Self-adaptive combined maintenance control method for structural system under environmental heterogeneous damage

Abstract

The invention provides a self-adaptive joint maintenance control method for a structural system under environmental heterogeneous damage. The method comprises the following specific steps: 1, modeling a damage accumulation trajectory of a structural component under an environmental heterogeneous mode; 2, predicting the service life of the part under the environment damage heterogeneous condition; 3, delayed maintenance control optimization of single structural components; 4, quantifying a single structural component-system joint maintenance mapping function; and 5, iterative grouping selection heuristic strategy solution algorithm planning is carried out. The method has the advantages that the dynamic capture of an environmental heterogeneity mode in a structural system damage rule is realized, so that a more accurate residual life prediction activity is carried out, and the sensitivity of a damage feature recognition process is ensured; self-adaptive dynamic updating of system maintenance schemes can be realized according to real-time state evaluation information and historical maintenance records of structural components, correlation in adjacent maintenance schemes can be quantified, the adaptability is high, and the intelligent level is high; the balance between the prolonging of the remaining service life of the structural component and the preparation buffering of the maintenance resources is realized, and the scientificity of the maintenance plan is ensured.

Description

Technical Field

[0001] The present invention relates to an adaptive joint maintenance decision and control method for a structural system that takes into account environmental damage heterogeneity. This method proposes a global adaptive joint intelligent maintenance control method that integrates state parameter update and dynamic selection grouping ideas for damage accumulation type structural systems under non-uniform environmental conditions. Specifically, a structural damage evolution trajectory model containing environmental heterogeneity patterns is constructed to form a monomer structure delayed maintenance strategy based on trajectory parameter self-update, and finally an iterative selection grouping mechanism is used to design a global adaptive maintenance decision and efficient control method for the structural system. This method is applicable to the field of intelligent maintenance of structural systems, and can significantly improve the accuracy and efficiency of system operation and maintenance control in complex and uncertain environments, and reduce the risk of damage and failure losses. Background Art

[0002] As an important carrier of digital technology in the field of system health management, intelligent maintenance based on autonomous decision-making plays an important role in reducing maintenance costs, ensuring personnel safety, and improving service reliability. As large-scale industrial equipment such as aircraft engines, inertial navigation, and high-speed trains gradually develop in the direction of scale and complexity, the cost of leasing, research and development, procurement, and maintenance costs continue to increase. Compared with reducing the costs incurred by the former, reducing maintenance costs is more technically feasible. Therefore, under the condition that catastrophic safety risks and major economic losses are of great concern, improving the scientificity and intelligence of operation and maintenance methods is of great practical significance.

[0003] Component-level maintenance usually only requires the formulation of independent maintenance strategies based on the operating status of a single device. However, the maintenance scheduling of large structural systems is often affected by factors such as resource sharing, structural associations, and fault interactions between structural components, resulting in the irrationality of the traditional maintenance method of simply combining component maintenance according to time proximity. For example, in offshore wind farm maintenance work, a large amount of costs are incurred for personnel transportation, material preparation, system shutdown, and disassembly activities before offshore maintenance activities are carried out; in railway train maintenance work, maintenance personnel need to first wait for the train to arrive, clean the vehicle, and disassemble the structure, which will cause a certain amount of time delay and downtime costs. Such related factors are widely present.

[0004] Traditional group maintenance strategies lack the utilization of failure opportunities, and the mathematical modeling is relatively complex, and the sensitivity to changes in system status is low. Therefore, the "rolling timeline method" was proposed. This method implements the iteration of maintenance plans in the long term by performing group operations within a given range and then moving to subsequent windows to repeat the operations until the planning period expires. However, existing methods face a series of challenges such as not fully considering the correlation between adjacent maintenance activities, limited planning time domain, unreasonable definition of repair operations, and ignoring maintenance opportunities brought by component failures.

[0005] Complex environments such as the ocean or space have strong spatial and temporal heterogeneity. When large structural systems serve in such heterogeneous environments, they usually have typical characteristics such as multi-stage mission profiles and non-uniform time-varying states, which leads to uncertainty in the fault evolution laws of various structural components and strong dynamic nonlinearity in the damage accumulation trajectory. The classic data-driven trajectory recognition method has poor sensitivity to state changes and is difficult to identify such time-varying differences. It is also severely limited by the amount of historical data and the choice of initial parameters, which can easily lead to distortion of state prediction results. This defect is particularly reflected in newly developed systems that lack prior information.

[0006] In order to solve the above problems, the present invention proposes an adaptive joint maintenance control method for structural systems under environmental heterogeneous damage. Based on identifying the impact of environmental heterogeneity on the damage accumulation process and the economic dependence characteristics within the structural system, a global dynamic delayed preventive joint maintenance control strategy is established. Summary of the invention

[0007] (1) The purpose of the present invention is to address the problems of traditional structural system maintenance methods in complex heterogeneous environmental conditions, such as poor responsiveness to time-varying damage data, insufficient mining of remaining life value, and insufficient understanding of the law of maintenance resource demand. A component damage online evolution trajectory correction model covering environmental heterogeneous parameters is comprehensively constructed, a dynamic delayed maintenance strategy is designed to improve the utilization rate of structural system maintenance resources, and an iterative selection grouping mechanism is used to form a global adaptive joint maintenance decision-making method oriented to environmental damage heterogeneity, thereby enhancing the real-time, accuracy and robustness of structural system joint maintenance.

[0008] (2) Technical solution:

[0009] The basic conditions proposed by the present invention are as follows:

[0010] Condition 1: Since the Wiener process has good mathematical analyticity and physical interpretation, it is often used to describe the damage accumulation process of real systems. The damage accumulation process of large structural system components is usually complex. Here, the generalized Wiener process x with a time-space scale conversion function is used. i (t) = x i (0)+v i Λ(t; β i )+σ i B(Λ(t;β i )) Model and analyze the damage accumulation process of components. Among them, v i Λ(t; β i ) is the drift function, where ν i is the key damage parameter, assuming it obeys normal distribution, Λ(t; β i ) is the space-time scale conversion function; β i and σi are deterministic parameters describing the commonality of the damage process trend of the same type of components, representing the time scale conversion parameter and the trend fluctuation parameter respectively; B(·) is the standard Brownian motion; the initial damage value x i (0)=0.

[0011] The advantages of this function are:

[0012] (1) The good mathematical properties of the linear Wiener process in the analytical calculation of lifetime are retained, that is, once the space-time scale conversion function Λ(t; β i ), the explicit distribution of the remaining life can be obtained in analytical form;

[0013] (2) By adjusting the space-time scale conversion function Λ(t; β i ) and the parameter β i The value of can be used to accurately characterize the failure patterns of different components.

[0014] Condition 2: Set the preventive maintenance threshold ξ based on the actual failure mode and damage form of the component i and the corrective maintenance threshold L i ,0<ξ i <L i When the damage accumulation value of component i is monitored to be within the maintenance threshold interval (ξ i ,L i ) (where ξ i is the preventive maintenance threshold, L i is the corrective maintenance threshold), in order to avoid the possible safety and economic impact of failure in advance and fully tap the remaining life value, it is necessary to carry out delayed preventive maintenance control on the component. Suppose the delay time is j i τ (where τ is a fixed monitoring period, j i is an integer variable to be optimized); when its damage accumulation value is monitored to exceed the corrective maintenance threshold L i , corrective maintenance control is initiated immediately.

[0015] The maintenance process of the structural system involves various costs: 1) The cost of monitoring each component C i,I , assuming that the inspection is non-destructive and takes negligible time to perform compared to the maintenance interval. 2) Shared maintenance cost C s,R (including maintenance personnel dispatch, maintenance material dispatch and other support costs) and independent maintenance costs of each component C i,R , only the replacement operation is considered, and the cumulative value of component damage after replacement is updated to 0. 3) Downtime loss per unit time C i,d , Losses due to untimely logistics support C i,fThe suddenness of corrective replacements means that there is insufficient buffer time for resource preparation, which results in additional losses.

[0016] Based on the above ideas, the present invention provides an adaptive joint maintenance control method for structural systems under environmental heterogeneous damage. Figure 1 The method is implemented in the following five steps:

[0017] Step 1: Modeling the damage accumulation trajectory of structural components under environmental heterogeneity mode.

[0018] The dynamic changes of temperature and humidity in the service environment and the presence of tiny particles and dust usually lead to uncertainty in the damage state of the system / component. This uncertainty can be characterized by defining the distribution of parameters in the damage model. Therefore, an adaptive key damage parameter ν is introduced. i , this key parameter is modified in real time under the global Bayesian update system to capture the impact of environmental changes on the system / component failure patterns.

[0019] (1) Offline training of common damage parameters

[0020] For component i, i∈{1,2,…,q} among q components of the same type, the key damage parameter ν i The prior distribution of

[0021] Satisfies the normal distribution, where is the mean of the prior distribution, and p0 is the variance of the prior distribution.

[0022] τ is a fixed monitoring period, in the state monitoring time vector T i =(0,τ,2τ,…,n i The damage accumulation data vector obtained on τ)′ is X i =(0,x i (τ),x i (2τ),…,x i (n i τ))′, where n i is the total number of monitoring times for component i. i The multidimensional normal distribution property of i ) and covariance Cov(X i )for:

[0023]

[0024] Therefore, based on the damage accumulation data of q components of the same type, the offline parameters (covering ν i The mean of the prior distribution of Variance p0, time scale transformation parameter β of the damage accumulation process i , trend fluctuation parameter σ i ) is:

[0025]

[0026] (2) Online correction of heterogeneous damage parameters

[0027] Based on the above common damage parameter offline training module, the key damage parameter ν is obtained i The prior distribution of the parameter ν i The prior distribution in Bayesian statistical inference can be regarded as the parameter ν i The empirical knowledge is obtained through historical data statistics, and the posterior distribution integrates the current sample observation information with the information in the prior distribution and corrects the parameter information in real time. The continuous iteration in the time dimension makes effective use of the sample information and transfers the understanding of the unknown parameters from the prior distribution to the posterior distribution, thereby reducing the uncertainty of the parameter estimate.

[0028] Taking any state monitoring point hτ as an example, the damage accumulation data observed at this moment is x i (hτ), according to the parameter ν i The normal distribution property of , under the global Bayesian update framework, obtains the parameter ν after online correction at time hτ i The latest distribution of is shown in the following formula (4), which is the same as the prior distribution mentioned above. To distinguish, the posterior distribution is used Indicates that is the mean of the posterior distribution, p h is the variance of the posterior distribution:

[0029]

[0030] Step 2: Component life prediction considering heterogeneous conditions of environmental damage.

[0031] The damage process of the components mentioned above is modeled using the generalized Wiener process. According to the statistical independence assumption of the Wiener process, at any state monitoring point hτ, the damage accumulation within Δt time has an increment Δx i (hτ) = x i (hτ+Δt)-x i (hτ) obeys the normal distribution. Therefore, when we get the parameter ν i The corrected posterior distribution at this time Then, according to the conditional probability density and marginal probability density formula of continuous random variables, the probability density function g of the damage increment within the modified Δt time at time hτ can be obtained: i (Δx|Δt; hτ), satisfies the following formula:

[0032]

[0033] Since the present invention involves a variety of probability density functions, in order to avoid confusion, subscripts are used here to distinguish them. v (ν i,h |Δt; hτ) represents the parameter ν at the time hτ i The probability density function of the posterior distribution, f Δ (x|ν i,h ; Δt; hτ) represents the known parameter ν i The posterior estimate ν i,h The probability density function of the damage increment in Δt under the condition of . In addition, Λ(hτ+Δt; β i ),Λ(hτ;β i ) represent the space-time scale conversion function values ​​corresponding to the time instants hτ+Δt and hτ respectively.

[0034] The remaining life of a component is defined as the time interval from the current state point to the functional failure point of a component that has not failed at a certain moment. Correspondingly, the remaining life prediction is the process of estimating the remaining useful life length based on known historical information and current monitoring data. The life of a component is usually random, so the remaining life is defined as a conditional random variable. The life T of the component damage cumulative evolution process based on Wiener process modeling in the present invention is defined by the first time it is reached, that is, the first time the corrective maintenance threshold L is reached i The time T = inf{t:x i (t)≥L i |x i (0)<L i The lifetime of the linear Wiener process satisfies the inverse Gaussian distribution (IG distribution), and its physical background is the damage accumulation process that obeys the standard linear drift Brownian motion. After introducing the parameter uncertainty caused by the space-time scale conversion function and the environmental damage heterogeneity, due to the randomness of the state, the lifetime of the generalized Wiener process described in the present invention no longer obeys the inverse Gaussian distribution and needs to be solved in combination with the total probability formula.

[0035] On this basis, the remaining life of the component after online correction at the condition monitoring point hτ is solved

[0036] T l ={t l =T-hτ:x i (T)≥L i |xi (hτ)<L i ,T>hτ}, its probability density distribution function f i (t l |hτ) and the distribution function F i (t l |hτ) can be expressed by the following formula:

[0037]

[0038]

[0039] Among them, Λ(hτ+t l β i ),Λ(hτ;β i ) represent hτ+t l The value of the space-time scale conversion function corresponding to the time hτ.

[0040] Step 3: Optimization of delayed maintenance control of single structural components.

[0041] The damage accumulation state of a component is continuous. When the damage accumulation of component i is between the maintenance threshold interval (ξ i ,L i ), the execution time of preventive maintenance control is delayed i τ time (where τ is a fixed monitoring period, j i is an integer variable to be optimized) to maximize the value of the remaining life of the component and reserve time for the preparation of maintenance resources. Therefore, the probability density function of the damage increment g is used i (·), assuming that the start time of a preventive maintenance plan is τ sta , the probability distribution of the average monitoring frequency k between adjacent preventive maintenance control activities is It can be expressed as:

[0042]

[0043] Further explanation of the variables and parameters in the formula: u, y, ω are all integral intermediate variables, and their physical meanings represent the damage increment; g i (u|j i τ; τ sta ),g i (ωτ;τ sta ),g i (yj i τ; τ sta ),g i (uτ; τ sta ),g i (ω(kj i -2)τ; τsta ) indicates that in τ sta The probability density function of the damage increment after online correction at the moment; L i -ω,ξ i -ω,L i -ω-u is the upper and lower limits of the integral, and its physical meaning is the damage increment; ξ i ,L i The thresholds for preventive and corrective maintenance.

[0044] To elaborate on the above equation: Since the initial damage level is zero, k≤j i +1 This situation does not exist; k = j i +2 and k ≥ j i +3 Both cases represent that the current damage accumulation value of component i is (ξ i ,L i ) and remains in this state after a period of delay, only in this case is there a probability that delayed preventive maintenance can be carried out normally.

[0045] If component i is monitored to exceed the corrective maintenance threshold L i , at this time, corrective maintenance will be started immediately to avoid further safety and economic losses. At this time, the maintenance planning time τ sta , solve the probability distribution of the average monitoring frequency k between adjacent corrective maintenance control activities It is represented by the following formula. It should be noted that the τ mentioned in the present invention sta It refers to the start time of all maintenance plans (including preventive and corrective maintenance). When it comes to a specific time, just substitute the relevant parameters and values ​​after online correction.

[0046]

[0047] in, represents the distribution function of the damage increment within a monitoring period, In formula (9), m and n are intermediate variables, which are integers and represent the probability summation range in the time increment dimension, which are 1 to j respectively. i and 1~+∞;υ,y,u,ω,ο all represent integral intermediate variables, and their physical meanings represent damage increments; g i (υ|τ;τ sta ),g i (y|(j i -m-1)τ; τ sta ),g i (u|τ; τ sta ),g i (ω|(n-1)τ; τ sta ),gi (ο|(k-2)τ; τ sta ) indicates that in τ sta The probability density function of the damage increment after online correction at the moment; L i -ω,ξ i -ω,L i -ω-u,L i -ω-uy is the upper and lower limits of the integral, and its physical meaning is the damage increment; ξ i ,L i is the maintenance threshold.

[0048] Since the fault state can only be identified through monitoring, this hidden nature will cause the system to shut down, and component i will be disconnected at the maintenance planning start time τ sta The possible average downtime in the future is T d The following equation gives:

[0049]

[0050] Get the long-term maintenance cost rate η of component i i (j i |τ sta )for:

[0051]

[0052] Among them, f i (t l |τ sta ) is the maintenance planning start time τ of component i sta The probability density distribution function of the remaining life after online correction; a is an intermediate variable, which takes an integer; aτ, (a-1)τ represent the upper and lower limits of the integral respectively, and their physical meaning is the time range of component failure; C i,I is the single monitoring cost, C s,R is the shareable maintenance cost, C i,R is the independent maintenance cost of each component, C i,d is the downtime loss per unit time, C i,f Losses due to untimely logistics support; They represent the maintenance planning start time τ sta The probability of the average monitoring frequency k between adjacent preventive maintenance control activities / corrective maintenance control activities.

[0053] Since the maintenance cost can be shared by a fixed amount C S,R It exists in every maintenance activity, so whether C is considered in structural component level maintenance is S,R It has no effect on the optimization results. By solving the minimum value of the above function The optimal delayed maintenance time for component i can be obtained accordingly

[0054] Step 4: Quantification of the monomer structure component-system joint maintenance mapping function.

[0055] The core of system joint maintenance is to share downtime and maintenance resources, but this joint maintenance will cause the maintenance time to deviate from the optimal maintenance execution time of each structural component. This deviation will extend (or shorten) the service life of the component, avoid (or cause) failure downtime, and may also trigger urgent needs for logistics support. A joint cost loss function is defined to quantify this phenomenon. According to the deviation of the joint maintenance time, it can be divided into two cases: (1) early maintenance and (2) deferred maintenance.

[0056] Due to the hiddenness and uncertainty of faults, in the advance maintenance scenario (N i τ<j i * τ), the joint loss function can be divided into three sub-scenario solutions. At the start time of maintenance planning τ sta , the optimal delayed maintenance time for each component is Assume that the delayed joint maintenance time of components is N i τ,s represents the time variable to the next failure.

[0057] If s<N i τ, since corrective repair is required immediately after the component failure is detected, at this time, no matter whether the repair time is moved forward or not, only the corrective replacement operation will be performed, so no additional loss will be caused in this case. The joint loss function in this case is:

[0058] CO 11 =0 (12)

[0059] If N i τ<s<j i * τ, the maintenance execution time is advanced to avoid downtime cost C i,d ·T d and untimely logistics support costs C i,f , but at the same time it shortens the service life of the components, and this loss is reduced by the optimal long-term cost rate at this time. To reflect. The joint loss function in this case is:

[0060]

[0061] Where, τ is the fixed state monitoring interval; F i (j i * τ|τ sta ),F i (Ni τ|τ sta ) is τ sta Based on the remaining life probability density function after online correction at time j i * τ,N i τ is the probability of failure; N i τ is the delayed joint maintenance time.

[0062] If i * τ<s, since failure is a hypothetical future event, the occurrence of failure has no effect on the deviation of component maintenance tasks, but the deviation will shorten the service life. The joint loss function in this case is:

[0063]

[0064] In summary, the joint cost loss function of maintenance control under proactive maintenance is:

[0065]

[0066] Similar to proactive maintenance, deferred maintenance It also contains three similar sub-scenarios, and its loss function is determined by the following formula:

[0067]

[0068] It can be seen that the two joint loss functions CO1 and CO2 under advance maintenance and postponed maintenance have similar forms. This is because advance maintenance and postponed maintenance are inverse forms of each other in mathematical form and actual physical meaning. Integrating them into a unified expression CO i (N i |τ sta ):

[0069]

[0070] In the formula, sgn(j i * -N i ) is the symbol function,

[0071] Apart from the additional cost loss caused by the deviation of the joint maintenance time from the optimal maintenance time of each structural component, a major advantage of combining the maintenance activities of multiple structural component systems is that the maintenance cost C s,R Therefore, the joint maintenance gain function is defined as the difference between the saved shared maintenance cost and the loss function. For a multi-structure component system, some of the components form a joint maintenance group G, then there is a gain function C(G):

[0072]

[0073] Among them, |G| represents the number of components in the maintenance group, the maintenance group composition G and the delayed maintenance execution time N i τ is the decision variable.

[0074] Step 5: Iterate group selection heuristic strategy solution algorithm planning.

[0075] The dynamic programming algorithm is an optimization algorithm for solving multi-stage decision-making problems and has been widely used in many fields such as production scheduling and resource allocation. In the chain structure of the dynamic programming algorithm, the boundary conditions are used as the starting point of the algorithm, and the optimal solution is sought by recursion stage by stage. This recursive effect makes the entire strategy generation process constantly change based on the state. Correspondingly, in the maintenance strategy of the present invention, after the maintenance activity is completed, the status of the component will be updated, and each maintenance grouping structure depends on the last time, and it will continue to roll and iterate during the life cycle. Therefore, the dynamic programming algorithm can be used for rapid iterative solution.

[0076] On this basis, the maintenance optimization model described in formula (18) is extended to some components GL in the system, which meet the damage accumulation value of (ξ i ,L i ) and are not currently planned to any maintenance group. Arrange the optimal maintenance time of the individual components that meet this requirement in ascending order: The corresponding component order {i1,i2,i3,…,i n ,}, thus, the grouping problem is converted into a multi-stage decision problem of sequentially deciding whether to add the current component to the previous existing group. This "sequential grouping mechanism" can bring more reasonable algorithm starting points and decision variables. When the size of the maintenance group is the same, it has the same gain cost and lower loss cost. Decompose GL into a set of mutually exclusive subsets GL = {G1, G2, ..., G l}, each subset represents a joint maintenance group, and these connection groups are arranged sequentially to obtain the optimal maintenance grouping structure of GL:

[0077]

[0078] Where C(GL) is the set of joint maintenance groups GL = {G1, G2, ..., G l}, G b ∈GL represents any maintenance subset in GL, i∈G b Indicates maintenance group G b Part i in CO i (N i |τ sta ) is the component i in τsta The joint loss function at the moment N * τ,GL * = {G1 * ,G2 * ,…,G l *} respectively represent the optimal joint maintenance time and grouping structure corresponding to the optimal gain function C(GL).

[0079] The following is an iterative group selection heuristic dynamic programming algorithm to solve the optimal maintenance grouping. The solution process is regarded as an n-stage decision problem. Let qua p For the structural component i in "sequential grouping" p-1 The number of components in the maintenance group, representing the starting state of the sub-phase p (p≤n). p =0 means the decision is based on structural component i p Starting a new maintenance group, there is no change in the cost of joint maintenance. The indicator function v p (qua p ,dec p )=0.dec p =1 means the decision is to make structural component i p Join the current group (i.e. component i p-1 group), then a maintenance gain will be generated at the joint maintenance level, and the indicator function is Combination repair bonus: v p (qua p ,dec p )=argmax{C(G p )}. Accordingly, the state transition equation and Bellman equation for constructing the algorithm are as follows, and the solution is based on this equation:

[0080]

[0081] Since the state of the first stage is known, qua1=0, dec1=1, then the second stage is qua2=1, so the above formula only needs to be solved to the second stage. qua2=1 is equivalent to the initial state of the whole process being known, f2(qua2)=f1(qua1), at this time, the reverse dynamic programming algorithm can be used to iterate from the nth stage to the second stage to solve the optimal value.

[0082] Since adjacent maintenance groups are related and the damage status information is updated with maintenance and monitoring, the maintenance planning result only takes the first subset in the final optimal sequence. We divide the component state set into {0, 1, 2}, which represent the normal stage, wear stage and failure stage respectively, and the maintenance threshold ξ of the corresponding divided stage i, L i If the state set of the detected components at the monitoring point changes, a new round of system joint maintenance control scheduling will be immediately added to the original maintenance plan. If any component fails, corrective maintenance will be performed on the component immediately.

[0083] The advantages and beneficial effects of the present invention are:

[0084] ① The present invention realizes the dynamic capture of environmental heterogeneity patterns in the damage law of the structural system to carry out more accurate remaining life prediction activities and ensure the sensitivity of the damage feature identification process;

[0085] ② The present invention can realize adaptive dynamic update of system maintenance schemes according to real-time status evaluation information of structural components and historical maintenance records, and can quantify the correlation among adjacent maintenance schemes, with strong adaptability and high intelligence level;

[0086] ③ Compared with the traditional preventive maintenance method, the present invention achieves a balance between extending the remaining service life of structural components and preparing buffers for maintenance resources, thus ensuring the scientific nature of the maintenance plan;

[0087] ④ The iterative group selection heuristic dynamic programming algorithm designed by the present invention is fast and simple to iterate, which reduces the space and time complexity of model solution, has strong operability and high flexibility;

[0088] ⑤ The method described in the present invention is scientific, has theoretical support, and its application field is closely in line with reality, and has broad promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 The present invention is a flow chart of the method.

[0090] Figure 2 This is a schematic diagram of the structure of a certain type of train bogie. DETAILED DESCRIPTION

[0091] The present invention will be further described in detail below with reference to examples.

[0092] The method of the present invention can be verified on a certain type of train bogie. As a typical complex structural system, the train bogie includes a frame, a suspension device, an axle box, a wheel set and a gear box, etc. Figure 2As shown, it is responsible for (1) supporting the vehicle body, bearing and distributing external loads from the wheel rail, vehicle body, wheels, etc.; (2) guiding the vehicle to pass through the curve smoothly; (3) buffering the vibration and impact between the railway and the vehicle; (4) transmitting traction and braking force to ensure the normal operation of the vehicle. The functions of various structural components are different, resulting in various failure modes. At the same time, the train has a long mileage and the service environment is relatively changeable and harsh, which causes the slight dynamic time variation of the damage accumulation trajectory of each component. The present invention uses the test data provided by a company, takes the mileage as the measurement coordinate, and selects three key components of wheel tread, gearbox gear and axle box rolling bearing as maintenance components. For the wheel tread and gearbox gear, the vibration signal (unit: dB) is used as the damage characteristic value; for the rolling bearing, the root mean square data of the outer ring vibration signal after denoising (unit: dB) is selected as the damage characteristic value, and the feasibility and advantages of the method are verified by the following steps.

[0093] Step 1: Modeling the damage accumulation trajectory of structural components under environmental heterogeneity mode.

[0094] By analyzing the historical failure data and online damage information of the three types of components, it is determined that the expression of the time-space scale conversion function in the damage accumulation model is a power function: First, substitute the historical data into equations (1)-(3) to train the common damage parameters p0,β i ,σ i 2 The results are shown in the following table. On this basis, the key damage parameter v is continuously modified as the component status changes by using online damage information combined with formula (4). i Since the steps for each parameter correction are the same, only the results of a certain parameter correction are listed here. p h For demonstration (similarly, the following steps are carried out based on this data as a demonstration), as shown in Table 1. At the same time, according to the actual damage form and failure mode of different components, the maintenance threshold parameters ξ of all components are provided in Table 1 i ,L i .

[0095] Table 1 Damage parameter information of three types of components

[0096]

[0097]

[0098] Step 2: Component life prediction considering heterogeneous conditions of environmental damage.

[0099] Substituting the corrected parameter data directly into equations (5)-(7) can solve the remaining life probability density function and damage increment probability density function of each component.

[0100] Step 3: Optimization of delayed maintenance control of single structural components.

[0101] Table 2 provides the maintenance cost information of three types of components, which is defined based on actual usage. In addition, the maintenance interval is τ = 5 × 10 4 km, shareable maintenance cost C S,R =50×10 4 Substituting the remaining life probability density function and the damage increment probability density function obtained in step 2 into equations (8)-(11) and solving the minimum value of equation (11), we can obtain the optimal delayed maintenance time and the corresponding minimum long-term maintenance cost rate of each structural component, as shown in Table 3.

[0102] Table 2 Maintenance cost information of three types of parts

[0103]

[0104] Table 3 Structural component level maintenance optimization results

[0105]

[0106] Step 4: Quantification of the monomer structure component-system joint maintenance mapping function.

[0107] Substituting the parameter data, life function and increment function information of steps 1 to 3 into equations (12)-(16), we can obtain a joint loss function in the form of equation (17), which represents the cost loss caused by joint maintenance; then substituting it into equation (18) can obtain the decision variables (maintenance scale G and maintenance execution time N) i τ) is a cost-benefit function.

[0108] Step 5: Iterate group selection heuristic strategy solution algorithm planning.

[0109] The cost-benefit function obtained in step 4 is extended to some components GL in the system. These components satisfy the damage accumulation value of (ξ i ,L i ) and is not currently planned into any maintenance group, a maintenance grouping structure in the form of formula (19) can be obtained, and then the optimal solution of the maintenance grouping structure is solved by the heuristic optimization algorithm of the present invention according to the state transfer equation and Bellman equation provided by formulas (20)-(21). Whenever the component state set {0, 1, 2} changes (this state set represents the normal stage, the wear stage and the failure stage, respectively, the corresponding maintenance threshold ξ i , L i), then repeat steps 1 to 5, that is, immediately add a new round of system joint maintenance control scheduling on the basis of the original maintenance plan, and finally obtain the bogie structure system in 280×5×10 4 The results of preventive joint maintenance decision within kilometers of driving are shown in Table 4.

[0110] Table 4 Adaptive preventive maintenance optimization results

[0111]

[0112] The case results show that the method of the present invention can capture and predict the real-time damage cumulative distribution model of system components under the influence of dynamic environmental heterogeneous factors, thereby constructing a maintenance mapping function of a single structural component-system, and optimizing the joint maintenance control strategy for structural systems with maintenance economic relevance, achieving the expected purpose.

[0113] In summary, the present invention proposes an adaptive joint maintenance control method for structural systems under environmental heterogeneous damage. It targets multi-structural component systems under time-varying environmental conditions. First, it constructs a structural damage evolution trajectory model containing environmental heterogeneous patterns, carries out online correction of component failure trajectories and real-time update of remaining life, and then constructs a long-term maintenance cost rate function to form a delayed maintenance strategy for structural components based on self-updating trajectory parameters to optimize resource allocation information and extend service life. Furthermore, a two-stage modeling method from single structure level to system level is adopted, and the maintenance cost sharing property between components is used to quantify the joint maintenance loss function and gain function. Finally, a heuristic optimization algorithm is designed to achieve decoupling of the maintenance grouping problem. The case results show that the present invention is applicable to fields such as intelligent maintenance control of damage accumulation type structural systems under non-uniform environmental conditions, and has strong operability.

Claims

1. An adaptive joint maintenance control method for a structural system under environmental heterogeneous damage, characterized in that: Here are the steps: Step 1: Modeling the damage accumulation trajectory of structural components under environmental heterogeneity; Introducing adaptive key damage parameter ν i , the parameter is modified in real time under the global Bayesian update system to capture the impact of environmental changes on the system / component failure patterns; Assume that any state monitoring point is hτ, and the damage accumulation data observed at that moment is x i (hτ), according to the parameter ν i The normal distribution property of , under the global Bayesian update framework, obtains the parameter ν after online correction at time hτ i The latest distribution of , the posterior distribution is Indicates that is the mean of the posterior distribution, p h is the variance of the posterior distribution: Step 2: Component life prediction considering heterogeneous conditions of environmental damage; According to the statistical independence assumption of the Wiener process, at the state monitoring point hτ, the damage accumulation within the time Δt has an increment Δx i (hτ) = x i (hτ+Δt)-x i (hτ), obeys the normal distribution; therefore, when we get the parameter ν i The corrected posterior distribution at this time Then, according to the conditional probability density and marginal probability density of continuous random variables, the probability density function g of the damage increment within the modified Δt time at time hτ is obtained. i (Δx|Δt; hτ), satisfies the following formula: f v (ν i,h |Δt; hτ) represents the parameter ν at the time hτ i The probability density function of the posterior distribution, f Δ (x|ν i,h ; Δt; hτ) represents the known parameter ν i The posterior estimate ν i,h The probability density function of the damage increment in Δt under the condition of i ),Λ(hτ;β i ) represent the space-time scale conversion function values ​​corresponding to hτ+Δt and hτ respectively; Step 3: Optimization of delayed maintenance control of single structural components; The damage accumulation state of a component is continuous. When the damage accumulation of component i is between the maintenance threshold interval (ξ i ,L i ), the execution time of preventive maintenance control is delayed i τ is the fixed monitoring period, j i is an integer variable to be optimized; using the probability density function of the damage increment g i (·), assuming that the start time of a preventive maintenance plan is τ sta , the probability distribution of the average monitoring frequency k between adjacent preventive maintenance control activities is It is expressed as: Among them, u, y, ω are all integral intermediate variables, and their physical meanings represent the damage increment; g i (u|j i τ; τ sta ),g i (ω|τ; τ sta ),g i (y|j i τ; τ sta ),g i (u|τ; τ sta ),g i (ω|(kj i -2)τ; τ sta ) indicates that in τ sta The probability density function of the damage increment after online correction at the moment; L i -ω,ξ i -ω,L i -ω-u is the upper and lower limits of the integral, and its physical meaning is the damage increment; ξ i ,L i thresholds for preventive and corrective maintenance; Step 4: Quantification of the mapping function of the monomer structure component-system joint maintenance; According to the deviation of the joint maintenance time, it is divided into two cases: early maintenance and deferred maintenance; Maintenance planning start time τ sta , the optimal delayed maintenance time for each component is Assume that the delayed joint maintenance time of components is N i τ, s represents the time variable to the next failure; the unified expression is CO i (N i |τ sta ): In the formula, sgn(j i * -N i ) is the symbol function, In addition to the additional cost loss caused by the deviation of the joint maintenance time from the optimal maintenance time of each structural component, the maintenance activities of the multi-structural component system are combined to share the maintenance cost C s,R ; Therefore, the joint maintenance gain function is defined as the difference between the saved shared maintenance cost and the loss function; for a multi-structure component system, some of the components form a joint maintenance group G, then there is a gain function C(G): Among them, G represents the number of components in the maintenance group, the maintenance group composition G and the delayed maintenance execution time N i τ is the decision variable; Step 5: Iterative group selection heuristic strategy solution algorithm planning; In the maintenance strategy, after the maintenance activity is completed, the status of the components will be updated, and each maintenance grouping structure depends on the previous one, and it will be continuously rolled out and iterated during the life cycle; dynamic programming algorithm is used for fast iterative solution.

2. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 is characterized by: In step one, offline training of common damage parameters is also included: For component i, i∈{1,2,…,q} among q components of the same type, the key damage parameter ν i The prior distribution of Satisfies the normal distribution, where is the prior distribution mean, p0 is the prior distribution variance; τ is a fixed monitoring period, in the state monitoring time vector T i =(0,τ,2τ,…,n i The damage accumulation data vector obtained on τ)′ is X i =(0,x i (τ),x i (2τ),…,x i (n i τ))′, where n i is the total number of monitoring times for component i; according to X i The multidimensional normal distribution property of i ) and covariance Cov(X i )for: Therefore, based on the damage accumulation data of q components of the same type, the comprehensive log-likelihood function of the offline parameters is obtained as follows: Among them, the offline parameters include ν i The mean of the prior distribution of Variance p0, time scale transformation parameter β of the damage accumulation process i , trend fluctuation parameter σ i .

3. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 2 is characterized by: In step 1, the online correction of heterogeneous damage parameters is also included: Based on the common damage parameter offline training module, the key damage parameter ν is obtained i The prior distribution of the parameter ν i Online correction work; the prior distribution in Bayesian statistical inference is regarded as the parameter ν i The empirical cognition is obtained through historical data statistics, while the posterior distribution integrates the current sample observation information with the information in the prior distribution to make real-time corrections to the parameter information.

4. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 is characterized by: In step 2, the life T of the component damage cumulative evolution process based on the Wiener process modeling is defined by the first arrival time, that is, the first time the corrective maintenance threshold L is reached. i The time T = inf{t:x i (t)≥L i |x i (0)<L i The lifetime of the linear Wiener process satisfies the inverse Gaussian distribution, and its physical background is the damage accumulation process that obeys the standard linear drift Brownian motion. After introducing the parameter uncertainty caused by the space-time scale conversion function and the environmental damage heterogeneity, the lifetime of the generalized Wiener process no longer obeys the inverse Gaussian distribution due to the randomness of the state, and needs to be solved in combination with the total probability formula. Calculate the remaining life T of the component after online correction at the condition monitoring point hτ l ={t l =T-hτ:x i (T)≥L i |x i (hτ)<L i ,T>hτ}, its probability density distribution function f i (t l |hτ) and the distribution function F i (t l |hτ) is expressed by the following formula: Among them, Λ(hτ+t l β i ),Λ(hτ;β i ) represent hτ+t l The value of the space-time scale conversion function corresponding to the time hτ.

5. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 is characterized in that: In step 3, since the initial damage level is zero, k≤j i +1 This situation does not exist; k = j i +2 and k ≥ j i +3 Both cases represent that the damage accumulation value of component i is (ξ i ,L i ) and remains in this state after a delay period. Only in this case is it possible to carry out delayed preventive maintenance normally; If component i is monitored to exceed the corrective maintenance threshold L i , then the corrective maintenance will be started immediately, at the maintenance planning time τ sta , solve the probability distribution of the average monitoring frequency k between adjacent corrective maintenance control activities It is expressed by the following formula; the τ mentioned here sta Refers to all maintenance planning start-up times. When it comes to a specific time, just substitute the relevant parameters and values ​​after online correction; in, represents the distribution function of the damage increment within a monitoring period, In formula (9), m and n are intermediate variables, which are integers and represent the probability summation range in the time increment dimension, which are 1 to j respectively. i and 1~+∞;υ,y,u,ω,ο all represent integral intermediate variables, and their physical meanings represent damage increments; g i (υ|τ;τ sta ),g i (y|(j i -m-1)τ; τ sta ),g i (u|τ; τ sta ),g i (ω|(n-1)τ; τ sta ),g i (ο|(k-2)τ; τ sta ) indicates that in τ sta The probability density function of the damage increment after online correction at the moment; L i -ω,ξ i -ω,L i -ω-u,L i -ω-uy is the upper and lower limits of the integral, and its physical meaning is the damage increment; ξ i ,L i is the maintenance threshold.

6. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 or 5, characterized in that: In step 3, since the fault state is only identified through monitoring, this hidden nature will cause the system to shut down. Component i is not available at the maintenance planning start time τ sta The possible average downtime in the future is T d The following equation gives: Get the long-term maintenance cost rate η of component i i (j i |τ sta )for: Among them, f i (t l |τ sta ) is the maintenance planning start time τ of component i sta The probability density distribution function of the remaining life after online correction; a is an intermediate variable, which takes an integer; aτ, (a-1)τ represent the upper and lower limits of the integral respectively, and their physical meaning is the time range of component failure; C i,I is the single monitoring cost, C s,R is the shareable maintenance cost, C i,R is the independent maintenance cost of each component, C i,d is the downtime loss per unit time, C i,f Losses due to untimely logistics support; They represent the maintenance planning start time τ sta The probability of the average monitoring frequency k between adjacent preventive maintenance control activities / corrective maintenance control activities; Since the maintenance cost can be shared by a fixed amount C S,R It exists in every maintenance activity, so whether C is considered in structural component level maintenance is S,R It has no effect on the optimization results; by solving the minimum value of the above function Corresponding to the best delayed maintenance time for each component 7. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 is characterized by: In step 4, the joint cost loss function of maintenance control under proactive maintenance is:

8. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 is characterized by: In step 4, the delayed maintenance loss function is determined by the following formula:

9. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 is characterized by: In step 5, the maintenance optimization model is extended to some components GL in the system, which meet the damage accumulation value of (ξ i ,L i ) and are not currently planned to any maintenance group; the optimal maintenance time of the individual components that meet this requirement is arranged in ascending order: The corresponding component order {i1,i2,i3,…,i n ,}, thus, the grouping problem is transformed into a multi-stage decision problem of sequentially deciding whether to add the current component to the previous existing group; when the size of the maintenance group is the same, it has the same gain cost and lower loss cost; decompose GL into a set of mutually exclusive subsets GL = {G1, G2, ..., G l }, each subset represents a joint maintenance group, and these connection groups are arranged sequentially to obtain the optimal maintenance grouping structure of GL: Where C(GL) is the set of joint maintenance groups GL = {G1, G2, ..., G l }, G b ∈GL represents any maintenance subset in GL, i∈G b Indicates maintenance group G b Component i in CO i (N i |τ sta ) is the component i in τ sta The joint loss function at the moment N * τ,GL * ={G1 * ,G2 * ,…,G l * } respectively represent the optimal joint maintenance time and grouping structure corresponding to the optimal gain function C(GL).

10. The method for adaptive joint maintenance control of a structural system under environmental heterogeneous damage according to claim 1 or 9, characterized in that: Design an iterative group selection heuristic dynamic programming algorithm to solve the optimal maintenance grouping; Consider the solution process as an n-stage decision problem, assuming qua p For the structural component i in "sequential grouping" p-1 The number of components in the maintenance group represents the starting state of the sub-phase p, p≤n; dec p =0 means the decision is based on structural component i p Starting a new maintenance group, there is no change in the cost of joint maintenance. The indicator function v p (qua p ,dec p )=0;dec p =1 means the decision is to make structural component i p Add to the current group, i.e. component i p-1 The group in which they are located will generate a maintenance gain at the joint maintenance level. The indicator function is to Combination repair bonus: v p (qua p ,dec p )=argmax{C(G p )}. Accordingly, the state transition equation and Bellman equation for constructing the algorithm are as follows, and the solution is based on this equation: Since the state of the first stage is known to be qua1=0, dec1=1, then the state of the second stage is qua2=1, so it is only necessary to solve to the second stage; qua2=1 is equivalent to the initial state of the whole process being known, f2(qua2)=f1(qua1), at this time, the reverse dynamic programming algorithm is used to iterate from the nth stage to the second stage to solve the optimal value; Since adjacent maintenance groups are related and the damage status information is updated with maintenance and monitoring, the maintenance planning result only takes the first subset in the final optimal sequence. The state set of the component is divided into {0, 1, 2}, which represent the normal stage, the wear stage and the failure stage respectively, and the maintenance threshold ξ corresponding to the divided stage is i , L i If the state set of the detected components at the monitoring point changes, a new round of system joint maintenance control scheduling will be immediately added to the original maintenance plan. If any component fails, corrective maintenance will be performed on the component immediately.

Images