A method for estimating the remaining useful life of equipment based on different degradation trends

Abstract

The present invention provides a method for estimating the remaining service life of equipment based on different degradation trends, comprising the following steps: Step S1: filtering the collected signal, and then adjusting the signal envelope to obtain a normalized health degree; Step S2: establishing an equipment degradation model for common degradation forms of the equipment; Step S3: setting a failure threshold and predicting the remaining service life. The method for estimating the remaining service life of equipment based on different degradation trends provided by the present invention realizes real-time online monitoring, intelligent early warning, intelligent diagnosis and life prediction of systems and equipment, improves equipment reliability analysis and management capabilities, and thus assists enterprise managers in making maintenance decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment reliability management, and in particular to a method for estimating the remaining useful life of equipment based on different degradation trends. Background Art

[0002] In the industrial field, online monitoring and life estimation of equipment have become an important means to improve equipment reliability. By collecting important parameters and real-time data closely related to the health status of the equipment, and processing and analyzing the data to obtain information on the health status of the equipment, it helps companies to specify more scientific and reasonable maintenance strategies for the equipment.

[0003] Since the equipment will produce losses during use, the losses will accumulate with the running time, operating conditions, etc., which will cause the service life of the equipment to gradually decrease. At the same time, with the continuous improvement of my country's signal processing technology, the information in the equipment signal is used to a greater extent, so the health information of the equipment can be obtained through signal processing. At present, there are many ways to process vibration signals, so it is necessary to process the data differently according to different application scenarios and data.

[0004] In the past decade, a large number of equipment remaining life prediction works have been widely studied and applied to various machines, such as rotating elements, turbine gearboxes, aircraft engines, etc. The remaining life prediction can be divided into model-based, data-driven and fusion model-based degradation modeling and life prediction methods in principle. Model-driven degradation modeling and life prediction methods mainly rely on the understanding of product degradation mechanisms. This method usually establishes a physical model of product or system degradation, and then reflects the health status of the equipment through the physical model; data-driven prediction methods refer to appropriate processing of product degradation data, combining mathematical methods such as statistics and random processes to model and estimate the degradation process of the product; fusion model-based degradation models and life prediction methods generally refer to combining the life degradation mechanism model with a data-driven model, or a combination of different data-driven models. Since the construction of different physical models is usually more complicated, data-driven methods have a wider range of applications. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects described in the prior art, thereby providing a method for estimating the remaining useful life of equipment based on different degradation trends. The method for estimating the remaining useful life of equipment based on different degradation trends realizes real-time online monitoring, intelligent early warning, intelligent diagnosis and life prediction of systems and equipment, improves equipment reliability analysis and management capabilities, and thus assists enterprise managers in making maintenance decisions, etc.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for estimating the remaining useful life of equipment based on different degradation trends comprises the following steps:

[0008] Step S1: After collecting the signal, filter it and then adjust the signal envelope to obtain the normalized health degree;

[0009] Step S2: Establishing a device degradation model for common degradation forms of the device;

[0010] Step S3: Set the failure threshold and predict the remaining service life.

[0011] Optionally, the step S1 specifically includes:

[0012] Step S11: collecting the original vibration signal x[n] at each moment;

[0013] Step S12: filtering in the sensitive frequency band [l,h] Hz of the signal to enhance degradation information;

[0014] Step S13: demodulating the vibration signal through Hilbert transform envelope to obtain a square envelope signal;

[0015] Step S14: Calculate the root mean square value of the square envelope signal, and obtain the normalized health degree according to the initial root mean square baseline.

[0016] Optional, health range is 0 to 1.

[0017] Optionally, the step S2 specifically includes:

[0018] Step S21: Determine initial parameters using a data fitting method;

[0019] Step S22: Obtain the expression of the parameter posteriori according to the Bayesian algorithm so that the parameters can be updated adaptively.

[0020] Optionally, the device degradation model is an exponential degradation model, and the exponential degradation model is:

[0021]

[0022] Where X(t) represents the acquired equipment health indicator data, t represents time, θ represents the amplitude of the exponential degradation trend, γ represents the time point when the equipment changes from the healthy stage to the degraded stage, α represents the mean value of the equipment health in the healthy stage, β' represents the slope of the exponential degradation trend, I(·) is the indicator function, σ1 and σ2 are the noise standard deviations in the healthy stage and the degraded stage, respectively.

[0023] Optionally, the remaining service life is obtained according to the exponential degradation model:

[0024]

[0025] Where f is the probability density function of the remaining useful life, T is the failure time, and X k represents the acquired equipment health indicator data, L represents the failure threshold, and represents the mean and standard deviation at t=k+1.

[0026] Optionally, the device degradation model is a linear degradation model, and the linear degradation model is

[0027] Optionally, the device degradation model is a polynomial degradation model, and the polynomial degradation model is: X k =b0+A k b+ε(t), where the error term is ε(t)=σ 2 I k ,

[0028] Among them, X k represents the acquired equipment health indicator data, b0 represents the constant term in the polynomial model, and b represents the time term. k represents the coefficient of the time term of the polynomial model,

[0029] The polynomial degradation model is specifically expressed as:

[0030] Optionally, the remaining service life obtained according to the polynomial degradation model is:

[0031]

[0032] Where f is the probability density function of the remaining useful life, T is the failure time, and X k represents the acquired equipment health indicator data, L represents the failure threshold, and represents the mean and standard deviation at t=k+1.

[0033] Optionally, the failure threshold is set to 70%.

[0034] Compared with the prior art, the method for estimating the remaining useful life of equipment based on different degradation trends provided by the present invention has the following beneficial effects:

[0035] According to the acquired vibration signal, the present invention can obtain enhanced degradation information and square envelope signal through filtering and signal envelope adjustment, and obtain normalized health degree according to the initial root mean square value. By preprocessing the obtained health degree, abnormal data under different working conditions such as shutdown and failure can be effectively removed, and data that can characterize the health of the equipment can be obtained for life prediction.

[0036] Based on the degradation data obtained by the above signal processing, the remaining service life of the equipment can be estimated by establishing mathematical models under different degradation trends and bringing them into the data for solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A technical roadmap for health calculation provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of implementation steps provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following is further explained in detail through specific implementation methods.

[0041] The present invention mainly includes two parts, namely, constructing equipment health indicators, establishing equipment degradation models and estimating remaining service life. The equipment health indicators are obtained by analyzing the equipment vibration signals. The vibration signals contain rich health status information of mechanical equipment. The present invention constructs health indicators by quantifying vibration signals to characterize the degradation trend of the health status of the equipment, and serves as the modeling object of the subsequent remaining life prediction algorithm. Establish an equipment degradation model and estimate the remaining service life of the equipment. The present invention is based on the extracted equipment health indicators and uses multiple degradation trends such as exponential degradation, linear degradation, and polynomial degradation to model the degradation process of the equipment. Finally, the remaining service life of the equipment is estimated based on the concept of soft threshold.

[0042] (1) Building equipment health indicators

[0043] The technical path to construct health indicators is as follows: Figure 1 As shown. Assume the sampling frequency F s The vibration signal collected is x=[x1…x N], where N is the signal length. To effectively estimate the health, if there is prior knowledge of the frequency band of the signal fault signal, the signal can be band-pass filtered with a frequency band of [l,h] ([l,h]∈[0,F s ]), and get the filtered signal x l,h .x l,h The corresponding square envelope signal can be calculated by Hilbert transform: SE l,h =|x l,h +jH(x l,h )| 2 =[SE1…SE N ], where H(·) represents the Hilbert transform. Then the square root mean value can be estimated as: In order to achieve the range of health within the interval [0,1], the health needs to be normalized and transformed as follows, and the health HD = RMS0 / RMS is obtained, where RMS0 is the square root mean when the health is 100%, and RMS is the square root mean at the current monitoring moment. During the program operation, the square root mean detected at the beginning can generally be used as RMS0.

[0044] (2) Establish equipment degradation model and estimate remaining service life

[0045] The present invention models common degradation forms of equipment, including exponential degradation, linear degradation and polynomial degradation, and estimates the remaining service life of the equipment based on the definition of a soft threshold.

[0046] 1) Exponential degradation model

[0047] This section proposes an exponential degradation model based on Brownian motion errors. This model is suitable for applications related to continuous error fluctuations in sensor readings. The present invention proposes a Bayesian update process that can automatically update parameters when new observations arrive. Under the assumption of a Brownian error process, the remaining useful life distribution of the monitoring equipment can be determined.

[0048] 1.1) Exponential degradation modeling

[0049] The degradation process of a product can generally be divided into a healthy stage and a degraded stage. In the healthy stage, the health status of the equipment fluctuates within a certain range and does not show any equipment degradation; in the degraded stage, the equipment begins to show performance degradation. Therefore, a unified model can be established using the following formula:

[0050]

[0051] Where X(t) represents the acquired equipment health indicator data, t represents time, θ represents the amplitude of the exponential degradation trend, γ represents the time point when the equipment changes from the healthy stage to the degraded stage, α represents the mean value of the equipment health in the healthy stage, β' represents the slope of the exponential degradation trend, I(·) is the indicator function, σ1 and σ2 are the noise standard deviations in the healthy stage and the degraded stage, respectively.

[0052] Among them, I(·) is the indicator function, which means that when the conditions in the brackets are met, the value of the indicator function is 1. Then for the second stage modeling, X(t) represents the continuous degradation process of health, so it can be assumed to have the following function form:

[0053]

[0054] W(t) is the standard Brownian motion error term, which follows a normal distribution with a mean of 0 and a variance of t. θ, β, and W(t) are independent of each other. Then the expectation of the above equation can be obtained as:

[0055]

[0056] The whole degradation process can be expressed in a simple form using a matrix, so the mathematical model here can be expressed as:

[0057] Y k =A′ k b′+e′ k

[0058] It should be noted that the sampling interval is Δ=t2-t1=t1.

[0059]

[0060]

[0061] Here, the Bayesian method is used to update the parameters. The posterior distribution of the Bayesian parameters is proportional to the product of the prior distribution and the likelihood function. Since the conjugate prior assumption can ensure that the prior distribution and the posterior distribution have the same distribution form, it is assumed that the prior distribution of b′ is with a mean of μ b' , the variance is Σ b' Normal distribution, that is, N(μ b' ,Σ b' ), then Y k obey Then the posterior distribution is:

[0062]

[0063] in, p(y k |b′) is the likelihood function. Then we can getk When the random variable lnX k+l At time point t k+l =t k +t l The mean is The variance is The normal distribution of

[0064]

[0065]

[0066] When I(Φ(Ψ1(t k ))>Φ(3))=1,

[0067]

[0068]

[0069] 1.2) Product life prediction

[0070] After the model parameters are given, the posterior distribution of the parameters can be obtained, and the failure time distribution of the product can be determined, that is, the time distribution when the product health reaches the failure threshold. In the model, L is used to represent the failure threshold. k+l is the observation point, given y k The remaining useful life distribution after the test can be expressed as follows:

[0071]

[0072] Substitute

[0073] Here,

[0074]

[0075] Where f is the probability density function of the remaining useful life, T is the failure time, and X k represents the acquired equipment health indicator data, L represents the failure threshold, and represents the mean and standard deviation at t=k+1.

[0076] According to this method, when a new signal observation value is obtained, the method used here will easily obtain the remaining life span and life span range.

[0077] 2) Linear degradation model

[0078] The above exponential degradation model is converted to a linear degradation model after logarithmization, that is, Therefore, the linear degradation model can be derived by canceling the logarithmic operation in the above exponential degradation model. The same is true for the remaining service life, so it will not be repeated here.

[0079] 3) Polynomial degradation model

[0080] A polynomial degradation model based on Gaussian error terms is proposed. Combined with Bayesian theory, the model parameters are automatically updated when new observation data arrive. At the same time, the remaining useful life distribution of the monitoring equipment is determined based on the concept of failure threshold.

[0081] 3.1) Polynomial degradation modeling

[0082] As mentioned above, the degradation process of a product can generally be divided into a healthy stage and a degradation stage. Here, only the failure part is modeled using polynomial degradation:

[0083] X k =b0+A k b+ε(t), where the error term is ε(t)=σ 2 I k Among them, X k represents the acquired equipment health indicator data, b0 represents the constant term in the polynomial model, and b represents the time term. k Represents the coefficients of the time term of the polynomial model.

[0084] Expanding the matrix in the above model yields:

[0085]

[0086] Here, the Bayesian method is used to update the parameters. The posterior distribution of the Bayesian parameters is proportional to the product of the prior distribution and the likelihood function. Since the conjugate prior assumption can ensure that the prior distribution and the posterior distribution have the same distribution form, it is assumed that the prior distribution of b is with a mean of μ b , the variance is Σ b Normal distribution, that is, N(μ b ,Σ b ), then the posterior distribution of b is:

[0087]

[0088] set up Then the posterior distribution of parameter b follows N(CD,C), then X k The posterior distribution of

[0089] Among them, p(X k |b) is the likelihood function,

[0090]

[0091]

[0092] Then at t k+l When X k+l The mean value of The variance is

[0093] 3.2) Product life prediction

[0094] After the model parameters are given, the posterior distribution of the parameters can be obtained, and the failure time distribution of the product can be determined, that is, the time distribution when the product health reaches the failure threshold. In the model, L is used to represent the failure threshold. k+l is the observation point, given X k The remaining useful life distribution after the test can be expressed as follows:

[0095] Since the failure time should be positive, we have:

[0096]

[0097] Derivative of the above formula l gives

[0098] in,

[0099] Where f is the probability density function of the remaining useful life, T is the failure time, and X k represents the acquired equipment health indicator data, L represents the failure threshold, and represents the mean and standard deviation at t=k+1.

[0100] The implementation steps of the present invention are as follows Figure 1 and Figure 2 As shown, the present invention constructs health indicators that characterize the degradation process of the equipment based on the vibration data of the equipment; the present invention pre-processes the health indicators extracted from the vibration signal, then performs parameter fitting and uses the above-mentioned model method to adaptively update the parameters, and finally realizes life prediction. The details are as follows:

[0101] (1) The present invention first establishes a decay index. First, the collected sensor signal is filtered, the signal after Hilbert transformation is envelope-adjusted, and then the decay index is obtained by quantifying it using the classic and relatively mature RMS index and normalizing it by RMS0, thereby obtaining a health degree in the range of 0-1.

[0102] (2) Data preprocessing. Due to the complex operating conditions of the equipment such as shutdown / switching / power reduction, the degradation index may be very small or almost zero. This type of data has an impact on subsequent calculations, so the degradation signal needs to be preprocessed.

[0103] (3) Determine the initial parameters. The initial parameters are determined by using the exponential fitting method using real data.

[0104] (4) Adaptive updating of parameters. The theoretical basis of the present invention is Bayesian theory. The above method is to obtain the expression of the parameter posterior according to Bayesian theory, so that the parameters can be updated adaptively, that is, the parameters can be updated according to the decay indicator data calculated by the new observation signal.

[0105] (5) Realize life prediction. Since the healthy operation of nuclear power plant equipment is of great significance, the failure threshold is set to 70% here. When the remaining service life is degraded to 70% of the predicted path, it is considered a failure. The time difference between the corresponding moment and the current moment is the remaining service life.

[0106] The present invention is applicable to systems that have installed sensors on key components related to equipment health and can realize signal monitoring. The algorithms for extracting the health index and estimating the remaining service life can be calculated and used separately, or integrated into the existing equipment monitoring system. The remaining service life estimation value and estimation interval finally obtained by the algorithm can provide effective information for engineers to reasonably replace equipment components.

[0107] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for estimating the remaining useful life of equipment based on different degradation trends, characterized in that: The following steps are involved: Step S1: After collecting the signal, filter it and then adjust the signal envelope to obtain the normalized health degree; Step S2: Establishing a device degradation model for common degradation forms of the device, which include exponential degradation, linear degradation, and polynomial degradation; the device degradation model is an exponential degradation model, which is: Where X(t) represents the acquired equipment health indicator data, t represents time, θ represents the amplitude of the exponential degradation trend, γ represents the time point when the equipment changes from the healthy stage to the degraded stage, α represents the mean value of the equipment health in the healthy stage, β' represents the slope of the exponential degradation trend, I(·) is the indicator function, σ1 and σ2 are the standard deviations of the noise in the healthy stage and the degraded stage, respectively, and W(t) represents the standard Brownian motion error term; Step S3: Set the failure threshold and predict the remaining service life.

2. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 1, characterized in that: The step S1 specifically includes: Step S11: collecting the original vibration signal x[n] at each moment; Step S12: filtering in the sensitive frequency band [l,h] Hz of the signal to enhance degradation information; Step S13: demodulating the vibration signal through Hilbert transform envelope to obtain a square envelope signal; Step S14: Calculate the root mean square value of the square envelope signal, and obtain the normalized health degree according to the initial root mean square baseline.

3. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 2, characterized in that: Health ranges from 0 to 1.

4. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 1, characterized in that: The step S2 specifically includes: Step S21: Determine initial parameters using a data fitting method; Step S22: Obtain the expression of the parameter posteriori according to the Bayesian algorithm so that the parameters can be updated adaptively.

5. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 1, characterized in that: According to the exponential degradation model, the remaining service life is: Where f is the probability density function of the remaining useful life, T is the failure time, and X k represents the acquired equipment health indicator data, L represents the failure threshold, and represents the mean and standard deviation at t=k+1.

6. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 1, characterized in that: The equipment degradation model is a linear degradation model. Among them, X(t) represents the acquired equipment health indicator data, θ represents the amplitude of the exponential degradation trend, β represents the degradation slope parameter, σ2 represents the noise standard deviation in the degradation stage, W(t) represents the standard Brownian motion error term, and t represents the time parameter.

7. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 1, characterized in that: The equipment degradation model is a polynomial degradation model, and the polynomial degradation model is: X k =b0+A k b+ε(t), where the error term is ε(t)=σ 2 I k , Among them, X k represents the acquired equipment health indicator data, b0 represents the constant term in the polynomial model, and b represents the time term, A k represents the coefficient of the time term of the polynomial model, The polynomial degradation model is specifically expressed as:

8. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 7, characterized in that: According to the polynomial degradation model, the remaining service life is obtained as: Where f is the probability density function of the remaining useful life, T is the failure time, and X k represents the acquired equipment health indicator data, L represents the failure threshold, and represents the mean and standard deviation at t=k+1.

9. The method for estimating the remaining useful life of equipment based on different degradation trends according to claim 1, characterized in that: The failure threshold was set at 70%.

Images