A fatigue life prediction method for rubber materials based on extreme learning machine
Through the fatigue life prediction method of rubber material optimization based on the extreme learning machine and dragonfly algorithm, the problem of low prediction accuracy in the existing technology is solved, and the impact of strain mean and strain amplitude is achieved, which improves the prediction accuracy and general applicability , significantly improves the durability of rubber parts design.
Patent Information
- Application Number
- CN202111358465.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-16
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing methods for predicting fatigue life of rubber materials are difficult to consider the impact of strain mean and strain amplitude on fatigue life at the same time, resulting in low prediction accuracy, and slow training speed and poor generalization performance of traditional neural network models.
The method based on the limit learning machine is adopted, combined with the dragonfly algorithm to optimize the input weight and hidden layer threshold, and a prediction model of the fatigue life of rubber materials is established. The effect of strain mean and strain amplitude on fatigue life is considered at the same time through the limit learning machine model. And the output weight matrix is calculated by Moore-Penrose generalized inverse.
It improves the accuracy and generality of the fatigue life prediction of rubber materials, can accurately predict life within two times the dispersion line, and significantly improves the durability of rubber parts design.
Smart Images

Figure CN114201911B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for predicting the fatigue life of a rubber material, and in particular to a method for predicting the fatigue life of a rubber material based on an extreme learning machine. Background Art
[0002] Rubber materials have high damping, low cost and excellent energy absorption characteristics, and are widely used in vehicle vibration isolation devices. Typical vehicle vibration isolation devices include: powertrain suspension, torsional vibration damper, tire and bushing, etc. Rubber vibration isolators usually operate under cyclic alternating loads, and their main vibration isolation rubber materials are prone to fatigue failure. Preventing or reducing fatigue failure of rubber materials during service is a key issue in the initial design stage of rubber vibration isolators. Therefore, accurately predicting the fatigue life of rubber materials and providing effective data reference for the initial design stage of rubber vibration isolators has important theoretical and engineering research significance.
[0003] The factors affecting rubber fatigue include mechanical load, loading environment, rubber formula and constitutive model, among which mechanical load is the main factor affecting rubber fatigue life. Mechanical load mainly includes applied stress or strain level, loading frequency and loading conditions. At present, the main methods for predicting material fatigue life are: crack initiation method based on continuum mechanics, crack propagation method based on fracture mechanics and intelligent algorithm based on machine learning. The crack initiation method predicts fatigue life by determining the power function relationship between fatigue damage parameter and fatigue life. Commonly used fatigue damage parameters include maximum principal strain, equivalent stress and strain energy density. This method can usually only consider the influence of one factor on fatigue life. The corresponding equivalent method can consider the influence of two factors, but the accuracy is not high. The crack propagation method calculates the integral of crack propagation rate, establishes the relationship between crack propagation rate and fatigue damage parameter, and thus determines the predicted fatigue life. The rapid development of machine learning in recent years has enabled many data analysis methods to be successfully applied to small sample fatigue data prediction (Fatigue Life Prediction for Vibration Isolation Rubber Based on Parameter-OptimizedSupport Vector Machine Model).
[0004] As a new algorithm for single hidden layer feedforward neural network, Extreme Learning Machine (ELM) calculates the output weight matrix between the hidden layer and the output layer through linear algebra instead of using the traditional forward propagation algorithm or Bayesian algorithm, which overcomes the shortcomings of the traditional artificial BP neural network model and has the advantages of fast training speed and high accuracy. The ELM algorithm has a certain randomness in the selection of input weights and hidden layer thresholds, which makes it difficult to ensure the prediction accuracy and has poor generalization performance. The dragonfly algorithm that has appeared in recent years is introduced to optimize the input weights and hidden layer thresholds. Compared with genetic algorithm and particle swarm algorithm, the dragonfly algorithm has the advantages of not being easy to fall into local optimality and simple parameter setting. At the same time, the five behaviors ensure the global and local search capabilities of the dragonfly population, so as to establish an extreme learning machine rubber material fatigue life prediction model with high prediction accuracy. Summary of the invention
[0005] The present invention provides a method for predicting the fatigue life of rubber materials based on an extreme learning machine, specifically a method for obtaining an extreme learning machine fatigue life prediction model after training by using rubber fatigue test data as input of an extreme learning machine neural network model. The method can simultaneously consider the influence of two influencing factors, strain mean and strain amplitude, on the fatigue life of rubber, and has a high prediction accuracy, thereby providing a reference for improving durability in the early stages of rubber component design.
[0006] The present invention is achieved by at least one of the following technical solutions.
[0007] A method for predicting fatigue life of rubber materials based on an extreme learning machine comprises the following steps:
[0008] S1. Setting fatigue test data of rubber materials, wherein the test data includes corresponding cycle life times under different strain means and strain amplitudes;
[0009] S2, preprocessing the test data;
[0010] S3, using the preprocessed test data as the input of the extreme learning machine model, determining the number of neurons in the input layer, hidden layer, and output layer of the extreme learning machine model, and randomly generating an input weight matrix between the input layer and the hidden layer and a threshold matrix of the hidden layer as initial values;
[0011] S4, optimizing the input weight matrix and the hidden layer threshold matrix to determine the input weight matrix and the hidden layer threshold matrix with the minimum mean square error between the predicted life and the actual life;
[0012] S5, returning the optimal input weight matrix and the hidden layer threshold matrix to the extreme learning machine model, combining the input and theoretical output, and calculating the Moore-Penrose generalized inverse of the hidden layer output matrix to obtain the output weight matrix;
[0013] S6. Combining the input weight matrix, the hidden layer threshold matrix and the output weight matrix, a fatigue life prediction model of rubber materials based on an extreme learning machine is established;
[0014] S7. Use the extreme learning machine's fatigue life prediction model for rubber materials to predict life, thereby completing the life prediction work for rubber materials.
[0015] Preferably, the life prediction simultaneously considers the influence of the strain mean and the strain amplitude on the fatigue life of the rubber material.
[0016] Preferably, the preprocessing includes de-extreme value, logarithmic averaging, normalization, test set division and test set processing.
[0017] Preferably, the extreme learning machine model is a neural network model with a single hidden layer, including an input layer and a hidden layer, the input layer and the hidden layer are connected by an input weight matrix, each neuron in the hidden layer has a threshold, and the hidden layer and the output layer are connected by an output weight matrix.
[0018] Preferably, the dragonfly algorithm is used to optimize the input weights and hidden layer thresholds of the extreme learning machine, thereby reducing the impact of randomly determined input weights and hidden layer thresholds on the prediction accuracy of the model.
[0019] Preferably, the hidden layer output matrix in step S5 is:
[0020]
[0021] Among them, H L×1 is the hidden layer output matrix, w Ln is the connection weight between the Lth neuron in the hidden layer and the nth neuron in the input layer, x n is the input of the nth sample, b L is the threshold of the nth neuron in the hidden layer, and L is the number of neurons in the hidden layer.
[0022] Preferably, the training set is used to train the input weights and hidden layer threshold parameters in the extreme learning machine model, and the test set is used to verify the prediction accuracy of the model.
[0023] Preferably, the dragonfly algorithm optimizes the input weights and hidden layer thresholds including: assigning a set of input weights and hidden layer thresholds to the dragonfly individual as position information, initializing the position of the dragonfly individual within a selected range, updating the position through the dragonfly's separation behavior, alignment behavior, cohesion behavior, attraction behavior, and avoidance behavior, using the extreme learning machine evaluation index as the fitness function of the dragonfly individual, and when the number of iterations reaches the set value or a dragonfly individual finds the optimal position, outputting the dragonfly individual with the highest fitness, whose position information is the optimal set of output weights and hidden layer thresholds.
[0024] Preferably, updating the dragonfly position includes: inputting weights and hidden layer threshold values are both small values within 5, and setting weight coefficients of various behaviors that affect the dragonfly position update, so that the local search capability of the population is better than the global search capability.
[0025] Preferably, among the extreme learning machine evaluation indicators, the weighted mean square error, mean absolute percentage error and determination coefficient between the actual output and the theoretical output are used as the fitness function of the optimization algorithm to obtain a set of extreme learning machine parameters with the minimum discreteness between the calculated and theoretical outputs.
[0026] Compared with the prior art, the present invention has the following advantages:
[0027] 1) The fatigue life prediction method of rubber materials based on extreme learning machine can simultaneously consider multiple factors affecting fatigue life, including load level, temperature, rubber material, etc.;
[0028] 2) Compared with the life prediction method based on damage parameters, the prediction accuracy of the rubber material fatigue life prediction method based on extreme learning machine is higher than that of the latter;
[0029] 3) Compared with the life prediction method based on damage parameters, the prediction results of the rubber material fatigue life prediction method based on extreme learning machine are all within the double dispersion line, while the latter is within the triple dispersion line;
[0030] 4) The fatigue life prediction method of rubber materials based on extreme learning machine adopts the dragonfly algorithm to determine the optimal extreme learning machine model parameters, and the prediction model has better versatility and stability. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a flow chart of a method for predicting fatigue life of rubber materials based on an extreme learning machine of the present invention;
[0032] Figure 2 It is a structural schematic diagram of the extreme learning machine in the present invention;
[0033] Figure 3 It is a schematic diagram of the prediction results of the training set in the present invention;
[0034] Figure 4 It is a schematic diagram of the prediction results of the test set in the present invention;
[0035] Figure 5 It is a schematic diagram of the life span of the training set in the present invention;
[0036] Figure 6 It is a schematic diagram of the life distribution of the test set in the present invention. DETAILED DESCRIPTION
[0037] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0038] like Figure 1 As shown, a method for predicting fatigue life of rubber materials based on extreme learning machine includes the following steps:
[0039] S1. Input the fatigue test data of rubber materials, i.e. the corresponding cycle life times under different strain mean values and strain amplitudes;
[0040] S2, perform processing such as removing extreme values, logarithmic averaging, normalization, and dividing the test set and the test set on the test data;
[0041] S3, using the processed test data as the input x of the model i , determine the number of neurons in the input layer, hidden layer, and output layer in the extreme learning machine model, and randomly generate the input weight matrix between the input layer and the hidden layer and the threshold matrix of the hidden layer within the specified range as the initial value;
[0042] The extreme learning machine model of step S3 has a structural diagram as shown in Figure 2 As shown in the figure, the extreme learning machine is a single hidden layer neural network model. The input layer and the hidden layer are connected by the input weight matrix w L×n Each neuron in the hidden layer has a threshold value, which constitutes the threshold matrix b L×1 , the hidden layer and the output layer are connected by the output weight matrix β m×L Connection. x represents the actual input vector of the model, and y represents the expected output vector of the model:
[0043] x=[x1,x2,…,x n ] T (4)
[0044] y=[y1,y2,…,ym ] T (5)
[0045] where x i (i=1,2,…n),y j (j=1,2,…m) respectively represent the i-th model input and j-th model output in a sample of n inputs and m outputs, and g(x) is the neuron activation function.
[0046] S4, input weight matrix w L×n and the hidden layer threshold matrix b L×1 The dragonfly algorithm is used to optimize the input weights and hidden layer thresholds of the extreme learning machine. The purpose is to reduce the randomly determined input weights w Ln and hidden layer threshold b L Impact on model prediction accuracy.
[0047] First, initialize the individual positions of dragonflies and set an initial input weight matrix w L×n and the hidden layer threshold matrix b L×1 Assign dragonfly individuals, expressed as (w L×n,i , b L×1,i ), w L×n,i represents the input weight matrix information carried by the i-th dragonfly in the population, b L×1,i Represents the hidden layer threshold matrix information carried by the i-th dragonfly in the population;
[0048] Then the dragonfly behavior is calculated and the position of the individual dragonfly is updated, thereby determining the input weight matrix and hidden layer threshold matrix with the minimum mean square error between the predicted lifespan and the actual lifespan;
[0049] S5, the optimal input weight matrix w determined by the dragonfly algorithm L×n,best and the hidden layer threshold matrix b L ×1, best Return to the extreme learning machine model, combine the input and theoretical output, and calculate the hidden layer output matrix:
[0050] The Moore-Penrose generalized inverse of the output weight matrix β is obtained. m×L ; where H L×1 is the hidden layer output matrix, w Ln is the connection weight between the Lth neuron in the hidden layer and the nth neuron in the input layer, x n is the input of the nth sample, b L is the threshold of the nth neuron in the hidden layer, and L is the number of neurons in the hidden layer.
[0051] Determine the optimal input weight matrix and hidden layer threshold matrix: assign a set of input weights and hidden layer thresholds to the dragonfly individual as position information, initialize the position of the dragonfly individual within the selected range, update the position through the dragonfly's separation behavior, alignment behavior, cohesion behavior, attraction behavior and avoidance behavior, and use the extreme learning machine evaluation index as the fitness function of the dragonfly individual. When the number of iterations reaches the set value or a dragonfly individual finds the optimal position, output the dragonfly individual with the highest fitness, and its position information is the optimal set of output weights and hidden layer thresholds.
[0052] Among the extreme learning machine evaluation indicators, the weighted mean square error, mean absolute percentage error and determination coefficient between the actual output and the theoretical output are used as the fitness function of the optimization algorithm to obtain a set of extreme learning machine parameters with the minimum discreteness between the calculated and theoretical outputs.
[0053] S6. Combine the input weight matrix w L×n , hidden layer threshold matrix b L×1 And the output weight matrix β m×L , a fatigue life prediction model for rubber materials based on extreme learning machine is established; the establishment method is: taking the strain mean and strain amplitude as input and the fatigue life as output, and it can simultaneously consider the influence of multiple factors such as strain mean and strain amplitude on fatigue life.
[0054] The comparison between the predicted values and the measured values of the DA-ELM model (the extreme learning machine model under the optimized parameters of the dragonfly algorithm, hereinafter referred to as the model) training set is shown in the figure. Figure 3 As shown in Figure 2, the comparison between the predicted value and the measured value of the model test set is as follows: Figure 4 The life span distribution of the model training set prediction value and the measured value is shown in Figure 5 As shown, the life span distribution of the model test set predicted values and measured values is as follows Figure 6 As shown, both are within the 2-fold dispersion line.
[0055] S7, input the strain mean and strain amplitude into the rubber material fatigue life prediction model based on the extreme learning machine, obtain the predicted life, and complete the life prediction of the rubber material. Specifically determine a set of output weights and hidden layer thresholds with the smallest mean square error between the predicted life value and the actual life value, and obtain the output weight matrix β in the extreme learning machine neural network through step S6 m×L , and then obtain a fatigue life prediction model with high prediction accuracy. The input weight and hidden layer threshold are both small values within 5. The setting of the weight coefficients of each behavior that affects the position update of the dragonfly needs to make the local search ability of the population better than the global search ability.
[0056] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A method for predicting fatigue life of rubber materials based on extreme learning machine, characterized in that: The following steps are involved: S1. Setting fatigue test data of rubber materials, wherein the test data includes corresponding cycle life times under different strain means and strain amplitudes; S2, preprocessing the test data; S3. Using the preprocessed test data as the input of the extreme learning machine model, determining the number of neurons in the input layer, hidden layer, and output layer of the extreme learning machine model, randomly generating an input weight matrix between the input layer and the hidden layer and a threshold matrix of the hidden layer as initial values; using the dragonfly algorithm to optimize the input weights and hidden layer thresholds of the extreme learning machine, reducing the influence of the randomly determined input weights and hidden layer thresholds on the prediction accuracy of the model; The dragonfly algorithm optimizes the input weights and hidden layer thresholds, including: assigning a set of input weights and hidden layer thresholds to the dragonfly individual as position information, initializing the position of the dragonfly individual within the selected range, updating the position through the dragonfly's separation behavior, alignment behavior, cohesion behavior, attraction behavior and avoidance behavior, using the extreme learning machine evaluation index as the fitness function of the dragonfly individual, and when the number of iterations reaches the set value or a dragonfly individual finds the optimal position, outputting the dragonfly individual with the highest fitness, whose position information is the optimal set of output weights and hidden layer thresholds; S4, optimizing the input weight matrix and the hidden layer threshold matrix to determine the input weight matrix and the hidden layer threshold matrix with the minimum mean square error between the predicted life and the actual life; S5, returning the optimal input weight matrix and the hidden layer threshold matrix to the extreme learning machine model, and obtaining the output weight matrix by calculating the Moore-Penrose generalized inverse of the hidden layer output matrix; S6. Combining the input weight matrix, the hidden layer threshold matrix and the output weight matrix, a fatigue life prediction model of rubber materials based on an extreme learning machine is established; S7. Use the extreme learning machine's fatigue life prediction model for rubber materials to predict life, thereby completing the life prediction work for rubber materials.
2. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 1, characterized in that: The life prediction simultaneously considers the influence of strain mean and strain amplitude on the fatigue life of the rubber material.
3. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 1, characterized in that: The preprocessing includes de-extreme value, logarithmic averaging, normalization, test set division and experimental set processing.
4. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 1, characterized in that: The extreme learning machine model is a neural network model with a single hidden layer, including an input layer and a hidden layer. The input layer and the hidden layer are connected through an input weight matrix. Each neuron in the hidden layer has a threshold. The hidden layer and the output layer are connected through an output weight matrix.
5. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 1, characterized in that: The hidden layer output matrix in step S5 is: Among them, H L×1 is the hidden layer output matrix, w Ln is the connection weight between the Lth neuron in the hidden layer and the nth neuron in the input layer, x n is the input of the nth sample, b L is the threshold of the nth neuron in the hidden layer, and L is the number of neurons in the hidden layer.
6. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 3, characterized in that: The training set is used to train the input weights and hidden layer threshold parameters in the extreme learning machine model, and the test set is used to verify the prediction accuracy of the model.
7. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 1, characterized in that: Updating the dragonfly position includes: input weights and hidden layer thresholds are both small values within 5, and the weight coefficients of each behavior that affects the dragonfly position update are set to make the local search ability of the population better than the global search ability.
8. The method for predicting fatigue life of rubber materials based on extreme learning machine according to claim 1, characterized in that: Among the extreme learning machine evaluation indicators, the weighted mean square error, mean absolute percentage error and determination coefficient between the actual output and the theoretical output are used as the fitness function of the optimization algorithm to obtain a set of extreme learning machine parameters with the minimum discreteness between the calculated and theoretical outputs.
