Metal rubber residual life prediction method based on physical information-temporal convolution network

Abstract

The application relates to a metal rubber residual life prediction method based on physical information-time sequence convolution network, in the proposed method, physical degradation information such as a metal rubber structure loss factor and a kinetic parameter can effectively reflect the characteristics of the metal rubber degradation trend, the physical information-time sequence convolution network method is extended to the life prediction of materials with nonlinear hysteresis characteristics, and meanwhile, a new idea is provided for modeling of the residual life prediction of the metal rubber.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal rubber, and particularly relates to a metal rubber residual life prediction method based on physical information-time sequence convolution network. BACKGROUND

[0002] Metal rubber is a key damping material for high-end equipment, but due to the complex preparation process, the internal space structure of each sample is different, and the traditional life prediction method often uses tests and empirical formulas, which has the problems of insufficient universality and poor prediction accuracy. The life prediction method of time sequence convolution network can couple the comprehensive characteristics of the fatigue life data of multiple metal rubbers into the prediction model, and to a certain extent, can solve the problem of insufficient universality of metal rubber life prediction, but the prediction accuracy of the time sequence convolution network has strict requirements for the data volume of the fatigue test, and insufficient data volume often leads to a significant decrease in prediction accuracy, so it is urgent to combine the physical information in the degradation process of the metal rubber with the time sequence convolution network to construct a time sequence convolution network (PI-TCN) based on the physical information of the fatigue damage of the metal rubber to predict the residual life of the metal rubber. SUMMARY

[0003] Therefore, the present application aims to provide a metal rubber residual life prediction method based on physical information-time sequence convolution network, which uses the physical information in the degradation process of the metal rubber to enhance the input space of the time sequence convolution network and optimize the gradient direction, and uses the ability of the time sequence convolution network (TCN) to couple multiple metal rubber degradation characteristics to improve the problem of insufficient universality of the traditional metal rubber life prediction and improve the accuracy of the metal rubber life prediction.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical scheme: a metal rubber residual life prediction method based on physical information-time sequence convolution network, comprising the following steps:

[0005] Step S1: degradation information of the metal rubber structure loss factor, the formula is as follows:

[0006]

[0007] In the formula, AW is the energy consumption of the metal rubber in a stress cycle period, and W is the maximum energy storage size of the metal rubber.

[0008] Step S2: constructing a nonlinear elastic restoring force model, a nonlinear damping force model, and a hysteresis damping force model according to the traditional dynamic model, and decomposing the hysteresis damping force by using Chebyshev polynomials;

[0009] Step S3: using a damage factor to represent the damage degree of the dynamic parameters of the metal rubber;

[0010] Step S4: coupling a plurality of kinetic parameter damage factors to construct a metal rubber kinetic comprehensive damage factor, whose formula is as follows:

[0011] d s = P i d i (2)

[0012] In the formula, d s is a comprehensive damage factor, d i is a damage factor of each kinetic parameter, P i is the contribution rate of each kinetic damage factor;

[0013] Step S5: curve fitting the mean value of the metal rubber comprehensive damage factor by using a two-parameter Weibull function, whose formula is as follows:

[0014]

[0015] In the formula: N is the cycle number, and λ and β are parameters related only to the loading condition and the properties of the metal rubber, and λ and β are regarded as to-be-identified parameters;

[0016] Step S6: based on the Weibull damage function, a physical information loss function is constructed by combining the mean square error function, whose formula is as follows:

[0017]

[0018] In the formula, y i is a real life label, is a predicted life label;

[0019] Step S7: a remaining useful life RUL is constructed to represent the fatigue life of the metal rubber, whose formula is as follows:

[0020] RUL(N) = N0-N (5)

[0021] In the formula, λ is a hyperparameter. y i is a real life label, is a predicted life label;

[0022] Step S8: the metal rubber degradation physical information constructed in steps 1 and 2 is introduced into the input layer of the physical information-time convolution network, and the physical information loss function constructed in step 6 is taken as the hybrid loss function of the network, and finally the construction of the physical information-time convolution network is completed.

[0023] In a preferred embodiment: the formula for decomposing the hysteresis damping force in step S2 is as follows:

[0024]

[0025] where k 2i-1 is the nonlinear elastic restoring force F k is the 2i-1 order stiffness coefficient of the nonlinear elastic restoring force (y), y0 is the initial pre-pressing amount; c 2i-1 is the nonlinear damping force is the 2i-1 order damping coefficient of the nonlinear damping force, N2 is the maximum order of the nonlinear damping force. k s is the equivalent linear stiffness of the hysteresis damping force, y s is the equivalent maximum slip distance, z s is the memory restoring force when the metal wire has relative slip, sgn(*) is a sign equation.

[0026] In a preferred embodiment: the damage factor is used to represent the damage degree of the metal rubber dynamic parameters in step S3, and the formula is as follows:

[0027]

[0028] wherein, is the tangent modulus of the metal rubber under the Nth cycle, is the initial tangent modulus of the metal rubber.

[0029] In a preferred embodiment: the damage equation of each parameter of the metal rubber dynamics is:

[0030]

[0031] wherein, k i (0) is the initial stiffness coefficient of each order of the metal rubber; k i (N) is the stiffness coefficient of each order after N cycles of load, c i (0) is the initial damping coefficient of each order of the metal rubber; c i (N) is the stiffness coefficient of each order after N cycles of load, k s (0) is the initial equivalent linear stiffness of the hysteresis damping force of the metal rubber; z s (0) is the memory restoring force when the metal rubber has initial relative slip; k s (N) is the equivalent linear stiffness of the hysteresis damping force after N cycles of load; z s (N) is the memory restoring force when the metal rubber has relative slip after N cycles of load;

[0032] The construction of the comprehensive damage factor mainly uses PCA principal component analysis:

[0033]

[0034] wherein: x i represents the damage factor of the dynamic parameters, μ i and σi respectively, X is the sample matrix of principal component analysis, R is the covariance matrix, X T is the transpose of X;

[0035] The mean square error loss function is:

[0036]

[0037] In the formula, y i is the real life label, is the predicted life label;

[0038] Construction of the hybrid loss function:

[0039]

[0040] In the formula, lambda is a hyperparameter.

[0041] Compared with the prior art, the present application has the following beneficial effects: the present application expands the input space of the physical-based time series convolution network by the physical information of the energy dissipation and the degradation of the kinetic parameters of the fatigue damage process of the metal rubber, further constructs the hybrid loss function of the metal rubber comprehensive damage factor Weibull equation network to constrain the optimization boundary of the physical-based time series convolution network (PI-TCN), and finally proposes a life prediction method of the physical-based time series convolution network (PI-TCN), which realizes effective prediction of the residual life of the metal rubber under limited samples. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 It is a system schematic diagram of the preferred embodiment of the present application.

[0043] Figure 2 It is a test tool diagram of the preferred embodiment of the present application.

[0044] Figure 3 It is a schematic diagram of the upper and lower ranges of the predicted real value of the PI-TCN model of the preferred embodiment of the present application.

[0045] Figure 4 It is a schematic diagram of the channel correlation of the preferred embodiment of the present application.

[0046] Figure 5 It is a heat map of the convolution dimension and the channel of the preferred embodiment of the present application. DETAILED DESCRIPTION

[0047] The present application will be further described below in combination with the drawings and embodiments.

[0048] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0049] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application; as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0050] The present embodiment provides a life prediction method of a physical-based timing convolutional network, as shown in Figure 1 The method comprises the following steps:

[0051] Step S1: Degradation information of the metal rubber structure loss factor, the formula is as follows:

[0052]

[0053] In the formula, ΔW is the energy consumption of the metal rubber in a stress cycle, W is the maximum energy storage size of the metal rubber,

[0054] Step S2: According to the traditional dynamic model, a nonlinear elastic restoring force model, a nonlinear damping force model, and a hysteresis damping force model are constructed, and the hysteresis damping force is decomposed by using Chebyshev polynomial, the formula is as follows:

[0055]

[0056] In the formula, k 2i-1 is the 2i-1 order stiffness coefficient of the nonlinear elastic restoring force F k (y), y0 is the initial pre-pressing amount. c 2i-1 is the 2i-1 order damping coefficient of the nonlinear damping force , N2 is the maximum order of the nonlinear damping force. k s is the equivalent linear stiffness of the hysteresis damping force, y s is the equivalent maximum slip distance, z s is the memory restoring force when the metal wire slips, and sgn(*) is the sign equation.

[0057] Step S3: A damage factor is used to represent the damage degree of the metal rubber dynamic parameters, the formula is as follows:

[0058]

[0059] In the formula, is the tangent modulus of the metal rubber for the Nth cycle, is the initial tangent modulus of the metal rubber.

[0060] Step S4: Since the metal rubber kinetic damage is a coupling effect of multiple kinetic parameter damage factors, considering only the change trend of a single kinetic damage factor cannot effectively reflect the degree of metal rubber fatigue damage. Therefore, it is necessary to couple multiple kinetic parameter damage factors to construct a metal rubber kinetic comprehensive damage factor, and its formula is as follows:

[0061] d s = P i d i (4)

[0062] In the formula, d s is the comprehensive damage factor, d i is the damage factor of each kinetic parameter, and P i is the contribution rate of each kinetic damage factor.

[0063] Step S5: The mean value of the metal rubber comprehensive damage factor is fitted by a two-parameter Weibull function, and its formula is as follows:

[0064]

[0065] In the formula: N is the number of cycles, λ and β are parameters related only to the loading condition and the properties of the metal rubber, and λ and β are regarded as to-be-identified parameters.

[0066] Step S6: Based on the Weibull damage function, a physical information loss function is constructed by combining the mean square error function, and its formula is as follows:

[0067]

[0068] In the formula, y i is the true life label, is the predicted life label.

[0069] Step S7: The remaining useful life (RUL) is constructed to represent the fatigue life of the metal rubber (the label of the data), and its formula is as follows:

[0070] RUL(N) = N0-N (7)

[0071] In the formula, λ is a hyperparameter. y i is the true life label, is the predicted life label.

[0072] Step S8: The physical information of the metal rubber degradation constructed in steps 1 and 2 is introduced into the input layer of the physical information-time sequence convolution network, and the physical information loss function constructed in step 6 is taken as the hybrid loss function of the network, and finally the construction of the physical information-time sequence convolution network is completed.

[0073] In this embodiment, 10 groups of metal rubber materials are subjected to cyclic load fatigue test. The test system mainly comprises a hydraulic station, a universal testing machine and a data acquisition system, and the test tooling is as shown in Figure 2 The 304 stainless steel is used to prepare a ring-shaped metal rubber sample with an outer diameter of 20 mm, an inner diameter of 9 mm, a height of 16 mm, and a density of 1.25 g / cm 3 The preparation process parameters are shown in Table 1. The upper clamp is driven to input a sinusoidal displacement excitation displacement with a frequency of 10 Hz and an amplitude of 1 mm to perform a fatigue test on the metal rubber sample. The data acquisition is performed in an interval sampling manner, and the test data is collected once every 2000 stress cycle periods. The sampling frequency of the test is set to 2560.

[0074] Table 1 Properties and preparation parameters of metal rubber

[0075]

[0076] The data of the test set is substituted into the three models for prediction comparison. The prediction accuracy R 2 of the Weibull model is 0.405, and the R 2 of the TCN model is only 0.2181. The prediction accuracy of the TCN model is reduced by 18.7%, which is because the TCN model is a pure data prediction method for predicting the remaining life of the metal rubber. This method has a great dependence on the quality and quantity of data. In the case of a small data set (set only 10 groups), the prediction of the TCN model is difficult to achieve the desired effect, and the fitting effect is even worse than that of the traditional model. The R 2 of the PI-TCN model is 0.967, which is improved by 56.2% compared with the traditional Weibull model.

[0077] Table 2 Comparison of training results of three prediction methods

[0078]

[0079] The PI-TCN model can greatly reduce the training loss value within 50 steps, while the TCN model needs more than 50 steps to reduce the training of the model. In the initial cycle load period, the fitting result of the Weibull model is closer to the true value, but as the cycle load time increases, the fitting result begins to deviate from the true value, and the overall prediction result of the Weibull model is higher than the actual value. The prediction result of the TCN model also deviates significantly from the true value of the metal rubber damage, and the prediction accuracy of the PI-TCN model is higher, and the range is up and down around the true value, as shown in Figure 3 .

[0080] The significance of the physical information features in the PI-TCN model is analyzed, the channel information model is established in the training network model, and the channel-dimension heat map is generated. The heat map can effectively represent the correlation between the dimension information and the channel. When the correlation of the dimension information in the channel is higher, the normalized value of the region tends to 1, and when the correlation of the dimension information in the channel is lower, the normalized value of the region tends to 0. Only feature 26 in the physical information feature information has low correlation with all channels, and the remaining physical features have high correlation in the channel, and the correlation of the first-order stiffness degradation information (feature 21) is the highest, as shown in Figure 4 .

[0081] Further study the influence of physical information in the convolution layer, select the input layer as the input and the middle convolution layer as the output, train the model and draw the heat map of the convolution dimension and the channel, the PI-TCN model improves the convolution feature dimension and the number of channels, and the number of convolution features with correlation increases from 7 to 8, and the number of channels also increases by one (channel 1), as shown in Figure 5 . This is because the hybrid loss function can increase the spatial boundary constraint after convolution processing, which can optimize the convolution features, increase the correlation between the convolution features and the channels, and finally optimize the output space and improve the prediction result of the model.

[0082] The above only describes the preferred embodiments of the present application, and any changes and modifications made within the scope of the application should be included in the scope of the application.

Claims

1. A metal rubber residual life prediction method based on a physical information-temporal convolution network, characterized by: Comprising the following steps: Step S1: the degradation information of the metal rubber structure loss factor, whose formula is as follows: In the formula, is the energy consumption of the metal rubber in a stress cycle, is the maximum energy storage of the metal rubber; Step S2: according to the traditional dynamic model, the nonlinear elastic restoring force model, the nonlinear damping force model and the hysteresis damping force model are constructed, and the hysteresis damping force is decomposed by using Chebyshev polynomial; Step S3: the damage factor is used to represent the damage degree of the metal rubber dynamic parameter; Step S4: the damage factors of multiple dynamic parameters are coupled to construct the comprehensive damage factor of the metal rubber dynamics, whose formula is as follows: （2） where d s is the overall impairment factor, d i is the impairment factor for each kinetic parameter, P i is the contribution rate of each kinetic impairment factor; Step S5: the mean value of the comprehensive damage factor of the metal rubber is curve fitted by using two-parameter Weibull function, whose formula is as follows: （3） wherein: N is the number of cycles, and are parameters related only to the loading conditions and to the properties of the metal rubber itself, and and are considered to be the parameters to be identified; Step S6: based on the Weibull damage function, the physical information loss function is constructed by combining the mean square error function, whose formula is as follows: （4） wherein is a true lifetime label, is a predicted lifetime label; Step S7: the residual useful life RUL is constructed to represent the fatigue life of the metal rubber, whose formula is as follows: （5） Step S8: the metal rubber degradation physical information constructed by step S1 and step S2 is introduced into the input layer of the physical information-time sequence convolution network, and the physical information loss function constructed by step S6 is taken as the hybrid loss function of the network, and finally the construction of the physical information-time sequence convolution network is completed; In step S3, the damage factor is used to represent the damage degree of the metal rubber dynamic parameter, whose formula is as follows: （7） wherein is the tangent modulus of the metal rubber at the Nth cycle, is the initial tangent modulus of the metal rubber.

2. The metal rubber residual life prediction method based on physical information-timing convolution network according to claim 1, characterized in that: In step S2, the formula for decomposing the hysteresis damping force by using Chebyshev polynomial is as follows: （6） wherein is a nonlinear elastic restoring force of order stiffness coefficient, is an initial pre-pressing amount; is a nonlinear damping force of order damping coefficient, is a maximum order of the nonlinear damping force; is a hysteresis damping force equivalent linear stiffness, is an equivalent maximum slip distance, z s is a memory restoring force when the metal wire occurs relative slip, is a symbolic equation.

3. The metal rubber residual life prediction method based on physical information-timing convolution network according to claim 1, characterized in that: The damage equation of each parameter of the metal rubber dynamics: （8） In the formula, represents the initial stiffness coefficients of the metal rubber at each order; for Stiffness coefficients of each order after the second cycle loading. denoted as the initial damping coefficients of the metal rubber at each order; for Stiffness coefficients of each order after the second cycle loading. The equivalent linear stiffness of the initial hysteresis damping force of the metal-rubber; The memory recovery force of the metal and rubber during initial relative slip; for Equivalent linear stiffness of hysteresis damping force after sub-cycle loading; for Memory recovery force during relative slip after sub-cycle loading; The construction of the comprehensive damage factor mainly uses PCA principal component analysis: （9） wherein: x i denotes the kinetic parameter damage factor, and denote the mean and standard deviation, respectively, X is the sample matrix of the principal component analysis, R is the covariance matrix, X T is the transpose of X; The mean square error loss function: （10） wherein is a true lifetime label, is a predicted lifetime label; The construction of the hybrid loss function: （11） In the formula, is a hyperparameter.

Images