Composite material performance degradation prediction method based on fusion of interval process and neural network
By combining interval process models and bidirectional long and short-term memory neural networks, the difficulty in adjusting model parameters and data dependence problems in the performance degradation prediction of composite materials are solved, and higher prediction accuracy and interpretability are achieved, and uncertainty can be effectively considered. Provide more accurate prediction results.
Patent Information
- Application Number
- CN202411244905.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-24
AI Technical Summary
In the existing composite material performance degradation prediction methods, it is difficult to adjust model parameters based on physical mechanisms, while the data-driven method relies on the quality of the training data and is difficult to effectively consider the uncertainty of the degradation process, resulting in insufficient prediction accuracy and interpretability.
A fusion method based on interval process and neural network is adopted, and an interval process model is established using gray system theory, combined with a bidirectional long and short-term memory neural network, and a more accurate performance degradation prediction model is constructed through interval process coordinate transformation and smooth interpolation processing.
Improve the conservatism of interval estimation in small samples, reduce the input and output digital scale of neural networks, significantly improve prediction accuracy and feasibility, effectively consider the uncertainty at different degradation moments, and provide more accurate prediction results. .
Smart Images
Figure CN119181447A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of material technology, and relates to composite material performance degradation prediction technology, and specifically to a composite material performance degradation prediction method based on the fusion of interval process and neural network. Background Art
[0002] The performance degradation prediction methods of composite materials can be divided into two categories, namely, physical mechanism-based methods and data-driven methods. For the former, many models based on physical mechanism analysis have been developed to describe the degradation process and study their applications. Another large category of data-driven models for degradation prediction are mostly based on machine learning (ML) methods, including support vector regression (SVR), random forests, and artificial neural networks. In addition, for degradation prediction using time series data sets, LSTM networks belong to recurrent neural networks (RNNs) and are one of the most popular data-driven models.
[0003] For the performance degradation model of composite materials in actual engineering, both physics-based and data-driven methods have different limitations. If the model is properly defined, the physics-based method may have good model interpretability and high prediction accuracy, but at the same time the parameters of complex models are difficult to adjust and these models may also be incomplete. On the other hand, data-driven methods can map complex system patterns or system mechanisms, but this depends largely on the availability of data. In other words, the degradation trends captured by the neural network model are mainly based on the quality of the training dataset, rather than the physical mechanism of the degradation process. Therefore, when the training dataset is biased, the prediction performance may be affected. Summary of the invention
[0004] Purpose of the invention: In order to overcome the deficiencies in the prior art, a composite material performance degradation prediction method based on the fusion of interval process and neural network is provided, which combines the interval degradation model with the bidirectional long short-term memory neural network to realize the composite material performance degradation prediction.
[0005] Technical solution: To achieve the above purpose, the present invention provides a composite material performance degradation prediction method based on the fusion of interval process and neural network, comprising the following steps:
[0006] S1: Using grey system theory to establish interval process model as composite material degradation model;
[0007] S2: Obtain the median function and radius function of the interval process model;
[0008] S3: Use prior information to perform smooth interpolation processing on the obtained degradation test data set;
[0009] S4: Based on the median function and the radius function, the smooth interpolation test data set is combined with the interval process model to perform interval process coordinate transformation to obtain the mapping coordinate set after coordinate transformation;
[0010] S5: determine the sequence of the mapping coordinate set and perform normalization preprocessing;
[0011] S6: Using the mapping data preprocessed in step S4 to train and predict the constructed bidirectional long short-term memory neural network, and obtain a prediction data set;
[0012] S7: De-map the prediction data set based on the interval process coordinates to obtain a prediction data set of performance degradation data.
[0013] Furthermore, the method for establishing the interval process model in step S1 includes:
[0014] A1: For a set of degraded data x={x1,x2,...,x M}, sort the test data from small to large to get the data sequence x (0) ;
[0015] A2: By sequentially adding up the sequence x (0) elements, and obtain a new cumulative sequence x (1) ;
[0016] A3: With sample sequence i as the horizontal axis, accumulate observation data is the vertical axis, calculate the difference Δ(i), and obtain the maximum difference Δ max ;
[0017] A4: Using the maximum difference Δ max and the gray recognition coefficient c to calculate the upper and lower boundaries of the interval variable;
[0018] A5: Obtain the interval variables at each moment, perform least squares fitting on the boundaries of the interval variables based on the prior information of composite material performance degradation, and establish an interval process model.
[0019] Furthermore, the calculation method of the upper and lower boundaries of the interval variable in step A4 is:
[0020] The standard deviation s of the test data is defined as:
[0021]
[0022] Use the standard deviation s to establish the upper and lower boundaries of the interval, expressed as:
[0023]
[0024] Among them, x m YesN s is the mean of the degraded data, and K is the expansion parameter of the interval boundary.
[0025] Furthermore, the median function X of the interval process model in step S2 m (t) and the radius function X r (t) represent the center line of the performance degradation trend and the boundary of uncertainty, respectively, expressed as:
[0026]
[0027] in, X (t) and Represent the lower and upper boundaries of the performance degradation model respectively.
[0028] Furthermore, the step S4 is specifically as follows:
[0029] The experimental data set after smooth interpolation processing is Combine the actual test data with the interval process model to perform interval process coordinate transformation Get the transformed mapping coordinate set δ i ;
[0030] For any single data point x i (t j ), whose value in interval process coordinates is:
[0031]
[0032] According to the definition of the formula, -1≤δ i (t j )≤1;
[0033] The degraded dataset x of a single sample i After the interval process coordinate transformation, the mapped coordinate value set δ i ={δ i (t1),δ i (t2),...,δ i (t n )}.
[0034] Furthermore, the bidirectional long short-term memory neural network in step S6 includes an input layer, a hidden layer and an output layer.
[0035] Furthermore, the training method of the bidirectional long short-term memory neural network in step S6 is: obtaining the prediction result of the next time step according to the training data, by feeding back the prediction information as the training data, and iteratively obtaining the prediction data set.
[0036] The interval process model in the present invention provides a prediction benchmark based on prior knowledge, and the degradation data from the test is mapped in the interval process. The bidirectional long short-term memory (Bi-LSTM) neural network will predict the performance degradation path of the composite material based on this highly nonlinear mapping.
[0037] Beneficial effects: Compared with the prior art, the present invention has the following advantages:
[0038] 1. The present invention uses grey system theory to establish an interval process model, which improves the conservatism of interval estimation under small sample conditions and reduces the demand for the number of test samples.
[0039] 2. The present invention uses a neural network to predict the relative position in the interval process coordinates, which can significantly reduce the digital scale in the input and output of the neural network.
[0040] 3. Compared with the existing deviation-based method, the present invention can effectively consider the uncertainty differences at different degradation moments and provide more accurate prediction results.
[0041] 4. The present invention uses a neural network to predict the relative position in the interval process coordinates, fully considering the degradation trend information of the physical mechanism provided by the interval process model. The output results of the neural network are strictly limited to the upper and lower boundaries of the interval process, thereby improving the overall prediction accuracy and feasibility. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 is a flow chart of the method of the present invention;
[0043] Figure 2 It is the degradation data and interval process diagram before and after the coordinate transformation in the method of the present invention;
[0044] Figure 3 Schematic diagram of interval residual stiffness degradation model in an embodiment of the present invention;
[0045] Figure 4 is the degradation data diagram of the test sample;
[0046] Figure 5 Graph showing the predicted results of the residual stiffness in an embodiment of the present invention. DETAILED DESCRIPTION
[0047] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.
[0048] The present invention provides a composite material performance degradation prediction method based on the fusion of interval process and neural network, such as Figure 1 As shown, it includes the following steps:
[0049] S1: Using grey system theory to establish interval process model as composite material degradation model;
[0050] Grey system theory reorganizes the original rough data into a strong regular sequence. For a set of degraded data x={x1,x2,...,x Ns}, sort the test data from small to large to get the data sequence x (0) ;
[0051]
[0052] in, i=1,2,...,N s -1;N s is the sample size;
[0053] By sequentially adding the sequence x (0) elements, and obtain a new sequence x (1) ,Right now:
[0054]
[0055] With sample sequence i as the horizontal axis, the accumulated observation data is the vertical axis, and the difference Δ(i) is defined as:
[0056]
[0057] If the observations have no uncertainty, then and x i (1) There is no difference, that is, Δ(i) = 0; the difference Δ(i) reflects the uncertainty of the parameter. The larger the difference Δ(i), the greater the uncertainty. The maximum difference is:
[0058] Δ max =max(Δ(1),Δ(2),...,Δ(N s ))
[0059] The uncertainty of the parameters is evaluated in a grey manner, and the standard deviation s of the test data is defined as:
[0060]
[0061] Wherein, c is the grey recognition coefficient, which is almost unaffected by the distribution of observed data and can be calculated by the grey evaluation model. In this embodiment, the coefficient c is taken as 2.5; then the upper and lower boundaries of the interval are established using the standard deviation s, which can be expressed as:
[0062]
[0063] Among them, x m YesN s The mean of the degraded data, K is the expansion parameter of the interval boundary, and according to Chebycheff inequality and statistical analysis, K is set to 3;
[0064] The interval variables at each moment are obtained, and the boundaries of the interval variables are fitted by the least square method according to the prior information of composite material performance degradation to establish the interval process model.
[0065] S2: Get the median function X of the interval process model m (t) and the radius function X r (t);
[0066] The median function X of the interval process model m (t) and the radius function X r (t) represent the center line of the performance degradation trend and the boundary of uncertainty, respectively, expressed as:
[0067]
[0068]
[0069] Among them, X(t) and They represent the lower and upper boundaries of the performance degradation model, that is, the upper and lower boundaries of the interval established using the standard deviation s.
[0070] S3: Get the degradation test dataset x i Considering the possible errors in the test observation, the obtained degradation test data set is smoothed and interpolated using the prior information to obtain the processed data set
[0071] S4: Based on the median function X m (t) and the radius function X r (t), combining the smooth interpolation processed test data set with the interval process model, performing interval process coordinate transformation, and obtaining a mapping coordinate set after coordinate transformation;
[0072] The actual test data (data set ) is combined with the interval process model to perform interval process coordinate transformation Get the transformed mapping coordinate set δ i ;
[0073] like Figure 2 As shown, for any single data point x i (t j ), whose value in interval process coordinates is:
[0074]
[0075] According to the definition of the formula, -1≤δ i (t j )≤1;
[0076] The degraded dataset x of a single sample i After the interval process coordinate transformation, the mapped coordinate value set δ i ={δ i (t1),δ i (t2),...,δ i (t n )}.
[0077] S5: Determine the mapping coordinate set δ i The sequence is processed by normalization preprocessing;
[0078] S6: Use the preprocessed mapping data to train and predict the constructed bidirectional long short-term memory neural network to obtain a prediction data set;
[0079] The bidirectional long short-term memory neural network includes an input layer, a hidden layer and an output layer.
[0080] The training method of the bidirectional long short-term memory neural network is to obtain the prediction result of the next time step based on the training data, and to obtain the prediction data set by feeding back the prediction information as training data and iterating. Among them, l is the number of prediction steps.
[0081] S7: Predicting datasets based on interval process coordinates Perform inverse mapping to obtain a prediction data set of performance degradation data
[0082] In order to verify the effectiveness and effect of the method of the present invention, this embodiment applies the method of the present invention to an example, as follows:
[0083] In this example, the residual stiffness degradation data of two-dimensional C / SiC ceramic matrix composite materials under room temperature fatigue is used as the prediction sample. Considering the uncertainty in the degradation process, the interval process model is established as follows: Figure 3The upper and lower boundaries of this interval process are two different types of degradation trajectories, where the upper boundary mainly includes the I and II stages of the fatigue degradation process, while the lower boundary has obvious III stage degradation.
[0084] In addition, a sample of residual stiffness data was selected from the literature for training and verification of the prediction method of the present invention. The residual stiffness data is as follows: Figure 4 shown.
[0085] This embodiment uses the least square method based on the prior model (Fittingbased on training data), the Bi-LSTM neural network based on bias (Bi-LSTM based on bias) and the fusion method based on interval processes (Bi-LSTM based on interval processes) of the present invention to predict the degradation data. The prediction results of the three methods are shown in Figure 5 .from Figure 5 It can be found that among the three methods, the fusion method based on interval process shows the best prediction effect.
Claims
1. A composite material performance degradation prediction method based on the fusion of interval process and neural network, characterized in that: The steps include: S1: Using grey system theory to establish interval process model as composite material degradation model; S2: Obtain the median function and radius function of the interval process model; S3: Use prior information to perform smooth interpolation processing on the obtained degradation test data set; S4: Based on the median function and the radius function, the smooth interpolation test data set is combined with the interval process model to perform interval process coordinate transformation to obtain the mapping coordinate set after coordinate transformation; S5: determine the sequence of the mapping coordinate set and perform normalization preprocessing; S6: Using the mapping data preprocessed in step S4 to train and predict the constructed bidirectional long short-term memory neural network, and obtain a prediction data set; S7: De-map the prediction data set based on the interval process coordinates to obtain a prediction data set of performance degradation data.
2. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 1 is characterized in that: The method for establishing the interval process model in step S1 includes: A1: For a set of degraded data x={x1,x2,...,x M }, sort the test data from small to large to get the data sequence x (0) ; A2: By sequentially adding up the sequence x (0) elements, and obtain a new cumulative sequence x (1) ; A3: With sample sequence i as the horizontal axis, accumulate observation data is the vertical axis, calculate the difference Δ(i), and obtain the maximum difference Δ max ; A4: Using the maximum difference Δ max and the gray recognition coefficient c to calculate the upper and lower boundaries of the interval variable; A5: Obtain the interval variables at each moment, perform least squares fitting on the boundaries of the interval variables based on the prior information of composite material performance degradation, and establish an interval process model.
3. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 2 is characterized in that: In step A1, the data sequence x (0) The expression is: in, N s is the sample size; In step A2, the accumulated sequence x (1) The expression is:
4. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 2 is characterized in that: The difference Δ(i) in step A3 is defined as: Maximum difference Δ max for: D max =max(Δ(1),Δ(2),...,Δ(N s ))。 5. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 2 is characterized in that: The calculation method of the upper and lower boundaries of the interval variable in step A4 is: The standard deviation s of the test data is defined as: Use the standard deviation s to establish the upper and lower boundaries of the interval, expressed as: Among them, x m YesN s is the mean of the degraded data, and K is the expansion parameter of the interval boundary.
6. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 1 is characterized in that: The median function X of the interval process model in step S2 m (t) and the radius function X r (t) represent the center line of the performance degradation trend and the boundary of uncertainty, respectively, expressed as: Among them, X(t) and Represent the lower and upper boundaries of the performance degradation model respectively.
7. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 6 is characterized in that: The step S4 is specifically as follows: The experimental data set after smooth interpolation processing is Combine the actual test data with the interval process model to perform interval process coordinate transformation Get the transformed mapping coordinate set δ i ; For any single data point x i (t j ), whose value in interval process coordinates is: According to the definition of the formula, -1≤δ i (t j )≤1; The degraded dataset x of a single sample i After the interval process coordinate transformation, the mapped coordinate value set δ i ={δ i (t1),δ i (t2),...,δ i (t n )}.
8. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 1 is characterized in that: The bidirectional long short-term memory neural network in step S6 includes an input layer, a hidden layer and an output layer.
9. The composite material performance degradation prediction method based on the fusion of interval process and neural network according to claim 1 is characterized in that: The training method of the bidirectional long short-term memory neural network in step S6 is: obtaining the prediction result of the next time step according to the training data, by feeding back the prediction information as the training data, and iteratively obtaining the prediction data set.