Method, device and medium for predicting remaining useful life of equipment

Abstract

The application relates to a device residual service life prediction method, a device and a medium, and the method comprises the following steps: S1, establishing a device degradation model containing a polynomial drift term; S2, performing parameter estimation on the device degradation model by using ridge regression combined with a maximum likelihood estimation method; S3, performing potential failure mode classification by using a Gaussian mixture model, and calculating a random failure threshold value according to the classification result by using kernel density estimation; S4, updating the Gaussian mixture model parameters by using a particle filtering algorithm based on fuzzy resampling, identifying the potential failure mode according to the Gaussian mixture model parameters, and updating the random failure threshold value; S5, calculating a residual service life probability density function value, and taking the time corresponding to the maximum value of the residual service life probability density function as the estimation result of the residual service life. Compared with the prior art, the application considers the influence of potential failure modes of the device on the failure threshold value, can be applied to the case where multiple potential failure modes exist, has wider applicability, and the prediction result is accurate.

Description

Technical Field

[0001] This invention relates to the field of equipment remaining service life prediction, and in particular to a method, device and medium for predicting equipment remaining service life. Background Technology

[0002] Degradation models and failure thresholds are two key factors affecting the accuracy of remaining useful life predictions. Currently, much research focuses on the construction of prediction models, while studies on failure thresholds are relatively scarce. Although some researchers have noted the randomness of failure thresholds, in related studies, the failure thresholds of equipment are usually assumed to follow a certain type of distribution, such as uniform, exponential, and normal distributions. Among these, the normal distribution is widely used due to its convenience in mathematical processing.

[0003] However, this approach has some limitations. Firstly, using typical distributions to describe random failure thresholds may be unreasonable; for example, the data distribution of equipment failure thresholds in real-world scenarios may not be perfectly normal. Secondly, since potential failure modes influence failure thresholds, the failure thresholds of samples within different potential failure modes can vary significantly. Directly using all failure threshold data to estimate the distribution could lead to highly random results, thus affecting the accuracy of remaining service life predictions. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art by providing a method, device and medium for predicting the remaining service life of equipment. This method takes into account the influence of potential failure modes on the failure threshold, can be applied to situations with multiple potential failure modes, has wider applicability, and provides accurate prediction results.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] According to a first aspect of the present invention, a method for predicting the remaining useful life of equipment is provided, the method comprising:

[0007] Step S1: Based on the Wiener process, establish a device degradation model containing polynomial drift terms;

[0008] Step S2: Estimate the parameters of the equipment degradation model using ridge regression combined with maximum likelihood estimation.

[0009] Step S3: Use a Gaussian mixture model to classify potential failure modes. Based on the classification results, use a kernel density estimation algorithm to calculate the corresponding random failure threshold.

[0010] Step S4: Update the Gaussian mixture model parameters using a particle filter algorithm based on fuzzy resampling, identify potential failure modes based on the Gaussian mixture model parameters, and update the random failure threshold.

[0011] Step S5: Calculate the remaining useful life probability density function value based on the equipment degradation model, and take the time corresponding to the maximum value of the remaining useful life probability density function as the estimated result of the remaining useful life.

[0012] Preferably, the device degradation model containing a polynomial drift term in step S1 has the following mathematical expression:

[0013]

[0014] In the formula: x(t) represents the health status of the equipment at time t; x k This indicates the device at the time t when degradation begins. k The state at time; the second term is the polynomial drift term, The coefficient function of the drift term is considered as the P-order polynomial derivative of the independent variable τ, θ p The variable parameter is σ; the third term is the diffusion term, σ is the diffusion term coefficient, and B(τ) is the standard Brownian motion.

[0015] Preferably, step S2 specifically includes:

[0016] Step S21: Obtain device degradation data from N samples. Indicates that the nth sample is in The state observation at time K n This represents the number of observations in the nth sample;

[0017] Step S22: Estimate the parameters of the polynomial drift term using a ridge regression model, specifically including:

[0018] 1) Construct a ridge regression model for the polynomial drift term of the equipment degradation model, with the mathematical expression as follows:

[0019]

[0020] Where: initial time t n,1 =0;

[0021] 2) Calculate the coefficient estimation results of the ridge regression model in formula (2) under different ridge parameters a, determine the current ridge parameter through the ridge trace diagram, and use the coefficient estimation results corresponding to the current ridge parameter a as the drift term parameter estimation results of the equipment degradation model.

[0022] The loss function for ridge regression is expressed as:

[0023]

[0024] In the formula: a is the ridge parameter;

[0025] Step S23: Estimate the diffusion term parameters using the maximum likelihood estimation algorithm, specifically including:

[0026] Let vector Δx n,k For adjacent observation time points t k and t k+1 The degradation increment between them is expressed as:

[0027]

[0028] In the formula: ζ n,k ~N(0,σ 2 Δt n,k ), Δt n,k =t n,k+1 -t n,k ;

[0029] Δx is obtained from the properties of the nonlinear Wiener process. n ~N(T) n ,σ 2 Ω n ),in

[0030]

[0031] In the formula:

[0032] Sample data X = [Δx1, Δx2, ..., Δx N The log-likelihood function of ] is expressed as:

[0033]

[0034] make σ 2 The expression for the estimation result is:

[0035]

[0036] Substituting the drift term parameter estimation result from step S22 into equation (7) yields the diffusion term parameter estimation value.

[0037] Preferably, step S3 uses a Gaussian mixture model for potential failure mode classification, specifically including:

[0038] Parameter estimation results for N samples There are M distinct potential failure modes, where the parameter estimation result for the m-th potential failure mode belongs to a multivariate Gaussian distribution N(μ). m,∑ m ), m=1,…,M; for a Gaussian mixture model, its log-likelihood function is expressed as:

[0039]

[0040] In the formula: α represents the parameters in the Gaussian mixture model. m express The probability of belonging to the m-th distribution, and f(θ n |μ m ,∑ m ) represents θ n The probability density;

[0041] The parameters in the Gaussian mixture model are calculated iteratively using the expectation-maximization algorithm, specifically including:

[0042] 1) Determine the initial cluster centers and calculate the mean of each cluster. and variance As the parameter values ​​of the initial Gaussian distribution, the initial probability is... M represents the number of potential failure modes;

[0043] 2) Calculate the parameter estimation results The probability of coming from the m-th distribution:

[0044]

[0045] 3) Calculate the parameters of the Gaussian mixture model after iteration:

[0046]

[0047] 4) Repeat steps 2) and 3) until the changes in all Gaussian mixture model parameters are less than the pre-set threshold ε, then use the result of the i-th iteration. As parameter estimation results of Gaussian mixture models

[0048] Preferably, in step S3, the kernel density estimation algorithm is used to calculate the random failure thresholds corresponding to different potential failure modes, specifically including:

[0049] Failure thresholds based on training samples with similar potential failure thresholds Where N λ To represent the number of samples, construct the original kernel density estimation function:

[0050]

[0051] In the formula: h is the bandwidth, Let K(*) be the standard deviation of the sample, and K(*) be the kernel function.

[0052] Based on equation (13), an adaptive bandwidth is introduced:

[0053]

[0054] In the formula: β is the sensitivity factor, which satisfies 0≤β≤1;

[0055] Choosing a Gaussian kernel function and considering the constraint that the failure threshold is greater than 0, the probability density function of the random failure threshold is expressed as:

[0056]

[0057] Preferably, in step S4, the Gaussian mixture model parameters are updated using a particle filter algorithm based on fuzzy resampling, specifically including:

[0058] 1) The parameter estimation results of the training samples As the initial particle set at time t1 And assign the same weight to each particle, i.e., w0 = 1 / N;

[0059] 2) In the k-th iteration, the prediction result of the online sample observation data is expressed as:

[0060]

[0061] In the formula: x k Indicates online samples at t k The observed value at time; This represents the prediction result of the nth particle in the kth iteration, and the update result of the (k-1)th iteration is used as the prediction result; Δt k =t k+1 -t k ;

[0062] 3) Obtain t k+1 Observation data at time x k+1 Then, calculate the weight of each particle:

[0063]

[0064] 4) Resample according to the particle weights, and add random noise to the parameter particles in the three resampled samples. The calculation expression is:

[0065]

[0066] In the formula: This indicates that particles are obtained through resampling; ξ n,k ~N(0,∑ k / N) represents random noise, ∑ k For particle set The covariance matrix is ​​positive semi-definite, and can be obtained by Cholesky decomposition ∑ k / N=WW′, where W is a lower triangular matrix, and the random noise is represented by ξ. k,k =WU, where U is a random vector that follows a standard normal distribution, and each component is an independent random number that follows a standard normal distribution;

[0067] 5) Calculation median This is used as the final parameter update result of the k-th iteration. After completing the iterative update, the weights of all particles are initialized.

[0068] When new observations are generated, update the model parameters by performing steps 2) through 5).

[0069] Preferably, in step S4, updating the random failure threshold based on the Gaussian mixture model parameter update result specifically includes:

[0070] According to the principle of Gaussian mixture model, after obtaining the new model parameters, the probability that it belongs to the m-th potential failure mode is:

[0071]

[0072] In the formula: and These are the parameter results of the Gaussian mixture model obtained through the expectation-maximization algorithm;

[0073] The potential failure mode corresponding to the maximum probability is considered to be the potential failure mode of the field equipment.

[0074] Whenever a new observation is generated, the potential failure modes of the equipment are identified, and the corresponding random failure threshold is updated.

[0075] Preferably, step S5 specifically includes:

[0076] Under a fixed failure threshold, the probability density function of the remaining useful life based on the equipment degradation model in equation (1) is expressed as:

[0077]

[0078] According to the law of total probability, under the random failure threshold condition, the probability density function of the remaining useful life based on the equipment degradation model is expressed analytically as follows:

[0079]

[0080] In the formula,

[0081]

[0082] After obtaining the updated results of the degradation model parameters and random failure threshold, substitute them into equation (21) to calculate the probability density function of the remaining service life of the device at the current time, and take the time corresponding to the maximum value of the probability density function as the estimated result of the remaining service life.

[0083] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement any of the methods described above.

[0084] According to a third aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the methods described herein.

[0085] Compared with the prior art, the present invention has the following beneficial effects:

[0086] 1) This invention classifies and calculates random failure thresholds based on potential failure modes, which can provide more accurate random failure threshold estimation results. Then, by adjusting the random failure threshold of the equipment based on degradation data, more accurate prediction of remaining service life can be achieved, providing a solid theoretical basis for predictive maintenance of equipment, which is conducive to improving the efficiency of maintenance work and reducing costs.

[0087] 2) This invention constructs an equipment degradation model containing a polynomial drift term and uses ridge regression combined with maximum likelihood estimation to estimate the model parameters, resulting in higher accuracy in estimating the equipment degradation model parameters. Attached Figure Description

[0088] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0089] Figure 2 A health index graph for 100 engine samples;

[0090] Figure 3 A graph showing the classification results of potential failure modes for the training samples;

[0091] Figure 4 Plots of random failure threshold probability density functions for different potential failure modes;

[0092] Figure 5 This is a schematic diagram of the random failure threshold update process;

[0093] Figure 6 A comparison chart of remaining useful life prediction results;

[0094] Figure 7 This is a comparison chart of cumulative relative accuracy indicators. Detailed Implementation

[0095] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0096] Example

[0097] The effectiveness of this invention was verified using the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset provided by NASA. The C-MAPSS dataset contains four subsets (FD001, ..., FD004), each containing a different number of training samples, test samples, operating conditions, and failure modes. FD003 represents a single operating condition and contains two failure modes: high-pressure compressor degradation and fan degradation; therefore, this subset was selected as the experimental dataset. Engine operating data was obtained from 21 sensors; monitoring data from sensors 2, 3, 4, 7, 9, 11, 12, 15, 20, and 21 were selected as degradation feature data.

[0098] The specific method for constructing engine health indicators is as follows: Sensor data is normalized; filtering is performed using the "smoothdata" function in MATLAB, with a window width of 15; a weighted fusion method based on evaluation indicators is used to construct the health indicators, calculated as follows:

[0099]

[0100] In the formula, N represents the number of training samples; HI n This represents the health indicator data of the nth sample; N represents s The weights of each sensor; corr(·), mon(·), and rob(·) represent the time correlation, monotonicity, and robustness indices, respectively.

[0101] like Figure 2 The table shows the health indicators of 100 engine samples under two failure modes, with sensor fusion coefficients of 0.48, 0.41, 0.95, 0.89, 0.51, 1.12, 1.25, 0.22, -0.27, and 0.042.

[0102] like Figure 1 As shown, this embodiment provides a method for predicting remaining useful life, which includes the following steps:

[0103] Step S1: Based on the Wiener process, establish a device degradation model containing polynomial drift terms.

[0104] In this embodiment, a fourth-order polynomial drift term is used. Therefore, the degenerate model expression containing the fourth-order polynomial drift term is:

[0105]

[0106] In the formula: x(t) represents the health status of the equipment at time t; x k This indicates the device at the time t when degradation begins. k The state at time; 4θ4τ 3 +3θ3τ 2 +2θ2τ+θ1 is the drift term coefficient function; σ is the diffusion term coefficient; θ p (p = 1, 2, 3, 4) are unit variable parameters; σ is a fixed parameter; t k Let t be the starting time of degradation. k =0.

[0107] Step S2: Use ridge regression combined with maximum likelihood estimation to estimate the parameters of the equipment degradation model.

[0108] The parameter estimation method consists of two steps. Assume that the degradation data of N samples are known. Indicates that the nth sample is in The state observation at time K, where K n This represents the number of observations in the nth sample. Since some samples have a lifetime of 400-500 cycles, feature scaling is performed during parameter estimation to prevent the coefficients of higher-order terms from becoming too small. The calculation method is as follows:

[0109]

[0110] In the formula, min(*) and max(*) represent the minimum and maximum values ​​of the elements in the matrix, respectively.

[0111] Next, the scaled features are used as the variables of the original polynomial for subsequent parameter estimation.

[0112] The first step is to estimate the drift term parameters using ridge regression. Assume the initial time is t. n,1 =0, based on the drift term of the degradation model, construct the following regression model:

[0113]

[0114] The loss function of ridge regression can be expressed as:

[0115]

[0116] In the formula, 'a' is the ridge parameter.

[0117] The coefficient estimation results of the ridge regression model (2) under different ridge parameters were calculated using the "ridge" function in MATLAB. Appropriate ridge parameters were selected using the ridge trace plot, and the corresponding coefficients were used as the drift term parameter estimation results of the degradation model.

[0118] The second step is to estimate the diffusion term parameters using maximum likelihood estimation. (Adjacent observation time points t) k and t k+1 The degradation increment Δx between n,k It can be represented as:

[0119]

[0120] In the formula ζ n,k ~N(0,σ 2 Δt n,k ), Δt n,k =t n,k+1 -t n,k .

[0121] Let vector Based on the properties of the nonlinear Wiener process, Δx n ~N(T) n ,σ 2 Ω n ),in

[0122]

[0123] In the formula, Ω n It is a K n A -1 dimensional diagonal matrix, with elements on the diagonal as...

[0124] Sample data X = [Δx1, Δx2, ..., Δx N The log-likelihood function of ] can be expressed as:

[0125]

[0126] make σ 2 The estimation result can be expressed as:

[0127]

[0128] Substituting the drift term parameter estimation result from the first step into equation (9) yields the following result:

[0129] Step S3: Use a Gaussian mixture model to classify potential failure modes. Based on the classification results, use a kernel density estimation algorithm to calculate the corresponding random failure threshold.

[0130] Random failure threshold estimation consists of two steps:

[0131] The first step is to classify potential failure modes based on the parameter estimation results using a Gaussian mixture model. Three potential failure modes are set. Based on the assumptions of the Gaussian mixture model, the parameter estimation result for the m-th potential failure mode belongs to a multivariate Gaussian distribution N(μ) m ,∑ m (m=1,…,3). For a Gaussian mixture model, its log-likelihood function can be expressed as:

[0132]

[0133] In the formula, α represents the parameters in the Gaussian mixture model. m express The probability of belonging to the m-th distribution, and f(θ n |μ m ,∑ m ) represents θ n The probability density.

[0134] The parameters in the Gaussian mixture model are calculated iteratively using the expectation-maximization algorithm:

[0135] 1) Use the "kmeans" function in MATLAB to determine the initial cluster centers and calculate the mean of each cluster. and variance As the parameter values ​​of the initial Gaussian distribution, the initial probability is...

[0136] 2) Calculate sample θ n The probability of coming from the m-th distribution:

[0137]

[0138] 3) Calculate the parameters of the Gaussian mixture model after iteration:

[0139]

[0140] Repeat steps 2 and 3 until the changes in all Gaussian mixture model parameters are less than a pre-set threshold ε, i.e. The result of the i-th iteration As parameter estimation results of Gaussian mixture models

[0141] The second step involves using kernel density estimation to calculate the random failure thresholds corresponding to different potential failure modes. This assumes that the failure thresholds of the training samples under the same type of potential failure thresholds are known. Where N λ This represents the number of samples. Construct the original kernel density estimation function:

[0142]

[0143] In the formula, For bandwidth, where is the standard deviation of the sample; K(*) is the kernel function.

[0144] Based on equation (13), an adaptive bandwidth is introduced:

[0145]

[0146] In the formula, (0≤β≤1) is the sensitivity factor, which is taken as 0.5.

[0147] Choosing the Gaussian kernel function and considering the constraint that the failure threshold is greater than 0, the probability density function of the random failure threshold can be expressed as:

[0148]

[0149] Step S4: Random Failure Threshold Update

[0150] The random failure threshold update consists of two steps. Assume x1, x2, ..., x... k For field equipment at t1, t2, ..., t k Real-time online degradation data.

[0151] The first step is to update the model parameters using a particle filter algorithm based on fuzzy resampling, based on online degradation data:

[0152] 1) The parameter estimation results of the training samples As the initial particle set at time t1 Each particle is assigned the same weight, i.e., w0 = 1 / N.

[0153] 2) In the k-th iteration, the prediction result of the online sample observation data can be expressed as:

[0154]

[0155] In the formula: x k Indicates online samples at t k The observed value at time; This represents the prediction result of the nth particle in the kth iteration, where the update result of the (k-1)th iteration is used as the prediction result. Δtk =t k+1 -t k ;

[0156] 3) When t is obtained k+1 Observation data at time x k+1 Then, the weight of each particle is calculated using the following expression:

[0157]

[0158] 4) Resample according to the particle weights, and add random noise to the resampled parameter particles. The calculation method is as follows:

[0159]

[0160] In the formula: This indicates that particles are obtained through resampling; ξ n,k ~N(0,∑ k / N) represents random noise, where ∑ k For particle set The covariance matrix. Since the covariance matrix is ​​positive semi-definite, it can be obtained through Cholesky decomposition ∑ k / N = WW′, where W is a lower triangular matrix. Therefore, random noise can be represented as ξ n,k =WU, where U is a random vector that follows a standard normal distribution, and each component is an independent random number that follows a standard normal distribution.

[0161] 5) Calculation median This result is then used as the final parameter update result for the k-th iteration. After completing the iterative update, the weights of all particles are initialized.

[0162] When new observations are generated, the model parameters are updated by executing steps 2) to 5).

[0163] The second step is to update the random failure threshold based on the model parameter update results. According to the principle of the Gaussian mixture model, after obtaining the new model parameters, the probability of it belonging to the m-th potential failure mode is:

[0164]

[0165] In the formula and These are the parameter results of the Gaussian mixture model obtained through the expectation-maximization algorithm.

[0166] The potential failure mode corresponding to the maximum probability is considered the potential failure mode of the field equipment. Whenever a new observation is generated, the potential failure mode of the equipment is identified using the above method, and its corresponding random failure threshold is updated.

[0167] Step S5: Calculate the remaining useful life probability density function value based on the equipment degradation model, and take the time corresponding to the maximum value of the remaining useful life probability density function as the estimated result of the remaining useful life.

[0168] Under a fixed failure threshold, the probability density function of the remaining useful life based on degradation model (2) can be expressed as:

[0169]

[0170] According to the law of total probability, under the random failure threshold condition, the analytical expression of the probability density function of the remaining useful life based on degradation model (2) can be expressed as:

[0171]

[0172] In the formula,

[0173]

[0174] After obtaining the updated results of the degradation model parameters and random failure threshold, substitute them into equation (23) to calculate the probability density function of the remaining service life of the device at the current moment, and take the time corresponding to the maximum value of the probability density function as the estimated result of the remaining service life.

[0175] To verify the superiority of this invention in predicting remaining useful life, 20% of the samples from each of two different failure modes were selected as test samples, and the remaining samples were used as training samples. Three potential failure modes were set, such as... Figure 3 The results shown are the classification results of potential failure modes of the training samples. It can be seen that the present invention can effectively distinguish between samples of high-pressure compressor degradation and fan degradation. Figure 4 The results show the calculation results of the failure threshold and random failure threshold probability density function for samples in different potential failure modes. Figure 5 The paper demonstrates the random failure threshold update process for 20 test samples, showing that as the amount of observed data increases, this invention can accurately identify the types of random failure thresholds for the test samples. The proposed method is then compared with traditional prediction methods based on the Wiener process model. Figure 6The diagram shows the remaining service life prediction results of two methods. It is clear that the method proposed in this invention achieves more accurate prediction results in most samples, especially high-pressure compressor degradation samples. Because traditional methods roughly estimate the random failure threshold based on the failure threshold of all samples, without considering the impact of potential failure modes on the failure threshold or the actual distribution of the failure threshold, there is a significant difference between the predicted and actual remaining service life results. To conduct quantitative analysis, the Cumulative Relative Accuracy (CRA) index is introduced to evaluate the merits of different methods; the smaller the error between the predicted and actual values, the larger the CRA. Figure 7 As shown, the method proposed in this invention performs better in CRA in most samples, and only slightly worse than the traditional method in one sample, further demonstrating its superiority in predicting remaining useful life.

[0176] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0177] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0178] The processing unit executes the various methods and processes described above, such as methods S1 to S5. For example, in some embodiments, methods S1 to S5 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1 to S5 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S5 by any other suitable means (e.g., by means of firmware).

[0179] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload programmable logic devices (CPLDs), and so on.

[0180] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0181] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0182] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the remaining service life of equipment, characterized in that, The method includes: Step S1: Based on the Wiener process, establish a device degradation model containing polynomial drift terms; The equipment degradation model containing polynomial drift terms is mathematically expressed as follows: In the formula: express Monitor the health status of the device at all times; Indicates the moment when the device begins to degrade. The state at time; the second term is the polynomial drift term, The coefficient function of the drift term is considered as the independent variable. of Polynomial derivative of order 1 The first term is a unit variable parameter; the third term is the diffusion term. The coefficient of the diffusion term, This is standard Brownian motion; Step S2: Estimate the parameters of the equipment degradation model using ridge regression combined with maximum likelihood estimation, specifically including: Step S21, Obtain Equipment degradation data for each sample , Indicates the first One sample in State observations at time 10:00 Indicates the first The number of observations in a sample; Step S22: Estimate the parameters of the polynomial drift term using a ridge regression model, specifically including: 1) Construct a ridge regression model for the polynomial drift term of the equipment degradation model, with the mathematical expression as follows: Where: initial time ; 2) The ridge regression model in formula (2) is calculated with different ridge parameters. The coefficient estimation results are used to determine the current ridge parameters through the ridge trace map, and the current ridge parameters are then... The corresponding coefficient estimation results are used as the drift term parameter estimation results of the equipment degradation model; The loss function for ridge regression is expressed as: In the formula: For ridge parameters; Step S23: Estimate the diffusion term parameters using the maximum likelihood estimation algorithm, specifically including: Let vector , For adjacent observation time points and The degradation increment between them is expressed as: In the formula: , ; Based on the properties of the nonlinear Wiener process, it is known that... ,in In the formula: ; Sample data The log-likelihood function is expressed as: make , The expression for the estimation result is: Substituting the drift term parameter estimation result from step S22 into equation (7) yields the diffusion term parameter estimation value. ; Step S3: Use a Gaussian mixture model to classify potential failure modes. Based on the classification results, use a kernel density estimation algorithm to calculate the corresponding random failure threshold. Step S4: Update the Gaussian mixture model parameters using a particle filter algorithm based on fuzzy resampling, identify potential failure modes based on the Gaussian mixture model parameters, and update the random failure threshold. Step S5: Calculate the remaining useful life probability density function value based on the equipment degradation model, and take the time corresponding to the maximum value of the remaining useful life probability density function as the estimated result of the remaining useful life.

2. The method for predicting the remaining service life of equipment according to claim 1, characterized in that, Step S3 uses a Gaussian mixture model to classify potential failure modes, specifically including: for Parameter estimation results for each sample ,exist There are 10 different potential failure modes, of which the 10th failure mode is the 10th failure mode. The parameter estimation results for each potential failure mode belong to a multivariate Gaussian distribution. For a Gaussian mixture model, its log-likelihood function is expressed as: In the formula: Represents the parameters in a Gaussian mixture model; express Belongs to the The probability of each distribution, and ; express The probability density; The parameters in the Gaussian mixture model are calculated iteratively using the expectation-maximization algorithm, specifically including: 1) Determine the initial cluster centers and calculate the mean of each cluster. and variance As the parameter values ​​of the initial Gaussian distribution, the initial probability is... , The number of potential failure modes; 2) Calculate the parameter estimation results From the The probability of each distribution: 3) Calculate the parameters of the Gaussian mixture model after iteration: 4) Repeat steps 2) and 3) until the changes in all Gaussian mixture model parameters are less than the preset threshold. , will the The result of the iteration As parameter estimation results of Gaussian mixture models .

3. The method for predicting the remaining service life of equipment according to claim 2, characterized in that, In step S3, the kernel density estimation algorithm is used to calculate the random failure thresholds corresponding to different potential failure modes, specifically including: Failure thresholds based on training samples with similar potential failure thresholds ,in To represent the number of samples, construct the original kernel density estimation function: In the formula: For bandwidth, The standard deviation of the sample. For kernel functions; Based on equation (13), an adaptive bandwidth is introduced: In the formula: It is a sensitivity factor that satisfies ; Choosing a Gaussian kernel function and considering the constraint that the failure threshold is greater than 0, the probability density function of the random failure threshold is expressed as: 。 4. The method for predicting the remaining service life of equipment according to claim 3, characterized in that, In step S4, the Gaussian mixture model parameters are updated using a particle filter algorithm based on fuzzy resampling, specifically including: 1) The parameter estimation results of the training samples As Initial particle set at time And assign the same weight to each particle, i.e. ; 2) in the In this iteration, the prediction results for the online sample observation data are expressed as follows: In the formula: Indicates online samples The observed value at time; Indicates the first The particle in the first The prediction results in the next iteration will The update result of the next iteration is used as the prediction result; , ; ; 3) Obtain Observational data at time Then, calculate the weight of each particle: 4) Resample according to the particle weights, and add random noise to the parameter particles in the three resampled samples. The calculation expression is: In the formula: This indicates that particles were obtained through resampling; Indicates random noise. For particle set The covariance matrix is ​​positive semidefinite and is obtained through Cholesky decomposition. , It is a lower triangular matrix that represents random noise as , It is a random vector that follows a standard normal distribution, and each component is an independent random number that also follows a standard normal distribution. 5) Calculation median and take it as the first The final parameter update result of the next iteration is used to initialize the weights of all particles after the iteration update is completed. When new observations are generated, update the model parameters by executing steps 2) through 5).

5. The method for predicting the remaining service life of equipment according to claim 4, characterized in that, In step S4, the random failure threshold is updated based on the Gaussian mixture model parameter update results, specifically including: According to the principle of Gaussian mixture models, when new model parameters are obtained, they belong to the... The probability of each potential failure mode is: In the formula: , and These are the parameter results of the Gaussian mixture model obtained through the expectation-maximization algorithm; The potential failure mode corresponding to the maximum probability is considered to be the potential failure mode of the field equipment. Whenever a new observation is generated, the potential failure modes of the equipment are identified, and the corresponding random failure threshold is updated.

6. The method for predicting the remaining service life of equipment according to claim 3, characterized in that, Step S5 specifically includes: Under a fixed failure threshold, the probability density function of the remaining useful life based on the equipment degradation model in equation (1) is expressed as: According to the law of total probability, under the random failure threshold condition, the probability density function of the remaining useful life based on the equipment degradation model is expressed analytically as follows: In the formula, After obtaining the updated results of the degradation model parameters and random failure threshold, substitute them into equation (21) to calculate the probability density function of the remaining service life of the device at the current time, and take the time corresponding to the maximum value of the probability density function as the estimated result of the remaining service life.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

Images