Methods for predicting the lifespan of steer-by-wire systems

Abstract

This invention discloses a method for predicting the lifespan of a steer-by-wire system, comprising: S1, identifying degradation characteristics affecting the lifespan of the steer-by-wire system and collecting condition monitoring quantities related to these degradation characteristics; S2, analyzing the correlation between the condition monitoring quantities and the remaining lifespan, calculating the correlation between each condition monitoring quantity and the degradation characteristic quantity, and screening condition monitoring quantities that can reflect the aging trend of the steer-by-wire system; S3, using singular value decomposition to denoise the screened condition monitoring quantities; S4, using CEEMDAN to decompose the degradation characteristic quantity into intrinsic modulus components and residual components; S5, modeling the global degradation trend by inputting the intrinsic modulus components into an MHA-BiLSTM model; S6, modeling local fluctuations by inputting the residual components into a Hidden Markov Model; and S7, fusing the prediction outputs of the global degradation trend model and the local fluctuation model. The Hidden Markov Model is used to predict local fluctuations caused by uncertainty, and the final fusion of the prediction outputs achieves the lifespan prediction of the steer-by-wire system.

Description

Technical Field

[0001] This invention relates to the field of steer-by-wire systems, and more specifically to a method for predicting the lifespan of steer-by-wire systems. Background Technology

[0002] As a steering technology originating from aircraft, steer-by-wire offers significant advantages in steering performance, space utilization, and overall vehicle comfort by eliminating the mechanical connection between the steering wheel and steering wheels. However, vehicles employing steer-by-wire systems have increased electrical components, system complexity, and the risk of loss of control, potentially leading to catastrophic consequences. Therefore, based on functional safety analysis, researching the lifespan prediction of steer-by-wire systems has become an urgent problem. Existing methods for predicting the lifespan of steer-by-wire systems are mainly divided into data-driven and model-driven approaches. Model-driven methods rely on mechanistic degradation models, but the degradation process of components in real-world steer-by-wire systems is extremely complex, making it difficult to establish accurate mechanistic models. Data-driven methods rely on raw data to build behavioral models. While they can accurately construct the mapping relationship from massive amounts of degradation data to lifespan information, data-driven methods heavily depend on high-quality training data. Due to the uncertainty of real-world operating conditions, the data degradation process has significant randomness, affecting prediction performance. Summary of the Invention

[0003] To address the significant randomness of data degradation under real-world working conditions, this invention proposes a life prediction method for steer-by-wire systems. It utilizes MHA-BiLSTM to predict the global degradation trend and a Hidden Markov Model to predict local fluctuations caused by uncertainty. Finally, the prediction outputs are fused to achieve life prediction for the steer-by-wire system.

[0004] The technical solution adopted in this invention is as follows: A method for predicting the lifespan of a steer-by-wire system, comprising the following steps:

[0005] S1, determine the degradation characteristics that affect the lifespan of the steer-by-wire system, and collect the condition monitoring quantities and degradation characteristic quantities related to the degradation characteristics;

[0006] S2, analyze the correlation between condition monitoring quantities and remaining life, calculate the correlation between each condition monitoring quantity and degradation characteristic quantity, and screen the condition monitoring quantities that can reflect the aging trend of the steering-by-wire system.

[0007] S3, using singular value decomposition to reduce noise in the filtered state monitoring quantities;

[0008] S4. Using CEEMDAN, the degradation features are decomposed into intrinsic modulus components and residual components. The intrinsic modulus components reflect the global degradation trend, while the residual components reflect the local fluctuations caused by uncertainty.

[0009] S5, global degradation trend modeling is performed by inputting intrinsic mode components into the MHA-BiLSTM model to predict global degradation trend;

[0010] S6, Local fluctuations are predicted by inputting residual components into a hidden Markov model to model local fluctuations;

[0011] S7 integrates the predicted outputs of the global degradation trend model and the local fluctuation model to predict the degradation process of the steering system.

[0012] Optionally, step S1 specifically includes the following steps:

[0013] The state monitoring parameters for collecting degradation characteristics of the steer-by-wire system are assumed to be n state monitoring parameters, and a total of t time points are monitored. Then, the state monitoring parameter x is represented as:

[0014] x = {x1, x2, ..., x} n}

[0015] Where, x i =[x i1 ,x i2 ,...,x it ] T Let i = 1, 2, ..., n, then the state monitoring quantity x can be expressed as:

[0016]

[0017] The degradation characteristic quantity corresponding to the state monitoring quantity of the steer-by-wire system is determined as the accuracy of the steering system. Degradation data is collected and represented as: Y = [y1, y2, ..., y t ] T

[0018] Where yj represents the accuracy value of the steer-by-wire system at time j, j = 1, 2, ..., t.

[0019] Optionally, step S2 specifically includes the following steps:

[0020] S21, calculate the correlation between each state monitoring quantity and the remaining lifetime. The formula for calculating the correlation coefficient R is:

[0021]

[0022] Where, x i For the i-th state monitoring sequence, For the state monitoring sequence x i The mean of x, where y is the sum of x and x. i The corresponding degradation characteristics, Let y be the mean of the time series y;

[0023] S22, Screening for highly correlated state monitoring quantities: The closer the correlation coefficient is to 1, the stronger the correlation between the state monitoring quantity and the degradation characteristic quantity. Based on this, the critical range of the correlation coefficient r is determined to be 0.9 to 1, and state monitoring quantities within this range are selected.

[0024] Optionally, step S3 specifically includes the following steps:

[0025] S31, based on phase space reconstruction theory, construct an m×n Hankel matrix H from the state monitoring quantities, and perform singular value decomposition on this matrix to obtain uncorrelated noise subspace and signal subspace, as shown in the formula:

[0026]

[0027] Where U represents an M×N orthogonal matrix; D represents the diagonal elements; D=diag(σ1,σ2,σ3......σ n ) represents an M×N diagonal matrix; V represents an N×N orthogonal matrix;

[0028] S32. Find the singular values ​​of the Hankel matrix A. Retain the singular values ​​in the diagonal matrix that reflect the effective signal, and set the smaller singular values ​​reflecting the noise signal to zero to initially remove noise from the high-frequency signal. Then perform the inverse process of singular value decomposition to finally obtain the matrix. Then the matrix The best approximation moment of rank r for H is calculated using the following formula:

[0029]

[0030] Where, σ i r represents the singular values ​​of matrix A; r represents the rank of matrix H; T represents the transpose; u i v represents the i-th eigenvector of matrix AAT; i This represents the i-th eigenvector of matrix A.

[0031] matrix The anti-diagonal elements in the data are added together and averaged to obtain the denoised state monitoring quantity.

[0032] Optionally, step S4 specifically includes the following steps:

[0033] S41, Gaussian white noise ε0E1(ω i (t)) Add the signal to be decomposed x(t), and then perform empirical mode decomposition to obtain the first intrinsic mode components IMF1. i (t), for IMF1 iThe first-order intrinsic mode component IMF1(t) is obtained by averaging (t), and its formula is:

[0034]

[0035] Where N is the number of intrinsic modulus components, ω i (t) represents the white noise added for the i-th time, ε i The standard deviation of white noise;

[0036] S42, calculate the first-order residual component r1(t), the formula is:

[0037] r1(t) = x(t) - IMF1(t)

[0038] S43, continue adding Gaussian white noise to r1(t), for the new signal r1(t)+ε1E1(ω i (t) is subjected to empirical mode decomposition to obtain the second-order eigenmode components:

[0039]

[0040] Wherein, E1(r1(t)+ε1E1(w) i )) represents performing an empirical mode decomposition;

[0041] S44, and so on, yields the k-th order residual r. k (t), the formula is:

[0042] r k (t)=r k-1 (t)-IMF k (t)

[0043] S45, calculate the (k+1)th order eigenmode component as follows:

[0044]

[0045] The signal is decomposed cyclically until the obtained residual is a monotonic function and cannot be further decomposed. The formula for the final residual component is:

[0046]

[0047] The obtained intrinsic modulus components It reflects the global degradation trend, while the residual component r(t) reflects the local fluctuations caused by uncertainty.

[0048] Optionally, step S5 specifically includes the following steps:

[0049] S51, constructing a multi-head attention module, specifically:

[0050] S511, assuming the tensor input to the module is `input`, after the linear transformation layer, three tensors of the same size as the input are mapped: Q, K, and V. These three tensors are then fed into the scaled dot product attention calculation part to obtain the attention weights and the reconstructed attention matrix. The calculation formula is as follows:

[0051]

[0052] S512 divides tensors Q, K, and V into h equal-sized segments along the same direction (usually the last dimension of the tensor). The attention matrix is ​​calculated by placing each segment of the tensor in a different subspace. After calculation, the matrices of these subspaces are stitched together along the division direction to obtain the multi-head attention matrix, which is calculated as follows:

[0053] Multihead(Q,K,V)=Concat(head1,head2,…,head h W O

[0054] Among them, head i Let W be the attention matrix computed in the i-th subspace, and Concat be the concatenation operation of these sub-attention matrices along the splitting direction. O The parameter matrix for the linear transformation;

[0055] S52 embeds the multi-head attention module into a BiLSTM-based LSTM unit structure, and its calculation formula is as follows:

[0056] f(t) = MultiHead(σ(w) f x(t)+w f h(t)+b f ))

[0057] i(t) = MultiHead(σ(w) i x(t)+w i h(t-1)+b i ))

[0058] o(t)=σ(w o x(t)+w o h(t-1)+b o )

[0059]

[0060] h(t)=o(t)⊙tanh(c(t))

[0061] Where f(t), i(t), and o(t) are the outputs of the forget gate, input gate, and output gate, respectively. Since the forget gate and input gate structures contain multi-head attention modules, the outputs f(t) and i(t) become tensors after attention reconstruction and participate in the calculation of the current state c(t) and the output state h(t); x(t) is the input layer at time t, i.e., the state monitoring quantity obtained after singular value decomposition in step S3; h(t-1) is the hidden layer at time t-1; h(t) is the hidden layer at time t; w f w i w o w c These are the weight matrices for the forget gate, input gate, output gate, and hidden layer, respectively; b f b i b o b c These are the forget gate, input gate, output gate, and bias term of the hidden layer, respectively; σ and tanh are both activation functions, σ is the sigmoid function, and tanh is the hyperbolic tangent function; in the BiLSTM structure, attention modules exist in both directions of the information flow and are used to adjust the weights of degenerate information;

[0062] S53, when training the prediction model, the root mean square error function is selected, and the formula is as follows:

[0063]

[0064] in, Let y represent the predicted value of the intrinsic modulus component in the i-th data to be predicted. i represents the actual value of the intrinsic modulus component in the i-th data to be predicted, n is the number of data points, the network is initialized by randomly initializing weights and biases, the parameters are updated by gradient descent algorithm until the loss function converges, and the final prediction model, i.e. global degradation trend model, is obtained.

[0065] S54 inputs the preprocessed new state monitoring values ​​into the global degradation trend model and outputs the intrinsic modulus components that reflect the global degradation trend.

[0066] Optionally, step S6 specifically includes the following steps:

[0067] S61, the data sequence r = {r1, r2, ..., r} of the residual components of the degradation features obtained in step S4 is processed. n} represents the set of observed states, and the data sequence of state monitoring quantities obtained in step S3 is x = {x1, x2, ..., x...} n Let} represent the set of hidden states, where n represents the number of state values;

[0068] S62, the Baum-Welch algorithm is used to estimate the model parameters, as shown in the following formula:

[0069] λ=(A,B,P)

[0070] Where λ is the local fluctuation prediction model, A={a ij} N×N It is the state transition matrix, where B is the probability of the output observation and P is the probability of the initial state;

[0071] The training parameters P and A are solved using the Baum-Welch algorithm to obtain the expected value P(r). t When |λ) is at its maximum, P and A correspond to it; calculate the output observation probability B using the following formula:

[0072]

[0073] Where j represents the state, K represents the number of Gaussian distributions, and φ jl μ represents the weight of the l-th Gaussian component. jl and Σ jl These are the mean and covariance values, respectively.

[0074] S63, Use the Viterbi algorithm to find the optimal hidden state sequence x corresponding to the observation sequence. t Specifically:

[0075] S631, the calculation formula is as follows:

[0076] δ1(i)=p i b i (r1), 1≤i≤N

[0077] ψ1(i)=0,1≤i≤N

[0078] Where N is the number of states, p i and b i It is the i-th element of the model parameters P and B, and δ1(i) and ψ1(i) are intermediate variables;

[0079] S632, recursively calculates the maximum probability of all individual paths and their nodes, using the following formula:

[0080]

[0081] Where N is the number of states, T is the total number of monitored moments, and δ t (i) represents all δ values ​​from the previous time step. t-1 (i) and element a in A ij and element b in B i (r t The maximum value of the product, where 1≤j≤N, ψ t(i) is what makes δ t-1 (i)a ij The index corresponding to the maximum;

[0082] S633, terminates the iteration, the calculation formula is as follows:

[0083]

[0084] Where N is the number of states, P * It is all ψ t The maximum value of the elements in (i), x T * Is to make ψ t (i) The index corresponding to the maximum value;

[0085] S634, perform optimal path backtracking to obtain the final optimal hidden state sequence, calculated as follows:

[0086]

[0087] Where, x t * It is the optimal hidden state sequence, where T is the total number of monitored times;

[0088] S64 inputs the preprocessed new state monitoring values ​​into the local fluctuation prediction model and outputs local fluctuations.

[0089] Optionally, step S7 specifically includes the following steps:

[0090] S71 utilizes a weighted fusion method to process the prediction outputs of the global degradation trend model and the uncertainty model, thereby achieving degradation process modeling;

[0091] S72, Predicted time t k Remaining lifetime L at the location k The calculation formula is as follows:

[0092] L k =inf{l k :X(l k +t k )>w}

[0093] Where X is the degraded data value at a certain moment, w is the failure threshold, and l k Indicates the remaining lifespan in hours.

[0094] The beneficial effects of this invention are as follows: Addressing the uncertainty problem in data-driven models, this invention proposes a life prediction method for steer-by-wire systems that combines a multi-head attention mechanism (Multi-HeadAttention-Bi-directional Long Short-Term Memory) with a Hidden Markov Model, considering quantified uncertainty. Degradation data is decomposed into global degradation trends and local fluctuations using Complete Ensemble Empirical Mode Decomposition (CEEMDAN). Accordingly, MHA-BiLSTM is used to predict the global degradation trend, and the Hidden Markov Model is used to predict the local fluctuations caused by uncertainty. Finally, the prediction outputs are fused to achieve life prediction of the steer-by-wire system. Attached Figure Description

[0095] Figure 1 This is a flowchart illustrating the principle of the life prediction method for the steer-by-wire system proposed in this embodiment of the invention.

[0096] Figure 2 This is a flowchart of the multi-head attention calculation method for the life prediction method of the steer-by-wire system proposed in this embodiment of the invention;

[0097] Figure 3 This is a schematic diagram of the BiLSTM neural network with embedded multi-head attention in the life prediction method for the steer-by-wire system proposed in this embodiment of the invention.

[0098] Figure 4 This is a flowchart of the hidden Markov model prediction process for the life prediction method of the steer-by-wire system proposed in this embodiment of the invention.

[0099] The labels in the attached figures are as follows: Detailed Implementation

[0100] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0101] like Figure 1-4 As shown in the figure, this embodiment discloses a method for predicting the lifespan of a steer-by-wire system, including the following steps:

[0102] S1, determine the degradation characteristics affecting the lifespan of the steer-by-wire system, and collect the condition monitoring quantities and degradation characteristic quantities related to these degradation characteristics, specifically:

[0103] The state monitoring parameters for collecting degradation characteristics of the steer-by-wire system are assumed to be n state monitoring parameters, and a total of t time points are monitored. Then, the state monitoring parameter x is represented as:

[0104] x = {x1, x2, ..., x} n}

[0105] Where, x i =[x i1 ,x i2 ,...,x it ] T Let i = 1, 2, ..., n, then the state monitoring quantity x can be expressed as:

[0106]

[0107] The degradation characteristic quantity corresponding to the state monitoring quantity of the steer-by-wire system is determined as the accuracy of the steering system. Degradation data is collected and represented as: Y = [y1, y2, ..., y t ] T

[0108] Where yj represents the accuracy value of the steer-by-wire system at time j, j = 1, 2, ..., t. The n state monitoring quantities come from the steering wheel angle, the rotor angle and phase current of the steering wheel road feel motor, the steering actuation angle, the rotor angle and phase current of the actuation motor, the rack displacement, and the vibration and noise signals of the steering gear housing, etc.

[0109] S2, analyze the correlation between condition monitoring quantities and remaining life, calculate the correlation between each condition monitoring quantity and degradation characteristic quantity, and screen the condition monitoring quantities that can reflect the aging trend of the steering-by-wire system, specifically:

[0110] S21, calculate the correlation between each state monitoring quantity and the remaining lifetime. The formula for calculating the correlation coefficient R is:

[0111]

[0112] Where, x i For the i-th state monitoring sequence, For the state monitoring sequence x i The mean of x, where y is the sum of x and x. i The corresponding degradation characteristics, Let y be the mean of the time series y;

[0113] S22, Screening for highly correlated state monitoring quantities: The closer the correlation coefficient is to 1, the stronger the correlation between the state monitoring quantity and the degradation characteristic quantity. Based on this, the critical range of the correlation coefficient r is determined to be 0.9 to 1, and state monitoring quantities within this range are selected.

[0114] S3, using singular value decomposition to reduce noise in the filtered state monitoring quantities, specifically:

[0115] S31, based on phase space reconstruction theory, construct an m×n Hankel matrix H from the state monitoring quantities, and perform singular value decomposition on this matrix to obtain uncorrelated noise subspace and signal subspace, as shown in the formula:

[0116]

[0117] Where U represents an M×N orthogonal matrix; D represents the diagonal elements; D=diag(σ1,σ2,σ3......σ n ) represents an M×N diagonal matrix; V represents an N×N orthogonal matrix;

[0118] S32. Find the singular values ​​of the Hankel matrix A. Retain the singular values ​​in the diagonal matrix that reflect the effective signal, and set the smaller singular values ​​reflecting the noise signal to zero to initially remove noise from the high-frequency signal. Then perform the inverse process of singular value decomposition to finally obtain the matrix. Then the matrix The best approximation moment of rank r for H is calculated using the following formula:

[0119]

[0120] Where, σ i r represents the singular values ​​of matrix A; r represents the rank of matrix H; T represents the transpose; u i Representing matrix AA T The i-th eigenvector; v i Representing matrix A T The i-th eigenvector of A;

[0121] matrix The anti-diagonal elements in the data are added together and averaged to obtain the denoised state monitoring quantity.

[0122] S4. CEEMDAN is used to decompose the degradation features into intrinsic mode components (IMFs) and residual components. The IMFs reflect the global degradation trend, while the residual components reflect local fluctuations caused by uncertainty. The degradation process of a steering-by-wire system is essentially a nonlinear, non-smooth sequence, following a global degradation trend accompanied by local fluctuations caused by uncertain operating conditions. CEEMDAN decomposition adds an IMF component with auxiliary noise after EMD decomposition, instead of directly adding Gaussian white noise to the original signal, thus solving the problem of white noise always existing in the IMFs obtained from EMD. Therefore, CEEMDAN is used to decompose the degradation features into global degradation trend and local fluctuations, which are then predicted separately. Specifically:

[0123] S41, Gaussian white noise ε0E1(ω i(t)) Add the signal to be decomposed x(t), and then perform empirical mode decomposition to obtain the first intrinsic mode components IMF1. i (t), for IMF1 i The first-order intrinsic mode component IMF1(t) is obtained by averaging (t), and its formula is:

[0124]

[0125] Where N is the number of intrinsic modulus components, ω i (t) represents the white noise added for the i-th time, ε i The standard deviation of white noise;

[0126] S42, calculate the first-order residual component r1(t), the formula is:

[0127] r1(t) = x(t) - IMF1(t)

[0128] S43, continue adding Gaussian white noise to r1(t), for the new signal r1(t)+ε1E1(ω i (t) is subjected to empirical mode decomposition to obtain the second-order eigenmode components:

[0129]

[0130] Wherein, E1(r1(t)+ε1E1(w) i )) represents performing an empirical mode decomposition;

[0131] S44, and so on, yields the k-th order residual r. k (t), the formula is:

[0132] r k (t)=r k-1 (t)-IMF k (t)

[0133] S45, calculate the (k+1)th order eigenmode component as follows:

[0134]

[0135] The signal is decomposed cyclically until the obtained residual is a monotonic function and cannot be further decomposed. The formula for the final residual component is:

[0136]

[0137] The obtained intrinsic modulus components It reflects the global degradation trend, while the residual component r(t) reflects the local fluctuations caused by uncertainty.

[0138] S5, global degradation trend modeling is performed by inputting intrinsic mode components into the MHA-BiLSTM model to predict the global degradation trend. For mechanical parts, the degradation information contained in the monitoring signal changes throughout their service life and usually becomes more pronounced with increasing degradation. The purpose of the attention module is to enable the network to learn hidden knowledge in time-series data closely related to the degradation process and to suppress redundant information in the feature learning process. Reflected in the internal computation of the structure, the attention matrix is ​​constructed by assigning different weights to each incoming degradation data, thereby achieving matrix reconstruction, specifically:

[0139] S51, constructing a multi-head attention module, specifically:

[0140] S511, assuming the tensor input to the module is `input`, after the linear transformation layer, three tensors of the same size as the input are mapped: Q, K, and V. These three tensors are then fed into the scaled dot product attention calculation part to obtain the attention weights and the reconstructed attention matrix, calculated as follows:

[0141]

[0142] S512 divides tensors Q, K, and V into h equal-sized segments along the same direction (usually the last dimension of the tensor). The attention matrix is ​​calculated by placing each segment of the tensor in a different subspace. After calculation, the matrices of these subspaces are stitched together along the division direction to obtain the multi-head attention matrix, which is calculated as follows:

[0143] Multihead(Q,K,V)=Concat(head1,head2,…,head h W O

[0144] Among them, head i Let W be the attention matrix computed in the i-th subspace, and Concat be the concatenation operation of these sub-attention matrices along the splitting direction. O The parameter matrix for linear transformation; the structure of the multi-head attention module is as follows: Figure 2 As shown.

[0145] S52, the multi-head attention module is embedded into a BiLSTM-based LSTM unit structure. Considering the favorable characteristics of the multi-head attention mechanism for sequence learning, the multi-head attention module is embedded into a BiLSTM-based LSTM unit structure. The unit structure is as follows: Figure 3 As shown, the calculation formula is as follows:

[0146] f(t) = MultiHead(σ(w) f x(t)+w f h(t)+b f ))

[0147] i(t) = MultiHead(σ(w) i x(t)+w i h(t-1)+b i ))

[0148] o(t)=σ(w o x(t)+w o h(t-1)+b o )

[0149]

[0150] h(t)=o(t)⊙tanh(c(t))

[0151] Where f(t), i(t), and o(t) are the outputs of the forget gate, input gate, and output gate, respectively. Since the forget gate and input gate structures contain multi-head attention modules, the outputs f(t) and i(t) become tensors after attention reconstruction and participate in the calculation of the current state c(t) and the output state h(t); x(t) is the input layer at time t, i.e., the state monitoring quantity obtained after singular value decomposition in step S3; h(t-1) is the hidden layer at time t-1; h(t) is the hidden layer at time t; w f w i w o w c These are the weight matrices for the forget gate, input gate, output gate, and hidden layer, respectively; b f b i b o b c These are the forget gate, input gate, output gate, and bias term of the hidden layer, respectively; σ and tanh are both activation functions, σ is the sigmoid function, and tanh is the hyperbolic tangent function; in the BiLSTM structure, attention modules exist in both directions of the information flow and are used to adjust the weights of degenerate information;

[0152] S53, when training the prediction model, the root mean square error function is selected, and the formula is as follows:

[0153]

[0154] in, Let y represent the predicted value of the intrinsic modulus component in the i-th data to be predicted. irepresents the actual value of the intrinsic modulus component in the i-th data to be predicted, n is the number of data points, the network is initialized by randomly initializing weights and biases, the parameters are updated by gradient descent algorithm until the loss function converges, and the final prediction model, i.e. global degradation trend model, is obtained.

[0155] S54 inputs the preprocessed new state monitoring values ​​into the global degradation trend model and outputs the intrinsic modulus components that reflect the global degradation trend.

[0156] S6, local fluctuations are modeled by inputting the residual components into the Hidden Markov Model to predict local fluctuations, specifically:

[0157] S61, the data sequence r = {r1, r2, ..., r} of the residual components of the degradation features obtained in step S4 is processed. n} represents the set of observed states, and the data sequence of state monitoring quantities obtained in step S3 is x = {x1, x2, ..., x...} n Let} represent the set of hidden states, where n represents the number of state values;

[0158] S62, the Baum-Welch algorithm is used to estimate the model parameters, as shown in the following formula:

[0159] λ=(A,B,P)

[0160] Where λ is the local fluctuation prediction model, A={a ij} N×N It is the state transition matrix, where B is the probability of the output observation and P is the probability of the initial state;

[0161] The training parameters P and A are solved using the Baum-Welch algorithm to obtain the expected value P(r). t When |λ) is at its maximum, P and A correspond to it; calculate the output observation probability B using the following formula:

[0162]

[0163] Where j represents the state, K represents the number of Gaussian distributions, and φ jl μ represents the weight of the l-th Gaussian component. jl and Σ jl These are the mean and covariance values, respectively.

[0164] S63, Use the Viterbi algorithm to find the optimal hidden state sequence x corresponding to the observation sequence. t Specifically:

[0165] S631, the calculation formula is as follows:

[0166] δ1(i)=p i bi (r1), 1≤i≤N

[0167] ψ1(i)=0,1≤i≤N

[0168] Where N is the number of states, p i and b i It is the i-th element of the model parameters P and B, and δ1(i) and ψ1(i) are intermediate variables;

[0169] S632, recursively calculates the maximum probability of all individual paths and their nodes, using the following formula:

[0170]

[0171] Where N is the number of states, T is the total number of monitored moments, and δt(i) is the sum of all δ values ​​from the previous moment. t-1 (i) and element a in A ij and element b in B i (r t The maximum value of the product, where 1≤j≤N, ψ t (i) is what makes δ t-1 (i)a ij The index corresponding to the maximum;

[0172] S633, terminates the iteration, the calculation formula is as follows:

[0173]

[0174] Where N is the number of states, P * It is all ψ t The maximum value of the elements in (i), x T * Is to make ψ t (i) The index corresponding to the maximum value;

[0175] S634, perform optimal path backtracking to obtain the final optimal hidden state sequence, calculated as follows:

[0176]

[0177] Where, x t * It is the optimal hidden state sequence, where T is the total number of monitored times;

[0178] S64 inputs the preprocessed new state monitoring values ​​into the local fluctuation prediction model and outputs local fluctuations.

[0179] S7 integrates the predicted outputs of the global degradation trend model and the local fluctuation model to predict the degradation process of the steering system, specifically:

[0180] S71 utilizes a weighted fusion method to process the prediction outputs of the global degradation trend model and the uncertainty model, thereby achieving degradation process modeling;

[0181] S72, Predicted time t k Remaining lifetime L at the location k The calculation formula is as follows:

[0182] L k =inf{l k :X(l k +t k )>w}

[0183] Where X is the degraded data value at a certain moment, w is the failure threshold, and l k Indicates the remaining lifespan in hours.

[0184] The degradation data is decomposed into global degradation trend and local fluctuations using CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise). Accordingly, MHA-BiLSTM is used to predict the global degradation trend, and a Hidden Markov Model is used to predict the local fluctuations caused by uncertainty. Finally, the prediction outputs are fused to achieve the life prediction of the steer-by-wire system.

[0185] It is understood that the specific embodiments described above are merely for explaining the relevant invention and not for limiting the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Multiple technical solutions in the same embodiment, as well as multiple technical solutions in different embodiments, can be arranged and combined to form new technical solutions that do not contradict or conflict with each other. Any equivalent structural transformations made based on the content of this specification and drawings, whether directly or indirectly applied to other related technical fields, are similarly included within the scope of protection of this invention.

Claims

1. A method for predicting the lifespan of a steer-by-wire system, characterized in that, Includes the following steps: S1, determine the degradation characteristics that affect the lifespan of the steer-by-wire system, and collect the condition monitoring quantities and degradation characteristic quantities related to the degradation characteristics; S2, analyze the correlation between condition monitoring quantities and remaining life, calculate the correlation between each condition monitoring quantity and degradation characteristic quantity, and screen the condition monitoring quantities that can reflect the aging trend of the steering-by-wire system. S3, using singular value decomposition to reduce noise in the filtered state monitoring quantities; S4. Using CEEMDAN, the degradation features are decomposed into intrinsic modulus components and residual components. The intrinsic modulus components reflect the global degradation trend, while the residual components reflect the local fluctuations caused by uncertainty. S5, global degradation trend modeling is performed by inputting intrinsic mode components into the MHA-BiLSTM model to predict global degradation trend; S6, Local fluctuations are predicted by inputting residual components into a hidden Markov model to model local fluctuations; S7 integrates the predicted outputs of the global degradation trend model and the local fluctuation model to predict the degradation process of the steering system.

2. The method for predicting the lifespan of a steer-by-wire system according to claim 1, characterized in that, Step S1 specifically includes the following steps: The state monitoring parameters for collecting degradation characteristics of the steer-by-wire system are assumed to be n state monitoring parameters, and a total of t time points are monitored. Then, the state monitoring parameter x is represented as: x={x1,x2,...,x n } Where, x i =[x i1 ,x i2 ,...,x it ] T Let i = 1, 2, ..., n, then the state monitoring quantity x is represented as: The degradation characteristic quantity corresponding to the state monitoring quantity of the steer-by-wire system is determined as the accuracy of the steering system. The degradation characteristic quantity is collected and represented as: Y=[y1,y2,...,y t ] T Where yj represents the accuracy value of the steer-by-wire system at time j, j = 1, 2, ..., t.

3. The method for predicting the lifespan of a steer-by-wire system according to claim 2, characterized in that, Step S2 specifically includes the following steps: S21, calculate the correlation between each state monitoring quantity and the remaining lifetime. The formula for calculating the correlation coefficient R is: Where, x i For the i-th state monitoring sequence, For the state monitoring sequence x i The mean of x, where y is the sum of x and x. i The corresponding degradation characteristics, Let y be the mean of the time series y; S22, Screening for highly correlated state monitoring quantities: The closer the correlation coefficient is to 1, the stronger the correlation between the state monitoring quantity and the degradation characteristic quantity. Based on this, the critical range of the correlation coefficient r is determined to be 0.9 to 1, and state monitoring quantities within this range are selected.

4. The method for predicting the lifespan of a steer-by-wire system according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31, based on phase space reconstruction theory, construct an m×n Hankel matrix H from the state monitoring quantities, and perform singular value decomposition on this matrix to obtain uncorrelated noise subspace and signal subspace, as shown in the formula: Where U represents an M×N orthogonal matrix; D represents the diagonal elements; D=diag(σ1,σ2,σ3......σ n ) represents an M×N diagonal matrix; V represents an N×N orthogonal matrix; S32. Find the singular values ​​of the Hankel matrix A. Retain the singular values ​​in the diagonal matrix that reflect the effective signal, and set the smaller singular values ​​reflecting the noise signal to zero to initially remove noise from the high-frequency signal. Then perform the inverse process of singular value decomposition to finally obtain the matrix. Then the matrix The best approximation moment of rank r for H is calculated using the following formula: Where, σ i r represents the singular values ​​of matrix A; r represents the rank of matrix H; T represents the transpose; u i Representing matrix AA T The i-th eigenvector; v i Representing matrix A T The i-th eigenvector of A; matrix The anti-diagonal elements in the data are added together and averaged to obtain the denoised state monitoring quantity.

5. The method for predicting the lifespan of a steer-by-wire system according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41, Gaussian white noise ε0E1(ω i (t)) The signal to be decomposed x(t) is added, and then empirical mode decomposition is performed to obtain the first eigenmode components. right By performing an average calculation, the first-order intrinsic mode component IMF1(t) is obtained, and its formula is as follows: Where N is the number of intrinsic modulus components, ω i (t) represents the white noise added for the i-th time, ε i The standard deviation of white noise; S42, calculate the first-order residual component r1(t), the formula is: r1(t) = x(t) - IMF1(t) S43, continue adding Gaussian white noise to r1(t), for the new signal r1(t)+ε1E1(ω i (t) is subjected to empirical mode decomposition to obtain the second-order eigenmode components: Wherein, E1(r1(t)+ε1E1(w) i )) represents performing an empirical mode decomposition; S44, and so on, yields the k-th order residual r. k (t), the formula is: r k (t)=r k-1 (t)-IMF k (t) S45, calculate the (k+1)th order eigenmode component as follows: The signal is decomposed cyclically until the obtained residual is a monotonic function and cannot be further decomposed. The formula for the final residual component is: The obtained intrinsic modulus components It reflects the global degradation trend, while the residual component r(t) reflects the local fluctuations caused by uncertainty.

6. The method for predicting the lifespan of a steer-by-wire system according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51, constructing a multi-head attention module, specifically: S511, assuming the tensor input to the module is `input`, after the linear transformation layer, three tensors of the same size as the input are mapped: Q, K, and V. These three tensors are then fed into the scaled dot product attention calculation part to obtain the attention weights and the reconstructed attention matrix. The calculation formula is as follows: S512 divides tensors Q, K, and V into h equal-sized segments along the same direction. The attention matrix is ​​calculated by placing each segment of the tensor in a different subspace. After calculation, the matrices of these subspaces are stitched together along the division direction to obtain the multi-head attention matrix, which is calculated as follows: Multihead(Q,K,V)=Concat(head1,head2,…,head h )W O Among them, head i Let W be the attention matrix computed in the i-th subspace, and Concat be the concatenation operation of these sub-attention matrices along the splitting direction. O The parameter matrix for the linear transformation; S52 embeds the multi-head attention module into a BiLSTM-based LSTM unit structure, and its calculation formula is as follows: f(t)=MultiHead(σ(w f x(t)+w f h(t)+b f )) i(t)=MultiHead(σ(w i x(t)+w i h(t-1)+b i )) o(t)=σ(w o x(t)+w o h(t-1)+b o ) h(t)=o(t)⊙tanh(c(t)) Where f(t), i(t), and o(t) are the outputs of the forget gate, input gate, and output gate, respectively. Since the forget gate and input gate structures contain multi-head attention modules, the outputs f(t) and i(t) become tensors after attention reconstruction and participate in the calculation of the current state c(t) and the output state h(t); x(t) is the input layer at time t, i.e., the state monitoring quantity obtained after singular value decomposition in step S3; h(t-1) is the hidden layer at time t-1; h(t) is the hidden layer at time t; w f w i w o w c These are the weight matrices for the forget gate, input gate, output gate, and hidden layer, respectively; b f b i b o b c These are the forget gate, input gate, output gate, and bias term of the hidden layer, respectively; σ and tanh are both activation functions, σ is the sigmoid function, and tanh is the hyperbolic tangent function; in the BiLSTM structure, attention modules exist in both directions of the information flow and are used to adjust the weights of degenerate information; S53, when training the prediction model, the root mean square error function is selected, and the formula is as follows: in, Let y represent the predicted value of the intrinsic modulus component in the i-th data to be predicted. i represents the actual value of the intrinsic modulus component in the i-th data to be predicted, n is the number of data points, the network is initialized by randomly initializing weights and biases, the parameters are updated by gradient descent algorithm until the loss function converges, and the final prediction model, i.e. global degradation trend model, is obtained. S54 inputs the preprocessed new state monitoring values ​​into the global degradation trend model and outputs the intrinsic modulus components that reflect the global degradation trend.

7. The method for predicting the lifespan of a steer-by-wire system according to claim 3, characterized in that, Step S6 specifically includes the following steps: S61, the data sequence r = {r1, r2, ..., r} of the residual components of the degradation features obtained in step S4 is processed. n } represents the set of observed states, and the data sequence of state monitoring quantities obtained in step S3 is x = {x1, x2, ..., x...} n Let} represent the set of hidden states, where n represents the number of state values; S62, the Baum-Welch algorithm is used to estimate the model parameters, as shown in the following formula: λ=(A,B,P) Where λ is the local fluctuation prediction model, A={a ij } N×N It is the state transition matrix, where B is the probability of the output observation and P is the probability of the initial state; The training parameters P and A are solved using the Baum-Welch algorithm to obtain the expected value P(r). t When |λ) is at its maximum, P and A correspond to it; calculate the output observation probability B using the following formula: Where j represents the state, K represents the number of Gaussian distributions, and φ jl μ represents the weight of the l-th Gaussian component. jl and Σ jl These are the mean and covariance values, respectively. S63, Use the Viterbi algorithm to find the optimal hidden state sequence x corresponding to the observation sequence. t Specifically: S631, the calculation formula is as follows: δ1(i)=p i b i (r1),1≤i≤N ψ1(i)=0,1≤i≤N Where N is the number of states, p i and b i It is the i-th element of the model parameters P and B, and δ1(i) and ψ1(i) are intermediate variables; S632, recursively calculates the maximum probability of all individual paths and their nodes, using the following formula: Where N is the number of states, T is the total number of monitored moments, and δ t (i) represents all δ values ​​from the previous time step. t-1 (i) and element a in A ij and element b in B i (r t The maximum value of the product, where 1≤j≤N, ψ t (i) is what makes δ t-1 (i)a ij The index corresponding to the maximum; S633, terminates the iteration, the calculation formula is as follows: Where N is the number of states, P * It is all ψ t The maximum value of the elements in (i), Is to make ψ t (i) The index corresponding to the maximum value; S634, perform optimal path backtracking to obtain the final optimal hidden state sequence, calculated as follows: Where, x t * It is the optimal hidden state sequence, where T is the total number of monitored times; S64: The new state monitoring values ​​are preprocessed and then input into the local fluctuation prediction model, which outputs local fluctuations.

8. The method for predicting the lifespan of a steer-by-wire system according to claim 1, characterized in that, Step S7 specifically includes the following steps: S71 utilizes a weighted fusion method to process the prediction outputs of the global degradation trend model and the uncertainty model, thereby achieving degradation process modeling; S72, Predicted time t k Remaining lifetime L at the location k The calculation formula is as follows: L k =inf{l k :X(l k +t k )>w} Where X is the degraded data value at a certain moment, w is the failure threshold, and l k Indicates the remaining lifespan in hours.

Images