Method for predicting the life of an electromagnetic directional control valve based on flow signals

Abstract

The present application relates to a kind of electromagnetic reversing valve life prediction method based on flow signal, belong to hydraulic component life prediction field.The present application utilizes improved lumped average modal empirical decomposition method, by adding positive and negative pairs of noise to reduce the degree of modal aliasing in modal decomposition, using permutation entropy to detect abnormal component, realize the accurate adaptive modal decomposition of nonlinear measured signal;Application kernel principal component method, introduce nonlinear function as kernel function, based on the principle of mapping, convert original space into high-dimensional space, form new data set, using principal component analysis for data dimension reduction in high-dimensional data space, form feature vector, get performance degradation fusion index;Through cubic exponential smoothing processing;Finally based on the trained adaptive neural network model, establish the life prediction model of electromagnetic reversing valve, calculate the life of electromagnetic reversing valve.This method can effectively predict the pressure drop trend and life of electromagnetic reversing valve.

Description

Technical Field

[0001] This invention belongs to the field of hydraulic component life prediction technology, specifically relating to a method for predicting the life of an electromagnetic directional valve based on flow signals. Background Technology

[0002] Solenoid directional control valves are used for switching hydraulic lines and changing the direction of actuation of actuators. They are fundamental components of hydraulic systems, widely used in engineering due to their simple operating principle and low requirements for hydraulic precision. However, in complex electromechanical-hydraulic systems, solenoid directional control valves are prone to failure during prolonged high-speed operation. Failure can lead to anything from downtime and substandard work quality to catastrophic accidents involving personnel safety, with potentially disastrous consequences. Therefore, reliability assessment and lifespan prediction of solenoid directional control valves can effectively prevent losses caused by failures and maintain the normal operation of the hydraulic system. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] The technical problem to be solved by the present invention is how to provide a method for predicting the life of an electromagnetic directional valve based on flow signals, so as to solve the problem of reliability assessment and life prediction of electromagnetic directional valves.

[0005] (II) Technical Solution

[0006] To address the aforementioned technical problems, this invention proposes a method for predicting the lifespan of an electromagnetic directional valve based on flow signals. This method includes the following steps:

[0007] Step S1: Obtain the initial state data of the solenoid directional valve;

[0008] The flow and pressure signals at the inlet and outlet of the electromagnetic directional valve are collected using a flow meter and a pressure sensor; the leakage signal is obtained by subtracting the inlet and outlet flow signal data; and the pressure drop signal is obtained using wavelet packet decomposition. The initial state data includes both leakage and pressure drop signal data.

[0009] Step S2: Obtain reconstructed signal data based on the improved empirical mode decomposition method;

[0010] The leakage signal data is processed using an improved empirical mode decomposition method to obtain stable mode function components and reconstruct the signal data.

[0011] Step S3: Based on the kernel principal component method and adaptive neural network model, establish a life prediction model for the electromagnetic directional valve;

[0012] Step S4: Set the failure pressure drop threshold to predict the lifespan of the solenoid directional valve;

[0013] Obtain the life prediction model of the electromagnetic directional valve established in step S3, and use the failure pressure drop threshold to predict the life of the electromagnetic directional valve.

[0014] (III) Beneficial Effects

[0015] This invention proposes a method for predicting the lifespan of electromagnetic directional valves based on flow signals. Compared with existing technologies, the advantages of this invention are as follows:

[0016] (1) In this embodiment of the invention, the inlet and outlet pressure drop of the electromagnetic directional valve is used as the failure criterion of the electromagnetic directional valve. Based on the relationship between flow rate, leakage and pressure drop, the leakage signal data is decomposed into multiple scales using an improved empirical mode decomposition method. The noise reduction and reconstruction are performed using mode function component screening rules. This can improve the adaptability of the mode decomposition process, reduce the probability of mode aliasing, increase reconstruction accuracy and improve computational efficiency. It is more suitable for signal processing of electromagnetic directional valves. The comparative calculation results show that this method has better computational accuracy.

[0017] (2) The embodiments of the present invention apply the kernel principal component method for fusion processing, which can reduce the dimensionality of data while ensuring effective representation, fuse detailed features with high contribution rate, construct a fusion index for the performance degradation of electromagnetic directional valve, and reduce the impact of random interference on data training by three exponential smoothing processes, thereby improving the accuracy of the index; linear prediction of the degradation evaluation index of electromagnetic directional valve is performed using three smoothing prediction, the optimal degradation fusion index is selected, the degradation evaluation model of electromagnetic directional valve is completed, adaptive fuzzy neural network learning and training are performed, the pressure drop trend of electromagnetic directional valve is effectively predicted, and the life prediction result of electromagnetic directional valve is obtained by using the failure pressure drop criterion; the superiority of this method can also be seen by comparing the prediction results of different methods. Attached Figure Description

[0018] Figure 1 This is a control block diagram of the electromagnetic directional valve life prediction method based on flow signal of the present invention.

[0019] Figure 2 This is a flowchart illustrating the implementation of the electromagnetic directional valve life prediction method according to an embodiment of the present invention.

[0020] Figure 3 The results are the prediction results of the linear fitting algorithm, the cubic exponent and the adaptive neural network calculation method in the embodiments of the present invention. Detailed Implementation

[0021] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0022] By analyzing and discussing the failure modes and possible fault categories of electromagnetic directional valves, a valve core motion simulation model was built to simulate typical faults, clarifying that a decrease in valve core displacement is a direct representation of the electromagnetic directional valve's fault state. Dynamic characteristics and internal flow field characteristics of the electromagnetic directional valve were simulated and analyzed, revealing that under the same operating pressure, the pressure drop at the inlet and outlet of the electromagnetic directional valve increases significantly with the increase of the valve opening; similarly, under the same displacement, the pressure drop at the inlet and outlet also tends to increase with the increase of the operating pressure. Based on these experimental conclusions, using the pressure drop at the inlet and outlet of the electromagnetic directional valve as the failure criterion, the degradation state of the electromagnetic directional valve was analyzed from the perspective of leakage. A degradation fusion index characterizing the health state of the electromagnetic directional valve was constructed, and the ANFIS algorithm was introduced to complete the life prediction of the electromagnetic directional valve.

[0023] To overcome the shortcomings of existing technologies, this invention utilizes an improved lumped-means mode empirical decomposition method. By adding paired positive and negative noise, the degree of mode aliasing in mode decomposition is reduced. Simultaneously, permutation entropy is used to detect anomalous components, achieving accurate adaptive mode decomposition of nonlinear measured signals. The kernel principal component method is applied, introducing a nonlinear function as the kernel function. Based on the principle of mapping, the original space is transformed into a high-dimensional space, forming a new dataset. Principal component analysis is then used in the high-dimensional data space for dimensionality reduction, ultimately forming feature vectors and obtaining performance degradation fusion indices. Furthermore, triple exponential smoothing is applied to reduce the impact of the volatility and sparsity of the degradation evaluation indices on classifier training and discrimination, improving the accuracy of the indices. Finally, based on the trained adaptive neural network model, a life prediction model for electromagnetic directional valves is established to calculate the lifespan of the electromagnetic directional valves.

[0024] To achieve the above objectives, the solution adopted by the present invention is as follows:

[0025] A method for predicting the lifespan of an electromagnetic directional valve based on flow signals, comprising the following steps:

[0026] Step S1: Obtain the initial state data of the solenoid directional valve;

[0027] The flow and pressure signals at the inlet and outlet of the electromagnetic directional valve are collected using a flow meter and a pressure sensor; the leakage signal is obtained by subtracting the inlet and outlet flow signal data; and the pressure drop signal is obtained using wavelet packet decomposition. The initial state data includes both leakage and pressure drop signal data.

[0028] Step S2: Obtain reconstructed signal data based on the improved empirical mode decomposition method;

[0029] The leakage signal data is processed using an improved empirical mode decomposition method to obtain stable mode function components and reconstruct the signal data.

[0030] Step S21: Introduce white noise into the leakage signal data, and process the leakage signal data from step S1 using an improved empirical mode decomposition method. The method for obtaining the average value of the leakage signal components is as follows:

[0031]

[0032] In the formula: I1(t) represents the average value of the leakage signal components; N represents the number of wavelet packet decomposition layers; Ne represents the total number of processed signals; n represents the number of processed signals. This represents the processed signal after introducing positive white noise; This represents the processed signal after introducing negative white noise; t represents the time variable.

[0033] Step S22: Perform mode decomposition on the average value of the leakage signal components to filter out the leakage signal components whose permutation entropy exceeds a fixed threshold; repeat the filtering process until the steady-state leakage signal components are obtained.

[0034] Step S23: Use the improved empirical mode decomposition method to separate the modal components in the leakage signal data to obtain all modal function components; introduce the KL divergence method to screen all modal function components to obtain stable modal function components, obtain a clearer reconstructed signal, and form characteristic parameters;

[0035] Step S3: Based on the kernel principal component method and the adaptive neural network model, establish a life prediction model for the electromagnetic directional valve;

[0036] Step S31: Use the kernel principal component method to fuse the characteristic parameters of the reconstructed signal and establish a mapping model of the electromagnetic commutator degradation fusion index on the voltage drop;

[0037] To preserve the detailed features of the electromagnetic directional valve, the kernel principal component method is used to fuse the extracted feature parameters. Based on the principle of kernel functions, the input space is projected to a high-dimensional feature space through nonlinear mapping. Then, principal component analysis is performed in the high-dimensional feature space to obtain feature vectors. The obtained feature parameter samples are assembled into a dataset, as shown below:

[0038] X = [x1, x2, ..., x m ,…,x M ], x m ∈R, m=1,2,……,M;

[0039] In the formula: X represents the dataset composed of feature parameter samples; x m This represents the m-th sample dataset; M represents the number of samples.

[0040] Define a nonlinear mapping relationship Φ(f):R K →R DIf D >> K, then each independent vector in the dataset is mapped to a feature space F, resulting in a feature space scaling matrix, as shown in the following equation:

[0041] Φ(f)=[Φ(x1),Φ(x2),……,Φ(x P )];

[0042] In the formula: Φ(f) represents the characteristic space scale matrix; Φ(x) represents the characteristic space scale matrix. P () represents the p-th sample point in the kernel symmetric matrix; P represents the number of variables in an independent vector;

[0043] Calculate the feature space scale matrix based on the feature parameter dataset, construct the kernel symmetric matrix K, and define the kernel symmetric matrix K = {k}. ij} P×P Introducing the feature space F, we perform an inner product transformation of the vectors to reconstruct the kernel symmetric matrix K, as shown in the following equation:

[0044] K=(Φ(x i ),Φ(x j ));

[0045] In the formula: K represents the kernel symmetric matrix; Φ(x i ) represents the i-th sample point in the kernel symmetric matrix; Φ(x) j () represents the j-th sample point in the kernel symmetric matrix;

[0046] The kth, k = 1, 2, ..., Pth principal component T is calculated in the feature space F. k As shown in the following formula:

[0047]

[0048] In the formula: T K α represents the k-th principal element in the feature space F; v represents the eigenvector of the kernel symmetric matrix K; Φ(f) represents the scale matrix of the feature space; α mk This represents the constant coefficient of the k-th pivot element;

[0049] To reduce the impact of the volatility and sparsity of degradation evaluation indicators on classifier training and discrimination, a triple exponential smoothing method is used to establish a mapping model of the degradation fusion index of electromagnetic directional valve to pressure drop.

[0050] Step S32: The data from the optimal degradation point of pressure drop to the initial prediction position is used to form a training set, which is then input into an adaptive neural network for training to establish a life prediction model for the electromagnetic directional valve.

[0051] The adaptive neural network prediction model consists of five layers forming a complete feedforward network. The fuzzification layer is responsible for fuzzifying each input variable, transforming it into a fuzzy set, and obtaining the membership function of the fuzzy set. The rule inference layer is responsible for multiplying the input signals to obtain the excitation intensity of the fuzzy rules. The normalization layer normalizes the above excitation intensity. The defuzzification layer is responsible for defuzzifying the normalized result and outputting the accurate result. The final output layer is responsible for calculating the total output result of the entire network.

[0052] The Adaptive Neural Network (ANFIS) prediction model is a complete feedforward network that calculates the relative fuzzy set membership function of each input component. The input layer directly connects each node to the input component, and the connected components are passed to the membership calculation layer as new inputs to build a life prediction model for electromagnetic directional valves.

[0053] Step S4: Set the failure pressure drop threshold to predict the lifespan of the solenoid directional valve;

[0054] Obtain the life prediction model of the electromagnetic directional valve established in step S3, and use the failure pressure drop threshold to predict the life of the electromagnetic directional valve.

[0055] Preferably, the step S1 of obtaining the voltage drop signal data according to the wavelet packet decomposition method specifically involves:

[0056] After the pressure signal is decomposed into N layers of wavelet packets, the energy in the flow signal of the electromagnetic directional valve is evenly distributed across 2 N On these two orthogonal frequency bands N Each orthogonal frequency band contains energy reflecting state information. The method for obtaining voltage drop signal data is as follows:

[0057]

[0058] In the formula: E Ns This represents the s-th voltage drop signal data; Represents the coefficients of the l-th wavelet packet; e represents the dimension number of the high-dimensional space; s and n represent the decomposition nodes; M e represents the dimension of the high-dimensional space; l represents the number of decomposition levels.

[0059] Preferably, in step S22, modal decomposition is performed on the average value of the leakage signal components to filter out leakage signal components whose permutation entropy exceeds a fixed threshold; this filtering is repeated until steady-state leakage signal components are obtained, specifically:

[0060] The first p-1 modal components corresponding to the original signal are decomposed as shown in the following equation:

[0061]

[0062] In the formula: r(t) represents the residual signal; S(t) represents the total signal; I h (t) represents the selected stable signal; p represents the total number of original signals;

[0063] The leakage signal components are grouped into a sequence {y(w), w = 1, 2, ..., c}, and their phase space is reconstructed:

[0064]

[0065] In the formula: c represents the embedding dimension; τ represents the time delay; Y represents the sequence of leakage signal components; y represents the elements in the sequence Y;

[0066] Given any vector Y(w), a sequence of symbols can be obtained, as shown below:

[0067] S(a) = [j1,j2,…,j c ];a=1,2,…,E,E≤c! ;

[0068] In the formula: S(a) represents the symbol sequence corresponding to the vector; a represents the symbol sequences that map to different c-dimensional phase spaces; j c E represents a single symbol corresponding to a vector; E represents the total number of symbol sequences.

[0069] The method for obtaining the permutation entropy of the leakage signal component sequence is as follows:

[0070]

[0071] In the formula: H p (c) represents the entropy of a non-normalized permutation;

[0072] For H p (c) Perform standardization as follows:

[0073] H p =H p (c) / ln(c!);

[0074] In the formula: H p Represents the permutation entropy;

[0075] The final permutation entropy H p The larger the value, the more disordered the time series; permutation entropy H p The smaller the value, the more regular the time series. The leakage signal components whose permutation entropy exceeds a fixed threshold are filtered out until the steady-state leakage signal components are obtained.

[0076] Preferably, in step S23, an improved empirical mode decomposition method is used to separate the modal components in the leakage signal data to obtain all modal function components; the KL divergence method is introduced to filter all modal function components to obtain stable modal function components, resulting in clearer reconstructed signal data, which forms feature parameters, specifically:

[0077] The obtained r(t) is decomposed using an improved empirical mode decomposition method to obtain all mode function components. Due to the occurrence of mode aliasing, empty mode function components will be displayed in the decomposed signal. A sequence Z = {z1, z2, ..., z...} is formed for all mode function components. c The sequence Y = y1, y2, ..., y1, y2, ..., y3, consisting of the leakage signal components and the leakage quantity signal components. c The effectiveness of the IMF is determined by calculating the differences between the sequences using the KL divergence method, as shown in the following formula:

[0078]

[0079] In the formula: δ(p,q) and δ(q,p) represent the KL distance between sequences Z and Y, respectively; p(z) represents the kernel density function of sequence Z; q(y) represents the probability distribution of sequence Y; z and y represent the elements in sequences Z and Y, respectively; C represents the sequence length;

[0080] The divergence values ​​of sequences Z and Y are obtained as shown in the following formula:

[0081] D(p, q)=δ(p, q)+δ(q, p);

[0082] In the formula: D(p, q) represents the divergence values ​​of sequences Z and Y;

[0083] After the leakage signal components are decomposed using an improved empirical mode decomposition method, the KL divergence values ​​of all obtained modal function components are calculated. This yields a set of similarity sequences for all modal function components. All similarity sequences are then filtered based on their correlation coefficient r and the inverse divergence value d. The selected sequences with r... max and d max The similarity sequences are combined to form the optimal solution sequence, as shown in the following equation:

[0084] λ=(r max d max );

[0085] In the formula: λ represents the optimal solution sequence; r max d represents the maximum correlation coefficient; max This represents the reciprocal of the maximum divergence value;

[0086] Substituting the optimal solution sequence into the Euclidean distance formula, as shown in the following equation:

[0087]

[0088] In the formula: γ i The Euclidean distance from the similarity sequence of modal function components to the optimal sequence; r i The correlation coefficient represents the modal function components; d i It represents the reciprocal of the divergence value of the modal function components;

[0089] Based on the results obtained above, stable mode function components are selected to obtain a clearer reconstructed signal.

[0090] Preferably, in step S31, a cubic exponential smoothing method is used to establish a mapping model between the degradation fusion index of the electromagnetic directional valve and the pressure drop, thereby processing the degradation estimation index of the electromagnetic directional valve and reducing the impact of interference signals on data training. Specifically:

[0091] For any sequence at time [X1, X2, ..., X...] v First, an exponential smoothing process is applied to obtain a first-order exponential sequence. Then set All sequences are observed and known sequences. Exponential smoothing is applied consecutively for prediction, and the result is... Repeat the exponential smoothing process three times to obtain the updated time series. in Given a time series [X1, X2, ..., X...] v The smoothing result of ] The time-shift exponential smoothing prediction results are obtained after three processing steps;

[0092]

[0093] In the formula: This represents the result of a single exponential smoothing. This indicates the result of quadratic exponential smoothing; This represents the result of cubic exponential smoothing; α represents the smoothing coefficient. X represents the smoothed value of the first v-1 data points in a single smoothing process; v This represents the actual observed value at time v; This represents the smoothed value of the first v-1 data points in the secondary smoothing process; This represents the smoothed value of the first v-1 data points in the three smoothing processes;

[0094] Different feature parameters contribute differently to the description of the degradation process. By introducing the overall contribution rate, the higher the value, the more spatial information the principal component contains. By arranging the contribution rates according to their magnitude and introducing smoothing principal components, the performance degradation index can represent the growth trend and describe the performance degradation process of the electromagnetic directional valve within a certain range. By constructing a mapping model between the electromagnetic directional valve degradation fusion index and the electromagnetic directional valve pressure drop, accurate prediction of the degradation fusion index can be achieved.

[0095] Preferably, in step S32, the relative fuzzy set membership function of each input component is calculated. The input layer directly connects each node to the input component, and the connected components are transmitted to the membership calculation layer as new inputs to construct the life prediction model of the electromagnetic directional valve. Specifically:

[0096] First, the method for obtaining the membership function of a fuzzy set is as follows:

[0097]

[0098] In the formula: Represents the membership function of a fuzzy set; a 1g b represents the subparameter of the first membership function; 1g c represents the exponential parameter of the first membership function; 1g Indicates the parameter set above the first membership function; a 2g b represents the subparameter of the second membership function; 2g c represents the exponential parameter of the second membership function. 2g This indicates the parameter set on the second membership function; Represents the first membership function; A represents the second membership function; g B represents the set of results from the first fuzzy processing. g o1 represents the second fuzzy processing result set; o2 represents the first input variable; g represents the second input variable; g represents the number of fuzzy sets.

[0099] Match the antecedents of the fuzzy rule and calculate the applicability of the fuzzy rule, as shown in the following formula:

[0100]

[0101] In the formula: Indicates the applicability of fuzzy rules; ω represents the membership degree of a fuzzy set. g Indicates the intensity of the incentive;

[0102] After normalizing the overall network, the applicability of the fuzzy rule normalization is calculated as follows:

[0103] ;

[0104] In the formula: Indicates the applicability of fuzzy rule normalization;

[0105] The normalized network is subjected to overall fuzzification, and the applicability of the fuzzy rule standardization is calculated as follows:

[0106]

[0107] In the formula: Indicates the applicability of fuzzy rule standardization; f represents the applicability of fuzzy rule normalization; g p represents the set of consequent parameters; g Indicates the first consequent parameter; q g Indicates the second consequent parameter; r g Indicates the third consequent parameter;

[0108] Finally, the normalized triggering degree of the fuzzy rules for the entire grid is calculated, as shown in the following formula:

[0109]

[0110] In the formula: Indicates the normalization triggering degree of the fuzzy rule;

[0111] Based on the fuzzy set membership function, fuzzy rule applicability, fuzzy rule normalization applicability, fuzzy rule standardization applicability, and fuzzy rule normalization triggering degree, an adaptive neural network-based life prediction model for electromagnetic directional valves is established.

[0112] Example 1:

[0113] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0114] This invention provides a detailed method for predicting the lifespan of electromagnetic directional valves based on flow signals, such as... Figure 1 The diagram shows the control block diagram of the electromagnetic directional valve life prediction method based on flow signal. This embodiment utilizes an improved Empirical Mode Decomposition (MEEMD) method to decompose the measured signal data, and performs noise reduction and reconstruction using Modal Function (IMF) component selection rules to reduce the probability of mode aliasing, increase reconstruction accuracy, and improve computational efficiency. Kernel Principal Component Analysis (KPCA) is applied to achieve data dimensionality reduction and construct a performance degradation fusion index. Triple exponential smoothing reduces the impact of random interference on the data and improves the accuracy of the index. Based on a trained Adaptive Neural Network (ANFIS) model, a life prediction model for the electromagnetic directional valve is established to calculate its lifespan.

[0115] This invention provides a method for predicting the lifespan of an electromagnetic directional valve based on flow signals. Figure 2 This is a flowchart illustrating the implementation of the electromagnetic directional valve life prediction method according to an embodiment of the present invention. To demonstrate the applicability of the present invention, it is applied to an example, specifically including the following steps:

[0116] S1: Obtain the initial state data of the solenoid directional valve;

[0117] Pressure and flow signals at the inlet and outlet of the three-position four-way solenoid directional valve were collected using pressure sensors and flow meters to obtain pressure drop and leakage data. The pressure and flow sensors used in this test output 4-20mA current signals, while the vibration sensor outputs a 0-5V voltage signal. The working pressure was adjusted to 35MPa for factory testing of the solenoid directional valve, and initial operating data were recorded. The relief valve was then adjusted to raise the pressure to 37MPa, and the test was conducted normally with data acquisition. Leakage signal data was obtained by subtracting the inlet and outlet flow signal data. Pressure drop signal data was obtained using wavelet packet decomposition. The initial state data included both leakage and pressure drop signal data.

[0118] After the pressure signal is decomposed into N layers of wavelet packets, the energy in the flow signal of the electromagnetic directional valve is evenly distributed across 2 N On these two orthogonal frequency bands N Each orthogonal frequency band contains energy reflecting state information. The method for obtaining voltage drop signal data is as follows:

[0119]

[0120] In the formula: E Ns This represents the s-th voltage drop signal data; Represents the coefficients of the l-th wavelet packet; e represents the dimension number of the high-dimensional space; s represents the decomposition node; M e represents the dimension of the high-dimensional space; l represents the number of decomposition levels.

[0121] S2: Obtain reconstructed signal data based on the improved empirical mode decomposition method;

[0122] The leakage signal data is processed using an improved empirical mode decomposition method (MEEMD) to obtain stable mode function components and reconstruct the signal data.

[0123] S21: White noise is introduced into the leakage signal data. The improved Empirical Mode Decomposition (MEEMD) method is used to process the leakage signal data of S1. The method for obtaining the average value of the leakage signal components is as follows:

[0124]

[0125] In the formula: I1(t) represents the average value of the leakage signal components; N represents the number of wavelet packet decomposition layers; Ne represents the total number of processed signals; n represents the number of processed signals. This represents the processed signal after introducing positive white noise; This represents the processed signal after introducing negative white noise; t represents the time variable.

[0126] S22: Perform mode decomposition on the average value of the leakage signal components to filter out the leakage signal components whose permutation entropy exceeds a fixed threshold; repeat the filtering process until the steady-state leakage signal components are obtained.

[0127] The first p-1 modal components corresponding to the original signal are decomposed as shown in the following equation:

[0128]

[0129] In the formula: r(t) represents the residual signal; S(t) represents the total signal; I h (t) represents the selected stable signal; p represents the total number of original signals;

[0130] The leakage signal components are grouped into a sequence {y(w), w = 1, 2, ..., c}, and their phase space is reconstructed:

[0131]

[0132] In the formula: c represents the embedding dimension; τ represents the time delay; Y represents the sequence of leakage signal components; y represents the elements in the sequence Y;

[0133] Given any vector Y(w), a sequence of symbols can be obtained, as shown below:

[0134] S(a) = [j1,j2,…,j c ];a=1,2,…,E,E≤c! ;

[0135] In the formula: S(a) represents the symbol sequence; a represents the element number in the symbol sequence; j c E represents the elements in the symbol sequence; E represents the total number of symbols in the sequence.

[0136] The method for obtaining the permutation entropy of the leakage signal component sequence is as follows:

[0137]

[0138] Where: H pc Represents the entropy of a non-standardized permutation;

[0139] For H pc The standardization process is performed as follows:

[0140] H p =Hp ( c) / ln(c!);

[0141] Where: H p Represents the permutation entropy;

[0142] The final permutation entropy H p The larger the value, the more disordered the time series; permutation entropy H p The smaller the value, the more regular the time series. The leakage signal components whose permutation entropy exceeds a fixed threshold are filtered out until the steady-state leakage signal components are obtained.

[0143] S23: The improved empirical mode decomposition method is used to separate the modal components in the leakage signal data to obtain all the mode function (IMF) components; the KL divergence method is introduced to screen all the mode function (IMF) components to obtain stable mode function (IMF) components, obtain a clearer reconstructed signal, and form characteristic parameters.

[0144] The obtained r(t) is decomposed using the improved Empirical Mode Decomposition (MEEMD) method to obtain all mode function (IMF) components. Due to the occurrence of mode aliasing, empty mode function (IMF) components will be displayed in the decomposed signal. A sequence Z = {z1, z2, ..., z...} is formed for all mode function (IMF) components. c The sequence Y = {y1, y2, ..., y} and the leakage signal components are composed of the following: c The effectiveness of the IMF is determined by calculating the differences between the sequences using the KL divergence method, as shown in the following formula:

[0145]

[0146] In the formula: δ(p, q) and δ(q, p) represent the KL distance between sequences Z and Y, respectively; p(z) represents the kernel density function of sequence Z; q(y) represents the probability distribution of sequence Y; z and y represent the elements in sequences Z and Y, respectively;

[0147] The divergence values ​​of sequences Z and Y are obtained as shown in the following formula:

[0148] D(p, q)=δ(p, q)+δ(q, p);

[0149] In the formula: D(p, q) represents the divergence values ​​of sequences Z and Y;

[0150] After the leakage signal components are decomposed using the improved Empirical Mode Decomposition (MEEMD) method, the KL divergence values ​​of all obtained Mode Function (IMF) components are calculated. This yields a set of similarity sequences for all IMF components. All similar sequences are then filtered based on their correlation coefficient r and the inverse divergence value d. The selected sequences with the highest correlation coefficient r are then... max and d max The similarity sequences are combined to form the optimal solution sequence, as shown in the following equation:

[0151] λ=(r mmax d max );

[0152] In the formula: λ represents the optimal solution sequence; r max d represents the maximum correlation coefficient; max This represents the reciprocal of the maximum divergence value;

[0153] The optimal solution sequence is then substituted into the Euclidean distance formula, and the calculation method is shown in the following equation:

[0154]

[0155] In the formula: γ i The Euclidean distance from the similarity sequence of modal function (IMF) components to the optimal sequence; r i The correlation coefficient represents the modal function (IMF) components; d i It represents the reciprocal of the divergence value of the modal function (IMF) components;

[0156] Table 1 Calculation results of Modal Function (IMF) components

[0157]

[0158] Based on the calculation results of the mode function (IMF) components in Table 1, stable mode function (IMF) components are selected to obtain a clearer reconstructed signal.

[0159] S3: Based on the kernel principal component method and adaptive neural network model, establish a life prediction model for electromagnetic directional valves;

[0160] S31: Use the kernel principal component method to fuse the characteristic parameters of the reconstructed signal and establish a mapping model of the electromagnetic commutator degradation fusion index on the voltage drop;

[0161] To preserve the detailed features of the electromagnetic directional valve, the kernel principal component analysis (KPCA) method was used to fuse the extracted feature parameters and construct a degradation fusion index. The resulting feature parameter samples were then compiled into a dataset, as shown below:

[0162] X = [x1, x2, ..., x m ], xm ∈R, m=1,2,……,M;

[0163] In the formula: X represents the dataset composed of feature parameter samples; x m This represents the m-th sample dataset; M represents the number of samples.

[0164] Define a nonlinear mapping relationship Φ(f):R K →R D If D >> K, then each independent vector in the dataset is mapped to a feature space F, resulting in a feature space scaling matrix, as shown in the following equation:

[0165] Φ(f)=[Φ(x1),Φ(x2),……,Φ(x P )];

[0166] In the formula: Φ(f) represents the characteristic space scale matrix; Φ(x) represents the characteristic space scale matrix. P () represents the p-th sample point in the kernel symmetric matrix; P represents the number of variables in an independent vector;

[0167] Calculate the feature space scale matrix based on the feature parameter dataset, construct the kernel symmetric matrix K, and define the kernel symmetric matrix K = {k}. ij} P×P Introducing the feature space F, we perform an inner product transformation of the vectors to reconstruct the kernel symmetric matrix K, as shown in the following equation:

[0168] K=(Φ(x i ), Φ(x j ));

[0169] In the formula: K represents the kernel symmetric matrix; Φ(x i ) represents the i-th sample point in the kernel symmetric matrix; Φ(x) j () represents the j-th sample point in the kernel symmetric matrix;

[0170] The k-th (k = 1, 2, ..., P) principal component T is calculated in the feature space F. k As shown in the following formula:

[0171]

[0172] In the formula: T k v represents the k-th principal component in the feature space F; k α represents the eigenvectors of the kernel symmetric matrix K; Φ(f) represents the eigenspace scale matrix; α ik This represents the constant coefficient of the k-th pivot element;

[0173] A mapping model of the degradation fusion index of the electromagnetic directional valve to the pressure drop is established by using the triple exponential smoothing method, which processes the degradation estimation index of the electromagnetic directional valve and reduces the impact of interference signals on data training.

[0174] For any sequence at time [X1, X2, ..., X...] v First, an exponential smoothing process is applied to obtain a first-order exponential sequence. Then set All sequences are observed and known sequences. Exponential smoothing is applied consecutively for prediction, and the result is... Repeat the exponential smoothing process three times to obtain the updated time series. in Given a time series [X1, X2, ..., X...] v The smoothing result of ] The time-shift exponential smoothing prediction results are obtained after three processing steps;

[0175]

[0176] In the formula: This represents the result of a single exponential smoothing. This indicates the result of quadratic exponential smoothing; This represents the result of quadratic exponential smoothing; α represents the smoothing coefficient. X represents the smoothed value of the first v-1 data points in a single smoothing process; v This represents the actual observed value at time v; This represents the smoothed value of the first v-1 data points in the secondary smoothing process; This represents the smoothed value of the first v-1 data points in the three smoothing processes;

[0177] Different feature parameters contribute differently to the description of the degradation process. Introducing an overall contribution rate, a higher value indicates that the principal component contains more spatial information. Arranging the contribution rates according to their magnitude and introducing smoothing treatment for the principal components, the performance degradation index can represent the growth trend and, within a certain range, describe the performance degradation process of the solenoid directional valve. Constructing a mapping model between the solenoid directional valve degradation fusion index and the solenoid directional valve pressure drop can facilitate accurate prediction of the degradation fusion index.

[0178] S32: The data from the optimal degradation point of pressure drop to the initial prediction position is used to form a training set, which is then input into an adaptive neural network for training to establish a life prediction model for the electromagnetic directional valve.

[0179] The Adaptive Neural Network (ANFIS) prediction model consists of 5 layers forming a complete feedforward network. It calculates the relative fuzzy set membership function of each input component. The input layer directly connects each node to the input component, and the connection is used as a new input to be transmitted to the membership calculation layer to build a life prediction model for electromagnetic directional valves.

[0180] The method for obtaining the membership function of a fuzzy set is as follows:

[0181]

[0182] In the formula: Represents the membership function of a fuzzy set; a 1g b represents the subparameter of the first membership function; 1g c represents the exponential parameter of the first membership function; 1g Indicates the parameter set above the first membership function; a 2g b represents the subparameter of the second membership function; 2g c represents the exponential parameter of the second membership function. 2g This indicates the parameter set on the second membership function; Represents the first membership function; A represents the second membership function; g B represents the set of results from the first fuzzy processing. g o1 represents the second fuzzy processing result set; o2 represents the first input variable; g represents the second input variable; g represents the number of fuzzy sets.

[0183] Match the antecedents of the fuzzy rule and calculate the applicability of the fuzzy rule, as shown in the following formula:

[0184]

[0185] In the formula: Indicates the applicability of fuzzy rules; Indicates the membership degree of a fuzzy set;

[0186] After normalizing the overall network, the applicability of the fuzzy rule normalization is calculated as follows:

[0187] ;

[0188] In the formula: ω represents the applicability of fuzzy rule normalization; g Indicates the intensity of the incentive;

[0189] The normalized network is subjected to overall fuzzification, and the applicability of the fuzzy rule standardization is calculated as follows:

[0190]

[0191] In the formula: Indicates the applicability of fuzzy rule standardization; f represents the applicability of fuzzy rule normalization; g p represents the set of consequent parameters; g Indicates the first consequent parameter; q g Indicates the second consequent parameter; r g Indicates the third consequent parameter;

[0192] Finally, the normalized triggering degree of the fuzzy rules for the entire grid is calculated, as shown in the following formula:

[0193]

[0194] In the formula: Indicates the normalization triggering degree of the fuzzy rule;

[0195] Based on the fuzzy set membership function, fuzzy rule applicability, fuzzy rule normalization applicability, fuzzy rule standardization applicability, and fuzzy rule normalization triggering degree, an adaptive neural network-based life prediction model for electromagnetic directional valves is established.

[0196] S4: Set the failure pressure drop threshold to predict the lifespan of the solenoid directional valve;

[0197] The life prediction model of the solenoid directional valve established by S3 is obtained, and the life of the solenoid directional valve is predicted by the failure pressure drop threshold.

[0198] To compare the prediction results, the results of the triple exponential smoothing prediction are interpolated into the overall prediction results for comparison. For example... Figure 3 The figure shows the results of linear fitting algorithm, cubic exponential algorithm, and adaptive fuzzy neural network (ANFIS) calculation methods according to embodiments of the present invention. The figure shows that the prediction result of the cubic exponential algorithm is 873,000, the prediction result after processing by the linear fitting algorithm is 854,000, and the prediction result after processing by the ANFIS algorithm is 868,000. The prediction results of the linear fitting algorithm differ significantly from those of the cubic exponential and ANFIS algorithms. However, the prediction curve of the cubic exponential algorithm differs significantly from the actual pressure drop curve, so the prediction results of the cubic exponential algorithm may be subject to sudden changes and should not be used for long-term predictions. Compared with the linear model and the cubic exponential algorithm, the ANFIS method provides more accurate predictions, and the predicted trend is more consistent with the actual trend. The adaptive fuzzy neural network (ANFIS) exhibits good prediction performance.

[0199] In summary, the prediction results of the electromagnetic directional valve life prediction method based on flow signals in this case demonstrate its excellent effectiveness.

[0200] (1) In this embodiment of the invention, the inlet and outlet pressure drop of the electromagnetic directional valve is used as the failure criterion of the electromagnetic directional valve. Based on the relationship between flow rate, leakage and pressure drop, the leakage signal data is decomposed into multiple scales using an improved empirical mode decomposition method. The noise reduction and reconstruction are performed using mode function component screening rules. This can improve the adaptability of the mode decomposition process, reduce the probability of mode aliasing, increase reconstruction accuracy and improve computational efficiency. It is more suitable for signal processing of electromagnetic directional valves. The comparative calculation results show that this method has better computational accuracy.

[0201] (2) The embodiments of the present invention apply the kernel principal component method for fusion processing, which can reduce the dimensionality of data while ensuring effective representation, fuse detailed features with high contribution rate, construct a fusion index for the performance degradation of electromagnetic directional valve, and reduce the impact of random interference on data training by three exponential smoothing processes, thereby improving the accuracy of the index; linear prediction of the degradation evaluation index of electromagnetic directional valve is performed using three smoothing prediction, the optimal degradation fusion index is selected, the degradation evaluation model of electromagnetic directional valve is completed, adaptive fuzzy neural network learning and training are performed, the pressure drop trend of electromagnetic directional valve is effectively predicted, and the life prediction result of electromagnetic directional valve is obtained by using the failure pressure drop criterion; the superiority of this method can also be seen by comparing the prediction results of different methods.

[0202] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the lifespan of an electromagnetic directional valve based on flow signals, characterized in that, The method includes the following steps: Step S1: Obtain the initial state data of the solenoid directional valve; The flow and pressure signals at the inlet and outlet of the electromagnetic directional valve are collected using a flow meter and a pressure sensor; the leakage signal is obtained by subtracting the inlet and outlet flow signal data; and the pressure drop signal is obtained using wavelet packet decomposition. The initial state data includes both leakage and pressure drop signal data. Step S2: Based on the improved empirical mode decomposition method, obtain the reconstructed signal data; including: Step S21: Introduce white noise into the leakage signal data, process the leakage signal data from step S1 using an improved empirical mode decomposition method, and obtain the average value of the leakage signal components. Step S22: Perform mode decomposition on the average value of the leakage signal components to filter out the leakage signal components whose permutation entropy exceeds a fixed threshold; repeat the filtering process until the steady-state leakage signal components are obtained. Step S23: Use the improved empirical mode decomposition method to separate the modal components in the leakage signal data to obtain all modal function components; introduce the KL divergence method to screen all modal function components to obtain stable modal function components, obtain a clearer reconstructed signal, and form characteristic parameters; Step S3: Based on the kernel principal component method and adaptive neural network model, establish a life prediction model for the electromagnetic directional valve; including: Step S31: Use the kernel principal component method to fuse the characteristic parameters of the reconstructed signal and establish a mapping model of the electromagnetic commutator degradation fusion index on the voltage drop; The extracted feature parameters are fused using the kernel principal component method. Based on the principle of kernel function, the input space is projected to a high-dimensional feature space through nonlinear mapping. Then, principal component analysis is performed in the high-dimensional feature space to obtain the feature vector. A mapping model of the degradation fusion index of the electromagnetic directional valve to the pressure drop was established using a cubic exponential smoothing method. Step S32: The data from the optimal degradation point of pressure drop to the initial prediction position is used to form a training set, which is then input into an adaptive neural network for training to establish a life prediction model for the electromagnetic directional valve. Step S4: Set the failure pressure drop threshold to predict the lifespan of the solenoid directional valve; Obtain the life prediction model of the electromagnetic directional valve established in step S3, and use the failure pressure drop threshold to predict the life of the electromagnetic directional valve.

2. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 1, characterized in that, The specific steps in step S1 of obtaining the voltage drop signal data using wavelet packet decomposition include: The pressure signal is decomposed into N layers of wavelet packets, and the energy in the flow signal of the electromagnetic directional valve is evenly distributed. On these orthogonal frequency bands, Each orthogonal frequency band contains energy reflecting state information. The method for obtaining voltage drop signal data is as follows: ； In the formula: Indicates the first s Individual voltage drop signal data; Indicates the first Wavelet packet coefficients; e Indicates the dimension number of a high-dimensional space; s,n Represents the decomposition node; M e The dimension of a higher-dimensional space; l Indicates the number of decomposition layers.

3. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 1, characterized in that, In step S21, the method for obtaining the average value of the leakage signal components is as follows: ； In the formula: This represents the average value of the leakage signal components; N Indicates the wavelet packet decomposition level; Ne Indicates the total number of signals processed; n Indicates the number of signals processed: This represents the processed signal after introducing positive white noise; This represents the processed signal after introducing negative white noise. t Represents a time variable.

4. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 3, characterized in that, In step S22, the average value of the leakage signal components is decomposed into modes to filter out the leakage signal components whose permutation entropy exceeds a fixed threshold. Repeat the filtering process until the steady-state leakage signal component is obtained, specifically including: The original signal corresponding to the previous The modal components are decomposed as shown in the following equation: ； In the formula: Indicates the remaining signal; Indicates the overall signal; This indicates the selected stable signals; p Indicates the total number of original signals; Combining leakage signal components into a sequence Reconstruct its phase space: ； In the formula: Indicates the embedding dimension; Indicates a time delay; Y This represents a sequence composed of leakage signal components; y Representing sequences respectively Y Elements in; Based on any vector A sequence of symbols can be obtained, as shown below: ； In the formula: This represents the symbol sequence corresponding to the vector; 'a' is represented as... c The phase space maps different symbol sequences; A single code element corresponding to a vector; E Indicates the total number of symbol sequences; The method for obtaining the permutation entropy of the leakage signal component sequence is as follows: ； In the formula: Represents the entropy of a non-standardized permutation; right The standardization process is performed as follows: ； In the formula: Represents the permutation entropy; The final permutation entropy The larger the value, the more disordered the time series; permutation entropy. The smaller the value, the more regular the time series. The leakage signal components whose permutation entropy exceeds a fixed threshold are filtered out until the steady-state leakage signal components are obtained.

5. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 4, characterized in that, In step S23, an improved empirical mode decomposition method is used to separate the modal components in the leakage signal data, obtaining all modal function components. The KL divergence method is then introduced to filter all modal function components, resulting in stable modal function components and clearer reconstructed signal data. The constituent feature parameters specifically include: The obtained data were decomposed using an improved empirical mode decomposition method. This yields all modal function components. Due to mode aliasing, the decomposed signal will show empty modal function components. A sequence is then formed from all the modal function components. A sequence composed of leakage signal components The effectiveness of the IMF is determined by calculating the differences in the sequences using the KL divergence method, as shown in the following formula: ； In the formula: Let Z and Y represent the KL distances, respectively. Represents the Z-kernel density function of the sequence; Let z represent the probability distribution of sequence Y; z and y represent the elements in sequences Z and Y, respectively; C represents the sequence length. The divergence values ​​of sequences Z and Y are obtained as shown in the following formula: ； In the formula: Represents the divergence values ​​of sequences Z and Y; After the leakage signal components are decomposed using an improved empirical mode decomposition method, the KL divergence values ​​of all obtained modal function components are calculated. This yields a set of similarity sequences for all modal function components. All similarity sequences are then ranked according to their correlation coefficients. r and the inverse of divergence d Filter the selected items. and The similarity sequences are combined to form the optimal solution sequence, as shown in the following equation: ； In the formula: Represents the sequence of optimal solutions; This represents the maximum correlation coefficient; This represents the reciprocal of the maximum divergence value; Substituting the optimal solution sequence into the Euclidean distance formula, as shown in the following equation: ； In the formula: This represents the Euclidean distance from the similarity sequence of modal function components to the optimal sequence; The correlation coefficient represents the modal function components; It represents the reciprocal of the divergence value of the modal function components; Based on the results obtained above, stable mode function components are selected to obtain a clearer reconstructed signal.

6. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in any one of claims 3-5, characterized in that, In step S31, the obtained feature parameter samples are assembled into a dataset, as shown below: ； In the formula: X The dataset represents the sample of features; Indicates the first m One sample dataset; M Indicates the number of samples; Define a nonlinear mapping relationship This maps each independent vector in the dataset to a feature space. This yields a feature space scale matrix, as shown in the following equation: ； In the formula: Represents the characteristic space scale matrix; Describes the nth kernel symmetric matrix p 100 sample points; P This represents the number of variables in an independent vector; Calculate the feature space scale matrix based on the feature parameter dataset and construct a kernel symmetric matrix. , kernel symmetric matrix Introducing feature space This involves performing an inner product transformation on the vectors to reconstruct a kernel-symmetric matrix. K As shown in the following formula: ； In the formula: Represents a kernel-symmetric matrix; Describes the nth kernel symmetric matrix i 100 sample points; Describes the nth kernel symmetric matrix j 100 sample points; In feature space F The calculation yields the first k individual principal As shown in the following formula: ； In the formula: Representing the feature space F The Middle k One principal element; Represents a kernel-symmetric matrix eigenvectors; Represents the characteristic space scale matrix; Indicates the first The constant coefficients of each principal component.

7. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 6, characterized in that, Step S31 employs a cubic exponential smoothing method to establish a mapping model between the electromagnetic directional valve degradation fusion index and pressure drop, specifically including: For any sequence at any time First, an exponential smoothing process is applied to obtain a first-order exponential sequence. Then set All sequences are observed and known sequences. Exponential smoothing is applied once consecutively for prediction, and the result is... Repeat the exponential smoothing process three times to obtain the updated time series. ,in Given time series The smoothing result The time-shift exponential smoothing prediction results are obtained after three processing steps; ； In the formula: The result of a single exponential smoothing; The result of quadratic exponential smoothing; Results of triple exponential smoothing; ; Indicates the first step in a smoothing process v -1 smoothed value of the data; This represents the actual observed value at time v; Indicating the first step in the secondary smoothing process v -1 smoothed value of the data; Indicates the first step in the three-stage smoothing process v -1 smoothed value of the data; Different feature parameters contribute differently to the description of the degradation process. An overall contribution rate is introduced, and the higher the value, the more spatial information the principal component contains. The contribution rates are arranged according to their magnitude, and a smoothing process is introduced for the principal components. The performance degradation index can represent the growth trend and describe the performance degradation process of the electromagnetic directional valve within a certain range. A mapping model of the electromagnetic directional valve degradation fusion index to the electromagnetic directional valve pressure drop is constructed to facilitate accurate prediction of the degradation fusion index.

8. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 7, characterized in that, Step S32 specifically includes: calculating the relative fuzzy set membership function of each input component; the input layer directly connects each node to the input component; the connected components are transmitted to the membership calculation layer as new inputs; and a life prediction model for the electromagnetic directional valve is constructed.

9. The method for predicting the lifespan of an electromagnetic directional valve based on flow signals as described in claim 8, characterized in that, Step S32 specifically includes: First, the method for obtaining the membership function of a fuzzy set is as follows: ； In the formula: Represents the membership function of a fuzzy set; This indicates the subparameter of the first membership function; Indicates the exponential parameter of the first membership function; This indicates the parameter set on the first membership function; This indicates the subparameter of the second membership function; Indicates the exponential parameter of the second membership function; This indicates the parameter set on the second membership function; Represents the first membership function; This represents the second membership function; This represents the set of results from the first fuzzy processing. This represents the set of results from the second fuzzy processing. Indicates the first input variable; Indicates the second input variable; g Indicates the number of fuzzy sets; Match the antecedents of the fuzzy rule and calculate the applicability of the fuzzy rule, as shown in the following formula: ； In the formula: Indicates the applicability of fuzzy rules; Indicates the membership degree of a fuzzy set; Indicates the intensity of the incentive; After normalizing the overall network, the applicability of the fuzzy rule normalization is calculated as follows: ； In the formula: Indicates the applicability of fuzzy rule normalization; The normalized network is subjected to overall fuzzification, and the applicability of the fuzzy rule standardization is calculated as follows: ； ； In the formula: Indicates the applicability of fuzzy rule standardization; Indicates the applicability of fuzzy rule normalization; Represents the set of consequent parameters; Indicates the first consequent parameter; Indicates the second consequent parameter; Indicates the third consequent parameter; Finally, the normalized triggering degree of the fuzzy rules for the entire grid is calculated, as shown in the following formula: ； ； In the formula: Indicates the normalization triggering degree of the fuzzy rule; Based on the fuzzy set membership function, fuzzy rule applicability, fuzzy rule normalization applicability, fuzzy rule standardization applicability, and fuzzy rule normalization triggering degree, an adaptive neural network-based life prediction model for electromagnetic directional valves is established.

Images