Residual life prediction method and device of equipment component, computer equipment and storage medium

Abstract

The invention discloses a residual life prediction method and device for an equipment component, computer equipment and a storage medium, and the method comprises the steps: building a state space model according to the degradation data of the equipment component, and the state space model comprises a degradation equation and an observation equation; monitoring the system state of the equipment component based on the state space model to obtain observation data; determining a parameter estimation value of the state space model based on the observation data; particle filtering is carried out on the observation data based on the parameter estimation value, posterior probability distribution corresponding to the system state of the equipment component is determined, and the posterior probability distribution represents the corresponding degradation state of the equipment component at each moment; the residual life of the equipment component is accurately predicted based on the posterior probability distribution and the preset nonlinear prediction model, and the prediction accuracy of the residual life of the equipment component can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of reliability engineering, and particularly to a method, device, computer device and storage medium for predicting the remaining life of equipment components. Background Art

[0002] In fields such as aerospace, power supply systems, chemical equipment, and machinery manufacturing, failures of key components may not only lead to equipment downtime, but also cause hazards such as personnel safety accidents, production losses, and environmental pollution. For strategic missiles, the impact of component damage is often immeasurable. Therefore, predicting the remaining life of key components is crucial for planned maintenance, reducing downtime, improving production efficiency, reducing costs, and even ensuring safety.

[0003] However, in practical applications, the degradation process of the system is usually affected by complex environmental factors, resulting in non-linear and non-Gaussian noise. Non-linearity is mainly caused by the interweaving of various influencing factors such as wear, corrosion, and fatigue of equipment components, while non-Gaussian noise is often caused by observation errors, sensor failures, or external environmental factors. In these cases, traditional remaining life prediction methods, such as prediction based on linear models, are difficult to accurately describe the system state and predict the remaining life. Summary of the Invention

[0004] The present application proposes a method, device, computer device and storage medium for predicting the remaining life of equipment components, which suppresses observation noise to improve the prediction accuracy of the remaining life of equipment components, and further improves the reliability of equipment components.

[0005] In a first aspect, a method for predicting the remaining life of equipment components is provided, including:

[0006] Establish a state space model based on the degradation data of the equipment component, where the state space model includes a degradation equation and an observation equation;

[0007] Monitor the system state of the equipment component based on the state space model to obtain observation data;

[0008] Determine the parameter estimation value of the state space model based on the observation data;

[0009] Perform particle filtering on the observation data based on the parameter estimation value to determine the posterior probability distribution corresponding to the system state of the equipment component, where the posterior probability distribution represents the degradation state corresponding to the equipment component at each moment;

[0010] Predict the remaining life information of the equipment component based on the posterior probability distribution and a preset non-linear prediction model.

[0011] Second aspect, a remaining life prediction device for a device component is provided, including:

[0012] A building module, configured to build a state space model according to degradation data of the device component, where the state space model includes a degradation equation and an observation equation;

[0013] A monitoring module, configured to monitor a system state of the device component based on the state space model to obtain observation data;

[0014] A first determination module, configured to determine a parameter estimation value of the state space model based on the observation data;

[0015] A second determination module, configured to perform particle filtering on the observation data based on the parameter estimation value to determine a posterior probability distribution corresponding to the system state of the device component, where the posterior probability distribution represents a degradation state corresponding to the device component at each moment;

[0016] A prediction module, configured to predict remaining life information of the device component based on the posterior probability distribution and a preset non-linear prediction model.

[0017] Optionally, in some embodiments of the present application, the first determination module includes:

[0018] A first determination sub-module, configured to determine a noise distribution type of the observation data;

[0019] A construction sub-module, configured to construct a likelihood function based on the noise distribution type and the observation data;

[0020] A second determination sub-module, configured to determine the parameter estimation value of the state space model based on the maximum likelihood estimation method and the likelihood function.

[0021] Optionally, in some embodiments of the present application, the noise distribution type includes a normal distribution, a uniform distribution, or an exponential distribution.

[0022] Optionally, in some embodiments of the present application, the first determination module includes:

[0023] A third determination sub-module, configured to determine the noise distribution type based on the observation data;

[0024] A fourth determination sub-module, configured to determine a first moment and a second moment of the observation data based on the noise distribution type;

[0025] A fifth determination sub-module, configured to determine the parameter estimation value of the state space model based on the first moment and the second moment.

[0026] Optionally, in some embodiments of the present application, the second determination module includes:

[0027] A generation sub-module, configured to generate a particle set based on the observation data;

[0028] An initialization sub-module, configured to initialize the particle set and the weight set corresponding to the particle set;

[0029] A sixth determination sub-module, configured to determine a state equation based on the parameter estimation value;

[0030] An update sub-module, configured to update the weights of the particles in the particle set based on the state equation;

[0031] A sampling sub-module, configured to sample new particles according to the updated weights;

[0032] Determine the posterior probability distribution based on each of the new particles.

[0033] Optionally, in some embodiments of the present application, the degradation equation is expressed as:

[0034] X(t) = X(0) + θt + σB(t)

[0035] Wherein, X(t) is a state variable, θ is a drift coefficient, σ is a diffusion coefficient, and B(t) is an increment in a Brownian motion process.

[0036] Optionally, in some embodiments of the present application, the observation equation is expressed as:

[0037] Y(t) = X(t) + ε

[0038] Wherein, Y(t) is an observation variable, X(t) is a state variable, and ε is observation noise.

[0039] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the remaining life prediction method for the above device components are implemented.

[0040] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the remaining life prediction method for the above device components are implemented.

[0041] The present application provides a method, apparatus, computer device and storage medium for predicting the remaining life of a device component. By establishing a state space model based on the degradation data of the device component, the state space model includes a degradation equation and an observation equation; monitoring the system state of the device component based on the state space model to obtain observation data; determining the parameter estimation value of the state space model based on the observation data; performing particle filtering on the observation data based on the parameter estimation value to determine the posterior probability distribution corresponding to the system state of the device component, and the posterior probability distribution represents the degradation state of the device component at each moment; and accurately predicting the remaining life of the device component based on the posterior probability distribution and a preset non-linear prediction model. In the remaining life prediction solution of the device component provided by the present application, the system state of the device component is monitored through the state space model and particle filtering is performed on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time, and then the posterior probability distribution of the device component is predicted based on the non-linear prediction model to determine the remaining service life of the device component. It can be seen that by performing particle filtering on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time to suppress the observation noise, and then predicting the remaining service life of the device component based on the posterior probability distribution of the device component by the non-linear prediction model, the prediction accuracy of the remaining life of the device component can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0043] Figure 1 It is an application environment diagram of the method for predicting the remaining life of a device component provided by an embodiment of the present application;

[0044] Figure 2 It is a flowchart of the method for predicting the remaining life of a device component provided by an embodiment of the present application;

[0045] Figure 3 It is a structural block diagram of the device for predicting the remaining life of a device component provided by an embodiment of the present application;

[0046] Figure 4 It is a structural block diagram of the computer device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present application. However, those skilled in the art will realize that the technical solutions of the present application may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be used. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the present application.

[0049] The block diagrams shown in the accompanying drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0050] The flowcharts shown in the accompanying drawings are only illustrative and do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps may be decomposed, while some operations / steps may be combined or partially combined, so the actual execution order may change according to the actual situation.

[0051] The method for predicting the remaining life of the device components provided by the embodiments of the present invention can be applied in an application environment such as Figure 1 Among them, the computer device 110 communicates with the server 120 through the network 130. The computer device 110 can establish a state space model based on the degradation data of the device components, and the state space model includes a degradation equation and an observation equation; monitor the system state of the device components based on the state space model to obtain observation data;

[0052] Determine the parameter estimation value of the state space model based on the observation data;

[0053] Perform particle filtering on the observation data based on the parameter estimation value to determine the posterior probability distribution corresponding to the system state of the device component, and the posterior probability distribution represents the degradation state corresponding to the device component at each moment;

[0054] Predict the remaining life information of the device component based on the posterior probability distribution and a preset non - linear prediction model, and display it through the computer device 110. In the present invention, the system state of the device component is monitored through a state - space model, and particle filtering is performed on the observed data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time. Then, based on the non - linear prediction model, the posterior probability distribution of the device component is predicted to determine the remaining service life of the device component. It can be seen that by performing particle filtering on the observed data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time to suppress the observation noise, and then predicting the remaining service life of the device component based on the non - linear prediction model for the posterior probability distribution of the device component, the prediction accuracy of the remaining life of the device component can be improved. Among them, the computer device 110 can be, but is not limited to, the tablet computer 110 - 1 and the notebook computer 110 - 2. The present invention will be described in detail through specific embodiments below.

[0055] Please refer to Figure 2 as shown in Figure 2 FIG. is a schematic flowchart of a method for predicting the remaining life of a device component provided by an embodiment of the present invention. This method can be applied to both the terminal and the server. This embodiment takes the application to the server as an example for illustration. The method for predicting the remaining life of the device component includes the following steps:

[0056] S101: Establish a state - space model according to the degradation data of the device component. The state - space model includes a degradation equation and an observation equation.

[0057] Among them, the device component can be a degradation device. A degradation device refers to a device whose performance gradually decreases or fails over time, such as electronic devices, mechanical devices, medical devices, environmental devices, computer devices, automotive parts, etc. It can also be parts, components or assemblies of a degradation device. Degradation data can be data collected during the process of the performance of the device component gradually decreasing over time.

[0058] The expression of the observation equation is as follows:

[0059] Y(t) = X(t)+ε

[0060] In the formula, Y(t) is the observed variable, X(t) is the state variable, and ε is the observation noise.

[0061] Among them, the observation noise can follow a normal distribution, that is, ε~N(0, a 2 ); the observation noise can follow an exponential distribution, that is, it is the observation noise with parameter λ. The observation noise can follow a uniform distribution, which is the observation noise with parameters a and b.

[0062] The implicit degradation process of the device component varying with time t is {X(t), t ≥ 0}. The degradation data X(t) of the device component with linear stochastic degradation at time t is described by a model driven by a standard Brownian motion process. That is, the expression of the degradation equation of the linear Wiener process is as follows:

[0063] X(0) = X(o) + θt + σB(t)

[0064] In the formula, X(t) is the state variable, θ is the drift coefficient of the degradation equation, σ (σ > 0) is the diffusion coefficient, and B(t) is the increment in the Brownian motion process.

[0065] Among them, when {B(t), t ≥ 0} is a standard Brownian motion process, when t > 0, there is σB(t) ~ N(0, σ 2 t), to describe the dynamic characteristics of the implicit degradation process of the device component, X(o) = x 0 is the initial degradation state, without loss of generality.

[0066] In this application, it is assumed that X(0) = x 0 = 0. Since the degradation equation is driven by {B(t), t ≥ 0}, it thus describes the inherent time-varying uncertainty of the stochastic degradation process. However, each individual device among the same type of device components will show different degradation amounts due to different operating conditions. Based on this, in this application, the observation noise ε is independently and identically distributed with the Brownian motion at any time t, and the observation noise ε of the observation equation, the drift coefficient θ of the degradation equation, and the increment B(t) in the Brownian motion process are mutually independent. The degradation equation describes the evolution of the system state of the device component over time, while the observation equation describes how to obtain the observable output, that is, the observation data, from the state variable.

[0067] Furthermore, for the degradation process {X(t), t ≥ 0} of the non-linear degradation device, it can be represented by a Wiener process as:

[0068]

[0069] In the formula, the degradation process {X(t), t ≥ 0} is driven by the standard Brownian motion process {B(t), t ≥ 0}, π(t, θ) represents the drift coefficient, σ B represents the diffusion coefficient, θ is the parameter vector, and π(t, θ) is a non-linear function of time t.

[0070] When π(t, θ) = θ, the degradation process of the device component is a linear Wiener process. X(0) is the initial degradation amount, which can be 0.

[0071] In this application, π(t, θ) = abt (b-1) , then the degradation equation can be transformed into:

[0072] X(t) = X(0) + at b + σ B B(t)

[0073] In the formula, assuming that a is a given parameter, when b = 1, the degradation process of the equipment component is a linear Wiener process.

[0074] S102: Monitor the system state of the equipment component based on the state space model to obtain observation data.

[0075] The state space model is used to describe the dynamic behavior of the equipment component. During the operation of the equipment component, observation data related to the system state is collected through the state space model. For example, sensor readings such as temperature, pressure, vibration, current, etc.

[0076] S103: Determine the parameter estimation value of the state space model based on the observation data.

[0077] Among them, the parameter estimation value is the model parameter of the state space model. Such as the drift coefficient θ, the diffusion coefficient σ, the parameters λ, a, b.

[0078] The parameter estimation value of the state space model can be estimated by using the maximum likelihood estimation method or the moment estimation method combined with the observation data.

[0079] S104: Perform particle filtering on the observation data based on the parameter estimation value to determine the posterior probability distribution corresponding to the system state of the equipment component.

[0080] Among them, the posterior probability distribution represents the degradation state corresponding to the equipment component at each moment. Specifically, the posterior probability distribution represents the degradation state corresponding to the equipment component at each moment after the current moment. The degradation state can represent the performance level of the equipment component (such as speed, temperature, pressure, vibration level, voltage, current, etc.) or the degradation amount. Among them, the degradation amount (i.e., degradation data) refers to the state change data of the equipment performance of the equipment component over time. Such as changes in physical dimensions, reduction of performance parameters, or any other quantifiable performance degradation.

[0081] The posterior probability distribution of the approximate system state can be obtained by performing particle filtering on the observation data through the Monte Carlo method.

[0082] In this step, particle filtering can eliminate the influence brought by the observation noise. Whether it is normal distribution noise, exponential distribution noise or uniform distribution noise, it can well fit the real degradation process, and over time, the filtering effect will gradually become better, which is conducive to improving the prediction accuracy of the remaining life prediction of the equipment component.

[0083] S105: Predict the remaining life information of the device component based on the posterior probability distribution and a preset non - linear prediction model.

[0084] Among them, the non - linear prediction model is used to make predictions for posterior probabilities greater than a certain value, obtaining the probabilities of possible failures of the device component at different time points and the probability that the remaining life of the device component exceeds a preset time point. That is, the remaining life information includes the probabilities of possible failures of the device component at different future time points and the probability that the remaining life exceeds a preset time point. Among them, the non - linear prediction model can be a non - linear regression model, neural network, support vector machine or other machine learning models.

[0085] When the stochastic degradation process {X(t), t≥0} first reaches a preset failure threshold ω, the device component fails. It should be noted that for critical devices (such as inertial navigation systems), even if the degradation returns to the normal state after reaching the first - passage time, replacement or maintenance strategies must be adopted. Therefore, once the degradation of some devices reaches the set threshold, the device components must stop operating to ensure safety. In view of this, using the concept of first - passage time, the life T can be defined as:

[0086] T = inf{t:X(t)≥ω|X(0)<ω}

[0087] According to the definition of the device life, define the remaining life L of the device component at the current time t k as: k For:

[0088] L k = inf{l k :X(t k +l k )≥ω}

[0089] The key to modeling a degraded device (i.e., a state - space model) based on a non - linear Wiener process and predicting the remaining life lies in solving the probability density function of the life T, and then realizing the prediction of the probability distribution of the remaining life L k .

[0090] Among them, the probability density function PDF and the cumulative density function CDF of T f are respectively and

[0091] In order to obtain the analytical expression, assume:

[0092] Before the failure time t, the device component operates normally and does not fail;

[0093] If the device component reaches the failure threshold at the first - passage time, the probability that the degradation process reaches the failure threshold before time t can be ignored.

[0094] Based on this, the probability density function of the first passage time for {X(t)|t≥0} to reach the failure threshold ω is:

[0095]

[0096] where: ω is the failure threshold, and S B (t) is the first passage time varying boundary of the standard BM.

[0097] For the case where the drift coefficient π(t;θ) = abt b-1 , the probability density function of the first passage time can be further expressed as:

[0098]

[0099] where a and b are set parameters.

[0100] Therefore, the degradation state of the device at time t k is x k = X(t k ). Then the PDF of the predicted remaining life at time t k is:

[0101]

[0102] where: ω k = ω - x k , ω is the failure threshold, and x k is the degradation state at time t k .

[0103] Since x k = X(t k ) is observed at time t k , then for any t≥t k , using the Markov property of the standard Brownian motion process {B(t),t≥0}, the degradation process can be expressed as In this case, if t is the time for {X(t),t≥t k} to reach the failure threshold, then the difference t - t k corresponds to the realization of the remaining life at time t k . Using the transformation the stochastic process {X(t),t≥t k} can be further transformed into:

[0104]

[0105] Therefore, t kThe remaining life at a moment is equal to the time when the stochastic process {Z(l k ), l k ≥0} first reaches the failure threshold ω k = ω - x k where:

[0106]

[0107] In the formula: Z(l k ) = X(l k + t k ) - x k , and Z(0) = 0.

[0108] Since {Z(l k ), l k ≥0}, then:

[0109] π(l k ; θ) = ab(l k + t k ) b-1

[0110]

[0111] Therefore, the influence brought by measurement noise needs to be considered in the remaining life prediction, especially different types of noise, such as Gaussian noise (i.e., normal distribution noise), exponential distribution noise, and uniform distribution noise. When the observed data is input, the Akaike Information Criterion (AIC) is first used to judge the closest noise distribution type, so as to determine the observation equation in the state space model.

[0112] In one embodiment, when the observation noise in the observation equation follows a Gaussian distribution, the prediction of the remaining life distribution is as follows:

[0113] Assume that the failure time is tf, and the remaining life at the current time t k is l k , so the failure time satisfies: Y M1 (t f ) = X(t f ) + ε k1 = ω, t k = t f - l k . Solving with the state space model gives:

[0114] ω = θ(t k + l k ) 2 + σB(t k + l k ) + ε k1

[0115] where ω is the failure threshold.

[0116] And the observation noise ε k1 ~ N(μ, β), where μ is the mean of the normal distribution (Gaussian distribution) that the observation noise follows, and the relationship between the remaining life and the observation noise is:

[0117]

[0118] Among them, And B(t k + l k ) - B(t k ) ~ N(0, l k ), it can be obtained that:

[0119] ΔX(t k ) ~ N(μΔX(t k ), σ 2 ΔX(t k ))

[0120] In the formula, μ is the mean of the normal distribution (Gaussian distribution) that the observation noise follows, ΔX(t k ) = θl k 2 + 2θt k l k , σ 2 ΔX(t k ) = σ 2 l k , let Z = ΔX(t k ) + ε k1 , there is:

[0121]

[0122] For the noise :

[0123]

[0124] After sorting out, it can be obtained that:

[0125]

[0126] Substitute the two probability density functions and into the convolution formula, and get:

[0127]

[0128] In the formula, g and z represent substituting the two probability density functions and The variable after fusion by substituting into the convolution formula.

[0129] Finally, according to the posterior probability distribution predicted from the observed data (i.e., the predicted state variable X(t)), the probability distribution of l is solved using numerical methods. k The probability distribution.

[0130] In one embodiment, when the observation noise in the observation equation follows an exponential distribution, the prediction of the remaining life distribution is as follows:

[0131] Based on the common conditions in the Gaussian distribution, the change of X(t) from t k to t f is:

[0132] ΔX(t k ) = X(t k +l k ) - X(t k ) = θ((t k +l k ) 2 -t k 2 ) + σ(B(t k +l k ) - B(t k ))

[0133] And the observation noise ε k2 ~Exp(λ), the probability density function of the observation noise can be obtained as:

[0134]

[0135] Define q = ω - Y M2 (t k ) = ΔX(t k ) + ε k2 , so convolving ΔX(t k ) with ε k2 gives:

[0136]

[0137] Among them, the increment of Brownian motion B(t k +R) - B(t k ) follows a normal distribution N(0, R), that is

[0138] B(t k +R) - B(t k ) ~ N(0,R)

[0139] In the formula, R represents the variance of the normal distribution.

[0140] Finally, based on the posterior probability distribution predicted from the observed data (i.e., the predicted state variable X(t)), the probability distribution of l is solved using numerical methods. k of the probability distribution.

[0141] In one embodiment, when the observation noise in the observation equation follows a uniform distribution, the prediction of the remaining life distribution is as follows:

[0142] Since B(t) is a standard Brownian motion and its distribution is a normal distribution, i.e., B(t) ∼ N(0, t), the distribution of X(t) is also a normal distribution:

[0143] X(t) ∼ N(θt 2 , σ 2 t)

[0144] According to the relationship between the degradation equation and the observation equation, we can obtain:

[0145] Y M3 (t) = X(t) + ε k3

[0146] where ε k3 (t) ∼ U(a, b). Since X(t) and ε k3 (t) are independent, the distribution of Y M3 (t) can be obtained by calculating the convolution. Among them, f X (x) represents the probability density function of X(t), represents the probability density function of ε k3 (t). Therefore, the probability density function of Y M3 (t) can be expressed as a convolution:

[0147]

[0148] Substituting X(t) and ε k3 (t) gives:

[0149]

[0150] After arrangement, we can get:

[0151]

[0152] In the formula, represents holds.

[0153] Finally, based on the posterior probability distribution predicted from the observed data (i.e., the predicted state variable X(t)), the probability distribution of l is solved using numerical methods. k of the probability distribution.

[0154] In this step, by combining the posterior probability distribution with the non-linear prediction model, the complexity and uncertainty of the degradation process of the equipment components can be captured more accurately, which is conducive to improving the accuracy of the remaining life prediction of the equipment components, thereby reducing unexpected failures and downtime.

[0155] In one embodiment, determining the parameter estimation value of the state space model based on the observation data includes:

[0156] Determining the type of noise distribution of the observation data;

[0157] Based on the type of noise distribution and the observation data, constructing a likelihood function;

[0158] Based on the maximum likelihood estimation method and the likelihood function, determining the parameter estimation value of the state space model.

[0159] Wherein, the type of noise distribution includes normal distribution, uniform distribution or exponential distribution.

[0160] Optionally, when the representation form of the observation equation is:

[0161]

[0162] In the formula, ε k1 is the observation noise, and the observation noise follows a normal distribution with a mean of 0 and a variance of β 2 .

[0163] Since B(t) ∼ N(0,t) is a standard Brownian motion, B(t) ∼ N(0,β 2 ), the observation data includes the observation values y 1 , y 2 , y 3 , y 4 ,..., y n corresponding to the moments t 1 , t 2 , t 3 , t 4 ,..., t n , and based on the observation equation, determining the conditional distribution of the observation noise Therefore, the value sequence of the state variable X(t) is x 1 , x 2 , x 3 , x 4 ,..., x n , and it is determined that the type of noise distribution of the observation data is normal distribution. Based on this, the probability density function of the observation value is expressed as:

[0164]

[0165] Meanwhile, to describe the dynamic behavior of the state variable x i based on the distribution of the state variable, that is it can be obtained that:

[0166]

[0167] Thus, the likelihood function is constructed:

[0168] Multiply the probability density functions of f X and and sum over all possible state variables:

[0169]

[0170] After calculation and arrangement, it can be obtained that:

[0171]

[0172] Take the log-likelihood function:

[0173]

[0174] Use the maximum likelihood estimation method to estimate the log-likelihood function and determine the parameters that maximize the log-likelihood function, that is, the parameter estimates:

[0175]

[0176] Optionally, the observation equation is expressed as:

[0177]

[0178] where ε k2 is the observation noise, and this observation noise follows an exponential distribution with parameter ε k2 .

[0179] If all the observed values in the observed data are greater than a preset value (such as 0), and the probability density function of the exponential distribution is:

[0180] f ε (e; λ) = λe -λe , e ≥ 0

[0181] The observation noise ε k2 follows an exponential distribution, that is, ε k2 ~Exp(λ). Based on this, the probability density function of the observed value can be determined as:

[0182]

[0183] Construct the likelihood function of the observed value:

[0184]

[0185] where f X (x i ) is the probability density function of x (refer to the above formula), and i the conditional probability density is proportional to the probability density f (x X ), and the log-likelihood function can be obtained as follows: i ) is proportional to the probability density f(x), and the log-likelihood function can be obtained as follows:

[0186]

[0187] where y represents the observed variable.

[0188] Using the maximum likelihood estimation method to estimate the log-likelihood function and determine the parameters that maximize the log-likelihood function, that is, the parameter estimates:

[0189]

[0190] Optionally, the observation equation is expressed as:

[0191]

[0192] where ε k3 is the observation noise, and this observation noise follows a uniform distribution with parameters a and b.

[0193] The probability density function of the observed values of the uniform distribution is:

[0194]

[0195] Based on the uniform distribution of the observation noise, that is, ε k3 ~U(a, b), the probability density function of the observed values is obtained:

[0196]

[0197] Construct the likelihood function of the observed values:

[0198]

[0199] Based on the observation equation, it can be obtained that Based on this, an integral can be performed on X i to obtain From this, it can be obtained that:

[0200]

[0201] Taking the logarithm and simplifying, it can be obtained that:

[0202]

[0203] Estimate the log-likelihood function using the maximum likelihood estimation method to determine the parameters that maximize the log-likelihood function, i.e., the parameter estimates:

[0204]

[0205] In one embodiment, determining the parameter estimates of the state space model based on the observed data includes:

[0206] Determine the type of noise distribution based on the observed data;

[0207] Based on the type of noise distribution, determine the first moment and the second moment of the observed data;

[0208] Determine the parameter estimates of the state space model based on the first moment and the second moment.

[0209] Among them, the type of noise distribution includes normal distribution, uniform distribution or exponential distribution.

[0210] In one embodiment, when the observation noise in the observation equation follows a normal distribution, for the observed value the first moment is:

[0211]

[0212] Since the observed data is discrete, the sample mean can be used to approximate the observed data to obtain the first moment of the observed value :

[0213]

[0214] The observed value The second moment is:

[0215]

[0216] Since the observation noise is independent of the state value Cov[X(t), ε k1 = 0, so E[X(t)ε k1 = 0, and we can get:

[0217] E[Y M1 (t) 2 = E[X(t) 2 + β 2

[0218] At the same time, use the unbiased estimate of the sample variance of the observed data to approximate the second central moment (second moment):

[0219]

[0220] Using the non - linear least squares method, minimize the sum of squared residuals (RSS):

[0221]

[0222] In one embodiment, when the observation noise in the observation equation follows an exponential distribution, the moment - estimation method can be used to obtain a preliminary estimate of θ. For the observed value The first - order moment is:

[0223]

[0224] Since the observation noise E[ε k2 = 1 / λ, thus:

[0225]

[0226] According to the observed values in the observed data, use the sample mean to estimate:

[0227]

[0228] An estimated value of θ can be approximately obtained.

[0229] Non - linear least squares: Take the estimated value of θ obtained by the estimation method as the initial value of θ, and use the non - linear least squares method to calculate θ, σ, λ. Thus, the optimal parameters can be solved:

[0230]

[0231] In this embodiment, the non - linear least squares method is used to solve the optimization problem. Use the moment - estimation method to obtain the initial parameter value θ, and through iterative solution until the convergence condition is met. Achieve the optimal parameter estimation and further obtain a more accurate

[0232] In one embodiment, when the observation noise in the observation equation follows a uniform distribution, the moment - estimation method can be used to obtain a preliminary estimate of θ. For the observed value The first - order moment is:

[0233] E[Y M3 (t)] = E[X(t)+ε k3 = θt 2 +E[ε k3

[0234] Since the observation noise follows a uniform distribution with parameters a and b, thus E[ε k3 =(a + b) / 2, substituting it in to obtain:

[0235]

[0236] ​Using the observed values in the observed data, estimate using the sample mean:

[0237]

[0238] Nonlinear least squares: Use the estimated value of θ obtained by the estimation method as the initial value of θ, and use the nonlinear least squares method to calculate θ, σ, a, b. Therefore, the following problem can be optimized to solve for the optimal parameters:

[0239]

[0240] Solve the optimization problem using the nonlinear least squares method. Use the method of moment estimation to obtain the initial parameter values, and solve iteratively until the convergence condition is met. Further obtain more accurate

[0241] It should be noted that in the state space model, the observation equation includes multiple distributions, while the degradation equation is general.

[0242] In one embodiment, performing particle filtering on the observed data based on the parameter estimate value to determine the posterior probability distribution corresponding to the system state of the device component includes:

[0243] Generating a particle set based on the observed data;

[0244] Initializing the particle set and the weight set corresponding to the particle set;

[0245] Determining the state equation based on the parameter estimate value;

[0246] Updating the weights of the particles in the particle set based on the state equation;

[0247] Sampling new particles according to the updated weights;

[0248] Determining the posterior probability distribution based on each new particle.

[0249] Determine the number of particles used in the particle filtering, and this number is used to ensure the approximate accuracy of the probability distribution.

[0250] The degradation equation in the state space model is expressed as:

[0251] X(t) = X(0) + at b + σ B B(t)

[0252] The observation equation is expressed as:

[0253] Y(t) = X(t) + ε

[0254] In the state space, the observed data is initialized to obtain an initial particle set {p 1 , p 2 ,... p n} and an initial weight set {w 1 , w 2 ,... w n}, where p i ~N(0, 1), i ranges from 1 to n; w i = 1 / n, and n is the number of particles.

[0255] For t = 1, 2,... T, the initial particle set is predicted, weighted updated, and resampled to obtain the filtering result, where each particle represents one of the possible states of the device component.

[0256] Specifically, for the initial particles with i = 1, 2,..., np;, prediction is performed to obtain:

[0257]

[0258] In the formula, if the observation equation follows a normal distribution, then ε i ~N(0, 1); if the observation equation follows an exponential distribution, then ε i ~exp(α), and if the observation equation follows a mean distribution, then ε i ~U(a, b).

[0259] For the updated particles corresponding to i = 1, 2,..., np;, the weights are re - determined (i.e., the weights are updated):

[0260]

[0261] Normalize all the updated weights:

[0262] If Y t follows an exponential distribution, then represents the density function of the observed value Y given the particle t .

[0263] According to the weight set resample the particle set to obtain the filtering result (posterior distribution probability):

[0264]

[0265] In the actual application process, the process of particle filtering for the observed data is as follows:

[0266] Initialize the particles;

[0267] Perform prediction through the state equation:

[0268]

[0269] Update particle weights:

[0270]

[0271] In the formula, the role of 0.00001 is to avoid a weight of 0.

[0272] Resample particles:

[0273] represents the weight of the i-th particle at time step k. Resampling is based on the weights to draw new particles where j = 1, 2,..., N represents the new particle index, N is the number of particles, and the sampling probability is expressed as:

[0274]

[0275] This formula represents that at time step k, the new particle is equal to the current particle with a probability proportional to the weight of the current particle. Therefore, after resampling, new particles will be obtained. These particles will better reflect the posterior probability of the system state.

[0276] The above is the remaining life prediction process of the device components of this application.

[0277] As described above, the present application provides a method, an apparatus, a computer device, and a storage medium for predicting the remaining life of a device component. By establishing a state space model based on the degradation data of the device component, the state space model includes a degradation equation and an observation equation; monitoring the system state of the device component based on the state space model to obtain observation data; determining a parameter estimation value of the state space model based on the observation data; performing particle filtering on the observation data based on the parameter estimation value to determine a posterior probability distribution corresponding to the system state of the device component, where the posterior probability distribution represents the degradation state of the device component at each moment; and predicting the remaining life of the device component based on the posterior probability distribution and a preset non-linear prediction model. In the remaining life prediction solution of the device component provided by the present application, the system state of the device component is monitored through a state space model, and particle filtering is performed on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time. Then, the posterior probability distribution of the device component is predicted based on the non-linear prediction model to determine the remaining service life of the device component. It can be seen that by performing particle filtering on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time to suppress observation noise, and then predicting the remaining service life of the device component based on the posterior probability distribution of the device component by the non-linear prediction model, the prediction accuracy of the remaining life of the device component can be improved.

[0278] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0279] In one embodiment, a device for predicting the remaining life of a device component is provided. The device for predicting the remaining life of the device component corresponds one-to-one to the method for predicting the remaining life of the device component in the above embodiment. Please refer to Figure 3 As shown, the device for predicting the remaining life of the device component includes:

[0280] A building module 201, configured to establish a state space model according to the degradation data of the device component, where the state space model includes a degradation equation and an observation equation;

[0281] A monitoring module 202, configured to monitor the system state of the device component based on the state space model to obtain observation data;

[0282] A first determination module 203, configured to determine a parameter estimation value of the state space model based on the observation data;

[0283] A second determination module 204, configured to perform particle filtering on the observation data based on the parameter estimation value, and determine a posterior probability distribution corresponding to the system state of the device component, where the posterior probability distribution represents the degradation state corresponding to the device component at each moment;

[0284] A prediction module 205, configured to predict the remaining life information of the device component based on the posterior probability distribution and a preset non - linear prediction model.

[0285] In the remaining life prediction solution of the device component provided in this application, a state - space model is established according to the degradation data of the device component. The state - space model includes a degradation equation and an observation equation; the system state of the device component is monitored based on the state - space model to obtain observation data; a parameter estimation value of the state - space model is determined based on the observation data; particle filtering is performed on the observation data based on the parameter estimation value to determine a posterior probability distribution corresponding to the system state of the device component, where the posterior probability distribution represents the degradation state corresponding to the device component at each moment; the remaining life of the device component is accurately predicted based on the posterior probability distribution and a preset non - linear prediction model. In the remaining life prediction solution of the device component provided in this application, the system state of the device component is monitored through the state - space model and particle filtering is performed on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time, and then the remaining service life of the device component is determined by predicting the posterior probability distribution of the device component based on the non - linear prediction model. It can be seen that by performing particle filtering on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time to suppress observation noise, and then predicting the remaining service life of the device component based on the non - linear prediction model for the posterior probability distribution of the device component, the prediction accuracy of the remaining life of the device component can be improved.

[0286] Optionally, in some embodiments of this application, the first determination module 203 includes:

[0287] A first determination sub - module, configured to determine the noise distribution type of the observation data;

[0288] A construction sub - module, configured to construct a likelihood function based on the noise distribution type and the observation data;

[0289] A second determination sub - module, configured to determine the parameter estimation value of the state - space model based on the maximum likelihood estimation method and the likelihood function.

[0290] Optionally, in some embodiments of this application, the noise distribution type includes normal distribution, uniform distribution or exponential distribution.

[0291] Optionally, in some embodiments of the present application, the first determination module 203 includes:

[0292] A third determination sub-module, configured to determine the type of noise distribution based on the observation data;

[0293] A fourth determination sub-module, configured to determine the first moment and the second moment of the observation data based on the type of noise distribution;

[0294] A fifth determination sub-module, configured to determine the parameter estimation value of the state space model based on the first moment and the second moment.

[0295] Optionally, in some embodiments of the present application, the second determination module 204 includes:

[0296] A generation sub-module, configured to generate a particle set based on the observation data;

[0297] An initialization sub-module, configured to initialize the particle set and the weight set corresponding to the particle set;

[0298] A sixth determination sub-module, configured to determine a state equation based on the parameter estimation value;

[0299] An update sub-module, configured to update the weights of the particles in the particle set based on the state equation;

[0300] A sampling sub-module, configured to sample new particles according to the updated weights;

[0301] Determine the posterior probability distribution based on each new particle.

[0302] Optionally, in some embodiments of the present application, the degradation equation is expressed as:

[0303] X(t) = X(0) + θt + σB(t)

[0304] Wherein, X(t) is a state variable, θ is a drift coefficient, σ is a diffusion coefficient, and B(t) is an increment in a Brownian motion process.

[0305] Optionally, in some embodiments of the present application, the observation equation is expressed as:

[0306] Y(t) = X(t) + ε

[0307] Wherein, Y(t) is an observation variable, X(t) is a state variable, and ε is observation noise.

[0308] In one embodiment, a computer device is provided, and the internal structure diagram of the computer device can be as Figure 4As shown in the figure. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, it realizes the functions or steps of a method for predicting the remaining life of a device component.

[0309] In one embodiment, a computer device is proposed, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0310] Establish a state space model based on the degradation data of the device component. The state space model includes a degradation equation and an observation equation; monitor the system state of the device component based on the state space model to obtain observation data; determine the parameter estimation value of the state space model based on the observation data; perform particle filtering on the observation data based on the parameter estimation value to determine the posterior probability distribution corresponding to the system state of the device component. The posterior probability distribution represents the degradation state of the device component at each moment; predict the remaining life of the device component accurately based on the posterior probability distribution and a preset non-linear prediction model.

[0311] In the remaining life prediction scheme of the device component provided in this application, the system state of the device component is monitored through a state space model, and particle filtering is performed on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time. Then, the posterior probability distribution of the device component is predicted based on a non-linear prediction model to determine the remaining service life of the device component. It can be seen that by performing particle filtering on the observation data based on the parameter estimation value to obtain the posterior probability distribution of the device component in real time to suppress observation noise, and then predicting the remaining service life of the device component based on the posterior probability distribution of the device component using a non-linear prediction model, the prediction accuracy of the remaining life of the device component can be improved.

[0312] In one embodiment, a computer-readable storage medium is proposed. The computer-readable storage medium stores a computer program. When the computer program is executed by the processor, the following steps are implemented:

[0313] A state space model is established based on the degradation data of the device component. The state space model includes a degradation equation and an observation equation. The system state of the device component is monitored based on the state space model to obtain observation data. Parameter estimates of the state space model are determined based on the observation data. Particle filtering is performed on the observation data based on the parameter estimates to determine the posterior probability distribution corresponding to the system state of the device component. The posterior probability distribution represents the degradation state of the device component at each moment. The remaining life of the device component is predicted accurately based on the posterior probability distribution and a preset non-linear prediction model.

[0314] In the remaining life prediction scheme of the device component provided in this application, the system state of the device component is monitored through a state space model, and particle filtering is performed on the observation data based on parameter estimates to obtain the posterior probability distribution of the device component in real time. Then, the posterior probability distribution of the device component is predicted based on a non-linear prediction model to determine the remaining service life of the device component. It can be seen that by performing particle filtering on the observation data based on parameter estimates to obtain the posterior probability distribution of the device component in real time to suppress observation noise, and then predicting the remaining service life of the device component based on the posterior probability distribution of the device component using a non-linear prediction model, the prediction accuracy of the remaining life of the device component can be improved.

[0315] It should be noted that for the functions or steps that can be realized by the above computer-readable storage medium or computer device, reference can be made to the relevant descriptions on the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described in detail here.

[0316] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in this application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0317] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0318] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of equipment components, characterized in that: include: Establishing a state space model according to the degradation data of the equipment components, wherein the state space model includes a degradation equation and an observation equation; Monitoring the system state of the equipment components based on the state space model to obtain observation data; determining parameter estimates of the state-space model based on the observed data; Performing particle filtering on the observed data based on the parameter estimation value to determine a posterior probability distribution corresponding to the system state of the equipment component, wherein the posterior probability distribution represents the degradation state corresponding to the equipment component at each moment; The remaining life information of the equipment component is predicted based on the posterior probability distribution and a preset nonlinear prediction model.

2. The remaining life prediction method according to claim 1, characterized in that: Determining the parameter estimation value of the state space model based on the observation data includes: determining a noise distribution type of the observation data; Constructing a likelihood function based on the noise distribution type and the observed data; Parameter estimates of the state-space model are determined based on a maximum likelihood estimation method and the likelihood function.

3. The remaining life prediction method according to claim 2, characterized in that: The noise distribution type includes normal distribution, uniform distribution or exponential distribution.

4. The remaining life prediction method according to claim 1, characterized in that: Determining the parameter estimation value of the state space model based on the observation data includes: determining the noise distribution type based on the observed data; Determining first-order moments and second-order moments of the observation data based on the noise distribution type; Parameter estimates of the state-space model are determined based on the first-order moment and the second-order moment.

5. The remaining life prediction method according to claim 1, characterized in that: The performing particle filtering on the observed data based on the parameter estimation value to determine the posterior probability distribution corresponding to the system state of the equipment component includes: generating a particle set based on the observation data; Initializing the particle set and the weight set corresponding to the particle set; determining an equation of state based on the parameter estimates; Based on the state equation, the weights of the particles in the particle set are updated; Extract new particles according to the updated weights; The posterior probability distribution is determined based on each new particle.

6. The life prediction method according to claim 1, characterized in that: The degradation equation is expressed as: X(t)=X(0)+θt+σB(t) Where X(t) is the state variable, θ is the drift coefficient, σ is the diffusion coefficient, and B(t) is the increment during the Brownian motion.

7. The life prediction method according to claim 1, characterized in that: The observation equation is expressed as: Y(t)=X(t)+ε Where Y(t) is the observed variable, X(t) is the state variable, and ε is the observation noise.

8. A device for predicting the remaining life of equipment components, characterized in that: include: An establishing module, used for establishing a state space model according to the degradation data of the equipment components, wherein the state space model includes a degradation equation and an observation equation; A monitoring module, used to monitor the system state of the equipment components based on the state space model to obtain observation data; A first determination module, configured to determine parameter estimates of the state-space model based on the observed data; A second determination module is used to perform particle filtering on the observation data based on the parameter estimation value to determine a posterior probability distribution corresponding to the system state of the equipment component, wherein the posterior probability distribution represents the degradation state corresponding to the equipment component at each moment; A prediction module is used to predict the remaining life information of the equipment component based on the posterior probability distribution and a preset nonlinear prediction model.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for predicting the remaining life of a device component according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the remaining life of a device component according to any one of claims 1 to 7 are implemented.

Images