and a2 is empty, then a1 = a2;

[0111] If 0.6*(a1 - 0.25) + 0.4*(a2 - 0.25) ≤ 0, it is EA; if 0.6*(a1 - 0.25) + 0.4*(a2 - 0.25) > 0, it is EB.

[0112] The operation repair cycle for this vehicle type is 300,000 kilometers, and the advanced repair cycle is 1,650,000 kilometers. The final turning cycle determination logic is as follows:

[0113] Table 6 Turning cycle determination table

[0114]

[0115]

[0116] Combining the above several models, a wheel life prediction model was integrated to predict the wheel life of this vehicle type. The evolution law of the new wheel life is as Figure 4 、 Figure 5 、 Figure 6 shown, Figure 4 、 Figure 5 、 Figure 6 are the evolution of the new wheel life of the head and tail cars, intermediate motor cars, and intermediate trailers respectively.

[0117] The results are shown in Table 7 below:

[0118] Table 7 Evolution of the new wheel life of car body types

[0119]

[0120] Based on the wheel turning data, historical turning data, wheel polygons, equivalent taper, and defects of the leading car body, an algorithm for accurately predicting the remaining life of wheels has been successfully established.

[0121] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0122] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made in these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

[0123] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An analysis algorithm for predicting the remaining life of a wheel, characterized in that, It includes the following steps: Step S1: First, establish a wear prediction model for wheel diameter and flange thickness; Step S2: Establish a wheel set turning model; Step S3: Establish a train rating model; Step S4: Based on the wear prediction model for wheel diameter and flange thickness, the wheel set turning model, and the train rating model, establish a remaining wheel life prediction model; In the above-mentioned step S1, establishing the wear prediction model for wheel diameter and flange thickness includes the following steps: S11: Extract the turning data of the machine tool; calculate the parameters of the wear amount of wheel diameter and flange thickness during two adjacent turnings of the wheel; S12: Establish a BP neural network model for training; S13: Use the genetic algorithm to optimize the neural network parameters to obtain the optimal model; S14: Save the trained model for future reference; In the above-mentioned step S2, establishing the wheel set turning model includes the following steps: S21: Extract the turning data of the train; calculate the wheel diameter value and the change amount of flange thickness before and after turning; S22: Analyze the turning data to establish the relationship between the turning amount and the flange thickness compensation; S23: Analyze the selection of flange thickness templates and establish a flange thickness template selection model; S24: Establish a wheel set turning model; In the above-mentioned step S3, establishing the train rating model includes the following steps: S31: Determine the wheel polygon rating standard according to the wheel polygon data; S32: Determine the wheel equivalent taper rating standard according to the wheel equivalent taper data; S33: Determine the head car defect rating standard according to the occurrence times of head car defects; S34: Comprehensively rate the train according to the above three rating standards and determine the next turning cycle; 2. The analysis algorithm for predicting the remaining life of a wheel according to claim 1, characterized in that Modeling the wear of wheel diameter and flange thickness, the specific steps are as follows: S11: Subtract the wheel diameter and flange thickness values before the next full train turning from the wheel diameter and flange thickness values after the last full train turning of the same wheel to obtain the wear amount of wheel diameter and flange thickness during this operation cycle; Corresponding the parameters of the last turned wheel diameter value, the last turned flange thickness value, the equivalent taper, the coaxial wheel diameter difference, the coaxial flange thickness difference, the operation mileage, the wheel diameter wear amount, and the flange thickness wear amount one by one, and separating the data of the head and tail cars, the middle motor cars, and the middle trailers, and establish wear models respectively; S12: Establish a BP neural network model for training: The neural network model is a three-layer neural network, including an input layer, a hidden layer, and an output layer. Determine the last turned wheel diameter value, the last turned flange thickness value, the coaxial wheel diameter difference, the coaxial flange thickness difference, and the operation mileage as input parameters, and the wheel diameter wear amount and the flange thickness wear amount as output parameters. Let the input be X = [X1,..., Xn], the output be Y = [Y1,..., Yl], and the number of hidden layers be m. The training model is as follows: Randomly set the connection weights W, the thresholds γ of the hidden layer and the output layer j , and θ k , and then perform the output calculation of the hidden layer as shown in Equation (1): j = 1, 2,..., m In formula (1): f is the activation function of the hidden layer; α j is the input of the j-th hidden layer; W ij is the connection weight between the input layer and the hidden layer; m is the number of nodes in the hidden layer; Calculation of the input of the output layer: k＝1，2，...，l In formula (2): g is the activation function of the output layer; β k is the input of the k-th output layer; W jk is the connection weight between the hidden layer and the output layer; l is the number of nodes in the output layer; Calculate the mean square error from the network predicted output Y and the expected output O: The updated estimation formula for any parameter v is: v = v + Δv Equation (4) The BP algorithm is based on the gradient descent strategy, adjusts the weights and thresholds in the negative gradient direction of the target. If the iteration ends, the relevant results are output. If the iteration algorithm does not end, return and repeat the training process; S13. Optimize the neural network parameters using the genetic algorithm to obtain the best model: Determine the number of each weight and threshold, and encode each individual; The fitness calculation method is shown in the following formula (5). In formula (5): n is the amount of predicted data; O i is the actual value; Y i is the predicted value; Calculate the fitness value of each individual in the initial population according to the formula; Determine that the probability of randomly selecting an individual is inversely proportional to its fitness function value, and the roulette wheel method is selected for the genetic algorithm; Crossover is to inherit the excellent gene combinations of the previous generation to the next generation. Randomly select a gene position of two individuals as the crossover position to form new excellent individuals; When the number of iterations of the genetic algorithm and the prediction result meet the expected error value, output the optimal individual; Assign the initial weights and thresholds optimized by the genetic algorithm to the neural network for training and prediction; S14. Train and save separately according to the vehicle type and car body type for later use in life prediction. Among them, the car body types include the head car, the middle motor car, and the middle trailer car.

3. The analysis algorithm for predicting the remaining life of a wheel according to claim 2, characterized in that, Establish a wheel turning model based on the turning data. The specific steps are as follows: S21. Input the vehicle model to be modeled to obtain the corresponding wheel turning data: If the number of turned wheels in the same car body within 7 days is greater than 50%, it is determined that a full train wheel turning has occurred; subtract the wheel diameter and flange thickness values after turning from the wheel diameter and flange thickness values before turning in the same full train wheel turning to obtain the turning amount R di and the flange thickness change amount R ti ; S22. Select the wheel flange thickness change amount R ti Select the wheel turning data with a wheel flange thickness change amount R ≤ -0.

2. Starting from the wheel flange thickness change amount of 0.2, divide the intervals to the left at an interval of 0.1, and record the amount of data in each interval as N i , [-0.3, -0.2], [-0.4, -0.3], [-0.5, -0.4] and so on, until the nth interval, such that N n > 100 and N n+1 <100 S23. For each interval of the rim thickness change amount divided, group the data of each interval together and calculate the mean value μ of the turning amount i and the standard deviation σ i ; conduct a second screening for the divided intervals, and save the data with the turning amount less than μ i +3*σ i in each interval. At this time, the data volume of each interval is re - recorded as N i , the total data volume is N, and recalculate the mean value μ i of the remaining data turning amount and the standard deviation σ i ; define the weight ω i of each interval = N i / N, and record the rim thickness change amount of each interval as the intermediate value t i ; S24. Let μ = a + b*t; Calculate the values of a and b according to the following formula (6): Take the partial derivatives of J(a, b) with respect to a and b respectively. The (a, b) that makes these two partial derivatives equal to 0 at the same time is the optimal solution; Denote |b| at this time as the wheel flange thickness compensation ratio for the corresponding vehicle type. Let the minimum turning repair amount of the machine tool be R dmin , denote R t0 = (R dmin - a) / b as the reference for flange thickness compensation; during turning repair, when the change amount of flange thickness R t ≥ R t0 , the turning repair amount is taken as R dmin ; when R t < R t0 , the turning repair amount is taken as |b|*(R t0 - R t ) + R dmin ; the minimum value of the turning repair amounts of the left and right wheels on the same axis is taken; S25. Select the minimum wheel flange thickness before turning and the corresponding wheel flange thickness template for the same axle of the corresponding vehicle type. If the application range of the wheel flange thickness is 26 - 34 mm and the minimum template is 28; When the minimum wheel flange thickness before turning of the same axle is less than 28 mm, the turning template adopts 28 mm; When the minimum wheel flange thickness before turning of the same axle is greater than 33.5 mm, the turning template adopts 33.5 mm; When starting from 28 mm and dividing into intervals every 0.5 mm, the probability corresponding to each interval using the corresponding template is denoted as p i , p i The sum is denoted as P1; The interval division is moved upward by 0.1 mm, and the corresponding P2 is calculated. By analogy, P3, P4 are calculated until P 10 ; Take P = max{P1, P2, ……, P 10}, and the interval corresponding to P is the one where the wheel flange thickness template selection logic is less than the lower limit of the smallest wheel flange thickness interval before coaxial turning. For the coaxial wheel flange thickness template, take 28 mm; S26. Integrate the turning wheel flange thickness template selection logic and the turning amount calculation formula to establish a wheel turning model.

4. The analysis algorithm for predicting the remaining life of a wheel according to claim 3, wherein Establish a wheel cycle determination model based on the wheel status monitoring data. The specific algorithm steps are as follows: S31. Determine the wheel polygon rating standard according to the wheel polygon data: Divide the number of wheels according to the measured value of the high-order polygon of the wheel into grade A, grade B, grade C, and grade D; S32. Determine the wheel equivalent taper rating standard according to the wheel equivalent taper data: Calculate the average value of the first 10% of the equivalent taper before the last full train turning, denoted as a1; Calculate the average value of the first 10% of the equivalent taper before the penultimate full train turning, denoted as a2; Judge the rating as EA or EB according to the situation that a1 and a2 are empty; S33. Conduct a comprehensive rating according to the results of the polygon rating and the equivalent taper rating; S34. Comprehensively determine the turning cycle according to the wheel comprehensive rating and the number of head car defects; S35. When predicting the remaining life, the mileage of the first turning cycle is determined according to the car body group rating. After that, the car body group rating increases by one level for each cycle until the mileage limit is reached; The running mileage of the first cycle after the major repair takes the mileage limit.

5. The analysis algorithm for predicting the remaining life of a wheel according to claim 4, characterized in that, According to the established wheel diameter, wheel flange thickness wear prediction model, wheel set turning model, and car body group rating model above, the remaining life prediction of the wheel can be realized. The specific algorithm steps are as follows: S41. Calculate the average turning cycle `m` and the average daily running mileage `s` of this vehicle type according to the historical turning data by vehicle depot and vehicle type; S42. Input the wheel information to be predicted, including the car number, car body, axle position, and wheel position information, and find the date of the last full train turning, the running mileage, the geometric dimensions after turning, and the status parameters of the wheel; S43. Calculate the wheel diameter wear amount w and the flange thickness wear amount w of the wheel during this period according to the wheel wear model. di ; D ti ; D i-wd is the wheel diameter value before the next turning, and T i-wt is the flange thickness value before the next turning; determine whether it reaches the limit. If it reaches the limit, enter S35. If it is not out of limit, enter S34, where i is the number of cycles, starting from 1; S44. According to the established wheel turning model, input the values of the wheel diameter and flange thickness before turning, judge the flange thickness turning template and the turning amount of the wheel diameter and flange thickness, and calculate the wheel diameter value D after turning. i , T i ; Judge whether the wheel diameter value after turning reaches the limit. If it reaches the limit, it means that the wheel operation repair reaches the limit, and enter S46; if it does not reach the limit, i = i + 1, and enter S33 to start the iterative loop until it reaches the limit. S45. The wheel reaches its limit during this operation. The mileage limit for this wheel is s = m*(D i - D lim ) / w di , and the expiration date d = d0 + s*ˋs, where d0 is the date after the last wheel turning; S46. When the wheel is out of limit, output the out-of-limit time, the cumulative running mileage, the values of each parameter, and the predicted turning situation data.

Images