Equipment maintenance decision-making method based on health state network and maintenance threshold interval

Abstract

The invention relates to the technical field of multi-component equipment maintenance, in particular to an equipment maintenance decision-making method based on a health state network and a maintenance threshold interval. An equipment maintenance decision method based on a health state network and a maintenance threshold interval provides a flexible, dynamic and efficient solution for maintenance decision of multi-component equipment by establishing the equipment health state network, setting the maintenance threshold interval and combining a particle swarm optimization algorithm with engineering constraints. According to the method, the maintenance cost, the system reliability and the operation and maintenance timeliness can be balanced through the flexible judgment and multi-dimensional optimization of the maintenance threshold interval under the condition of considering the key degree difference of the parts and the dynamic working condition requirements, so that the collaborative optimization of the maintenance opportunity and the maintenance strategy is realized, and the operation and maintenance efficiency of the equipment under the complex working condition is ensured to be improved.

Description

Technical Field

[0001] This invention relates to the field of multi-component equipment maintenance technology. It determines the optimal component maintenance scheme by combining the status and maintenance threshold range of the equipment health status network with actual maintenance engineering constraints and optimization objectives, thereby assisting in actual maintenance decision-making. Background Technology

[0002] Currently, the maintenance of multi-component equipment commonly uses fixed health thresholds as the basis for maintenance decisions. This method has significant technical limitations. Maintenance is triggered when the health of a component falls below the preset threshold, failing to differentiate the criticality of different components and struggling to adapt to dynamically changing operating conditions. Especially when dealing with multi-component systems with complex interrelationships, a single threshold decision often leads to two extremes: on the one hand, premature maintenance may result in wasted resources, and on the other hand, delayed maintenance may trigger a chain reaction of failures.

[0003] Furthermore, traditional solutions largely rely on static models, failing to update component status based on changing monitoring data, leading to decisions lagging behind actual operating conditions. Particularly when dealing with complex, interconnected multi-component systems, existing technologies lack the ability to update the health status network retrospectively, cannot integrate abnormal test indicators and non-test evidence, and struggle to accurately reflect fault propagation paths between components. This static analysis deficiency exacerbates the risk of misjudgment and lacks the ability to synergistically optimize maintenance costs, system reliability, and maintenance efficiency. Especially when facing real-world constraints such as fluctuating spare parts inventory and differences in personnel skills, dynamic adjustments to maintenance decisions are difficult. These deficiencies severely restrict the improvement of operational efficiency for complex equipment systems. Summary of the Invention

[0004] This invention addresses the shortcomings of existing maintenance decision-making methods by proposing a maintenance decision-making method based on an equipment health status network and threshold intervals. By establishing a dynamic health status threshold mechanism, a flexible decision-making interval is formed between the mandatory maintenance threshold and the observation threshold, effectively avoiding the rigid decision-making defects of traditional methods. When a component's health status falls within this interval, the system initiates a multi-dimensional optimization algorithm, comprehensively considering key factors such as the component's weight distribution in the fault sensitivity matrix, current spare parts inventory levels, and the skill matching degree of maintenance personnel, achieving coordinated optimization of maintenance timing and maintenance strategy. The method innovatively integrates the topological correlation characteristics of the health status network into the decision-making model, enabling maintenance decisions to ensure the reliability requirements of critical components while also considering the overall system reliability, achieving a multi-objective balance between maintenance cost, system reliability, and maintenance timeliness.

[0005] The method includes the following steps: Step 1: Based on expert knowledge and the identified failure modes, determine the basic structure of the equipment health status network. Then, based on the maximum likelihood method, learn the prior probabilities of the health status network to obtain the complete health status network.

[0006] The core idea is to find a set of parameters, given a model structure, that maximizes the probability of observed data occurring. For Bayesian networks, suppose we have a set of parameters containing... Historical fault statistics data set of a sample measurement and control system and a Bayesian network structure The goal is to estimate the parameters of the conditional probability distribution (CPD) of each node in the network. .set up It is the set of parameters of a Bayesian network. In the parameters The following dataset was observed The probability of finding the most likely value is also called the likelihood function. Maximum likelihood estimation aims to find the probability that the likelihood function is most likely. Largest parameter In practice, for ease of calculation, the log-likelihood function is taken, i.e.: ; For a node of a Bayesian network Assuming he has Parent nodes Each parent node has a different combination of values. Let... For nodes One of the possible values, Parent node A combination of values. Then the node Take the value from the given parent node. Take down conditional probability The maximum likelihood estimate is: ; in, It is a node in the dataset. The value is x and the parent node The number of samples with the value y. It is a node in the dataset. Values The number of samples.

[0007] Step two involves updating the network health status using variable elimination based on evidence derived from online monitoring data or non-testing methods. The goal of this inference is typically to calculate the conditional probability distribution of one or more query nodes given the evidence (i.e., the values ​​of some nodes are known). Suppose we want to calculate the conditional probability distribution of the query nodes given the evidence... Under the condition of querying nodes conditional probability distribution According to Bayes' theorem, we have: ; in It is a query node and evidence nodes Joint probability distribution of simultaneous specific values Evidence node The probability of taking a specific value. When the inference evidence is a probability value, the Monte Carlo method can be used to update the network state. The updated network contains information on the normal probability and failure probability of each critical maintenance node of the multi-component equipment.

[0008] Step 3: Set the maintenance threshold range Repair nodes with health values ​​above the right boundary of the repair threshold range are in good condition and repair is not recommended; repair nodes with health values ​​below the left boundary of the range are in a pre-failure state and are subject to forced repair; repair nodes with health values ​​within the range are pending and are included in the optimization model for further scheme optimization.

[0009] Step four involves establishing a maintenance effectiveness evaluation function and constructing a dynamic reliability evaluation function for optimization based on the maintenance node-system fault sensitivity matrix determined by the health status network. The maintenance effectiveness evaluation function includes the recovery update benchmark for the health of maintenance nodes in the health status network after implementing various maintenance strategies. The "replacement maintenance" action will restore the health of the maintenance node to "1," while the recovery benchmark for the "imperfect maintenance" action is configured by experts in the maintenance reference information database. (The last sentence appears to be incomplete and possibly refers to a separate step: "After maintenance recovery...") Health is defined as: ; in, This parameter serves as the recovery baseline for module health under different maintenance strategies. Each module is unique to this parameter, and it is configured by users in the database based on expert experience. To prevent the configured baseline from being lower than the original baseline... Here, a default coefficient greater than 1 is given. The default health level for replacement is 1; if no repair is performed, the original health level will not be restored. .

[0010] The failure probability of each maintenance node is obtained by simulating the failure of each maintenance node using a system health status network, thereby determining the maintenance node-system failure sensitivity matrix. Let... This represents a node in the health status network system, where C represents a maintenance node of interest to the health status network. Each node has two states: 0 indicates a fault, and 1 indicates normal operation. (The last sentence appears to be incomplete and possibly refers to a different node.) When ), the probability that system node A is normal is . ; and maintenance node failures ( When ), the probability that system node A is normal is . Maintenance node - A matrix for measuring the system's fault sensitivity. Defined as: ;; in This represents the i-th maintenance node. The denominator serves a normalization function, placing the difference in the numerator within a relatively reasonable range (-1 to 1). A value of 1 indicates a strong positive correlation, while a value of -1 indicates a strong negative correlation. In cases of negative correlation, "1 - health level" is used in the weighted calculation of system health. Based on this matrix reasoning system, the reliability evaluation function considering maintenance effects is as follows: ; This computational method avoids repeatedly calling Monte Carlo simulations to calculate virtual reliability using the health state network during particle swarm optimization. In particular, it can significantly accelerate the optimization solution rate when the health state network is large, and enhance the fast and real-time responsiveness of decision-making.

[0011] Step 5: Establish a maintenance reference information database. This database includes information such as the cost and time of different maintenance strategies at different maintenance nodes, the health recovery benchmark for corrective repairs, spare parts inventory, the working status that allows for maintenance, and the skill level of personnel required for maintenance. Based on this maintenance reference information, establish evaluation functions for total maintenance cost and total time.

[0012] ; Among them, the stage cost penalty factor Inflating maintenance expenses at specific stages in cost calculations; The cost of a single repair for a module under different repair strategies: ; Similarly, the total maintenance time is the product of the maintenance strategy adopted by the module and the corresponding single maintenance time, and then summed. However, considering the parallel execution of maintenance time, the total maintenance time cannot be simply added together. Therefore, the allocation of maintenance tasks must be taken into account during the calculation, which introduces the parameter. : ; in The single repair time of the module under different maintenance strategies: ; To assign tasks to maintenance personnel, different task allocation strategies directly affect the calculation of the total maintenance time. Based on the parallelism issue mentioned above, the following parallel scheduling model is adopted for task allocation to maintenance personnel: ; ; ; Where N represents the total number of tasks requiring repair; This represents the health status of the module corresponding to the i-th maintenance task; It is a sorted sequence of tasks; This indicates the maximum number of maintenance personnel that can be deployed; k represents the maintenance personnel's ID number. This refers to the cumulative maintenance time for the tasks currently assigned to the k-th maintenance worker; This represents the set of tasks assigned to the k-th maintenance worker; used to iterate through the sorted task sequence. The task in; This represents the maintenance time required for the j-th task.

[0013] Step six: Use the particle swarm optimization algorithm with engineering constraints to obtain the optimized maintenance strategy for the maintenance node. The optimization variable is the maintenance strategy adopted by the maintenance node. and maintenance priority , among which strategy Five strategies. In terms of prioritization, the initial maintenance priority is first determined based on the current health status. Then, nodes that can be maintained in parallel are arranged based on the total number of maintenance personnel and their skill matching. Finally, the priority after parallelization is determined to maximize maintenance efficiency and reduce the risk of worsening the fault.

[0014] Set up an engineering constraint checker to constrain the solution domain range and ensure that the algorithm optimization process conforms to engineering constraints: ; The above engineering constraints are, in order: spare parts inventory, working status of maintenance nodes that allow maintenance, matching degree of maintenance personnel skill level, maintenance budget, and estimated lower limit of system reliability after maintenance.

[0015] The fitness evaluation function of the optimization algorithm consists of three parts: the total maintenance cost evaluation function, the total maintenance time evaluation function, and the post-maintenance system reliability evaluation function. During the optimization process, one of them is selected as the main optimization objective, and the others are used as boundary constraints.

[0016] ; in These are the evaluation functions for cost, repair time, and system reliability, respectively. The optimized output repair decision scheme is as follows: ; The breakdown is as follows: maintenance strategy for each maintenance node, maintenance priority, maintenance personnel allocation plan, estimated total maintenance cost, estimated total maintenance time, and estimated equipment reliability after maintenance.

[0017] The advantages of this invention are: By innovatively combining the Bayesian update mechanism of the health status network with maintenance threshold ranges, flexibility and dynamism in maintenance decisions for multi-component equipment are achieved. Dynamic fusion of test data and observational evidence precisely quantifies the health status of components and its associated impacts, enabling flexible judgment through three-tiered threshold ranges (mandatory maintenance / optimization decision / observation and postponement).

[0018] Based on fault sensitivity weights, parallel scheduling algorithms, and actual engineering constraints and objectives, resource allocation is optimized while ensuring system reliability, ultimately forming a closed-loop decision-making system that combines real-time performance, adaptability, and engineering feasibility, significantly improving the operation and maintenance efficiency of complex equipment. Attached Figure Description

[0019] Figure 1 A flowchart of an equipment maintenance decision-making method based on a health status network and maintenance threshold range; Figure 2 For the health status network of the inertial integrated system; Figure 3 The decision range between the health status and maintenance threshold of the maintenance node; Figure 4 The decision-making logic for pre- and final maintenance plans; Figure 5 Overview of maintenance reference information data for each maintenance node (partial attributes); Figure 6 To minimize maintenance costs, the fitness variation during the iterative process is minimized. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0021] Example Taking a certain inertial navigation system as an example, we analyzed the abnormal signs of each level of the system structure and test items, and determined their correlation. The results are shown in Table 1.

[0022] Table 1. Composition and Test Items of the Inertial Navigation System: Failure Modes and Causes Analysis ; ;

[0023] Step 1: Based on the table above and expert experience, establish the basic structure of the node health status network for the multi-maintenance node system. Then, based on historical fault data, use a parametric learning model to obtain the prior inference probability of the health status network, ultimately obtaining the equipment's health status network as shown below. Figure 2 As shown in Table 2, the node definitions and parameters of the health status network are as follows.

[0024] Table 2 Classification of Bayesian Network Node Variables and States ;

[0025] Step 1: Based on the monitoring data, abnormal signs are identified or observed. The health status of the network is updated using the variable elimination method to obtain the posterior health information of the maintenance node, as shown in Table 3.

[0026] Table 3. Maintenance Node Health Update Results ;

[0027] Step 3: Set the maintenance threshold range to [0.2, 0.6] for initial screening of maintenance plans. Maintenance is not recommended for C1, C6, and C7 if their health is above 0.6; maintenance is mandatory for C2 if its health is below 0.2 (ignoring engineering constraints, this is an urgent recommendation); and further decisions are made by the maintenance optimization model if the health of C3, C4, and C5 is within the maintenance threshold range. Figure 3 As shown.

[0028] Step four involves establishing a maintenance effectiveness evaluation function and constructing a dynamic reliability evaluation function for optimization based on the maintenance node-system fault sensitivity matrix determined by the health status network. Specific formulas are detailed in the invention description, and the maintenance effectiveness parameters are provided by decision-making experts based on experience; specific values ​​are available in the maintenance reference information table.

[0029] Step 5: Establish a maintenance reference information database, which includes information such as cost, time, spare parts, and skill requirements for each maintenance node related to maintenance modeling. Taking the accelerometer as an example, the maintenance reference information table is shown in Table 4 and... Figure 4 As shown.

[0030] Table 4 Maintenance Reference Information Table (Taking Accelerometer as an Example) ;

[0031] The formulas for the evaluation functions of maintenance cost and maintenance time are described in the invention description, and the relevant parameters are shown in the maintenance reference information table.

[0032] In summary, the overall decision-making logic is as follows: Figure 5As shown in Table 5, the parameters involved in the maintenance optimization model not only consider the health information of each maintenance node, but also take into account engineering constraints, the decision-maker's system reliability requirements, budget constraints, and optimization objectives to provide maintenance solutions.

[0033] Table 5 Parameters of the Maintenance Decision Model ;

[0034] Step Six: Solve the maintenance decision model using the PSO algorithm with multiple constraints. The optimization objective is set as "minimizing maintenance cost," with "equipment reliability after maintenance" and "total maintenance time" used as optimization constraints in the calculation. The fitness changes during the optimization iteration process are as follows: Figure 6 As shown in Table 6, the suggested information obtained after solving the problem is as follows.

[0035] Table 6 Output of Maintenance Decision Optimization Scheme ;

[0036] The final maintenance plan output includes the maintenance strategy adopted by the module, the maintenance priority, the maintenance personnel allocation plan, and the estimated maintenance cost, total maintenance time, and equipment reliability for implementing this maintenance plan. Specifically: C1, C6, and C7 do not require repair. Repair is recommended for C2 to C5, with a "preventive replacement" strategy for C2 and a "preventive repair" strategy for C3 to C5. The maintenance optimization level and personnel allocation are as follows: Personnel 1 repairs [C2, C5] in sequence, and Personnel 2 repairs [C4, C3] in sequence, in order to maximize maintenance efficiency and minimize potential risks.

[0037] The estimated total cost of the repair using this method is 2150 yuan, the total repair time is 2.3 hours, and the equipment reliability after repair reaches 0.86, providing intuitive reference information for decision-makers' final decision.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for equipment maintenance decision based on health state net and maintenance threshold interval, which is suitable for the technical field of multi-component equipment maintenance, and determines the best component maintenance scheme by combining the state of the equipment health state net and the maintenance threshold interval with the actual maintenance engineering constraints and optimization objectives, characterized in that, The method comprises the following steps: Step one, combing the abnormal signs of equipment test items and the failure modes of each level of the system, analyzing the correlation and building a multi-component equipment health state network. Step two, updating the equipment health state network based on inference evidence to obtain the posterior maintenance node health state. Step three, preliminarily screening the pre-maintenance decision strategy of some nodes according to the posterior health degree sequence and the maintenance threshold interval. Step four, establishing a maintenance effect evaluation function and determining the maintenance node-system failure sensitivity matrix based on the health state network to construct a dynamic reliability evaluation function in optimization. Step five, establishing a maintenance reference information base and constructing a maintenance cost and maintenance time evaluation function. Step six, using a particle swarm optimization algorithm with engineering constraints to solve the optimized maintenance node maintenance strategy.

2. The method of claim 1, wherein: Based on expert knowledge and the failure modes combed above, the basic structure of the equipment health state network is determined, and then the prior probability of the health state network is learned based on a data-driven method to obtain a complete health state network.

3. The method of claim 2, wherein: Inference evidence includes evidence based on online monitoring data or non-test means. The inference evidence of the observation is injected into the health state network, and the variable elimination method is used to update the network health state. The updated network contains the normal probability and failure probability information of each key maintenance node of the multi-component equipment.

4. The method of claim 3, wherein: The maintenance node state is good if the health degree value is higher than the right boundary of the maintenance threshold interval, and maintenance is not recommended; the maintenance node is in a pre-failure state if the health degree value is lower than the left boundary of the interval, triggering forced maintenance; the maintenance node with a health degree between the interval is pending and is included in the optimization model for further scheme optimization.

5. The method of claim 4, wherein: The maintenance effect evaluation function includes the recovery update benchmark of the health degree of the maintenance node in the health state network after adopting various maintenance strategies. The "replacement maintenance" behavior will restore the health degree of the maintenance node to "1", and the recovery benchmark of the "imperfect maintenance" behavior is configured by experts in the maintenance reference information base. The failure probability of each maintenance node is simulated based on the system health state network to obtain the system failure probability, and the maintenance node-system failure sensitivity matrix is determined. Based on the matrix, the reliability evaluation function of the system after considering the maintenance effect is inferred, which accelerates the calculation rate of the fitness function of the optimization algorithm and enhances the rapid and real-time responsiveness of the decision.

6. The method of claim 5, wherein: The maintenance reference information base contains information such as the cost and maintenance time of different maintenance nodes adopting different maintenance strategies, the health degree recovery benchmark of repair maintenance, spare parts inventory, allowable maintenance working state, and personnel skill level required for maintenance. Based on the maintenance reference information, the total maintenance cost and total time evaluation function is established.

7. The method of claim 6, wherein: The optimization variables are the maintenance strategies and the maintenance priorities of the maintenance nodes. The strategies include no maintenance, post-maintenance imperfect repair, post-maintenance replacement, preventive imperfect repair and preventive replacement. The priorities are determined based on the current health degree, the total number of maintenance personnel and the skill matching degree. The optimization process maximizes the maintenance efficiency and reduces the risk of failure aggravation. A packaging engineering constraint verifier is used to constrain the solution domain range and ensure that the algorithm optimization process meets the engineering constraints. The constraints include spare parts inventory, current system working state, number and skill level of maintenance personnel, maintenance budget and lower limit of estimated system reliability after maintenance. The fitness evaluation function of the optimization algorithm includes three parts: total maintenance cost evaluation function, total maintenance time evaluation function and system reliability evaluation function after maintenance. One of them is selected as the main optimization target, and the others are used as boundary constraints. The output of the optimization is the maintenance decision scheme, which includes the maintenance strategy, maintenance priority, maintenance personnel allocation scheme, estimated total maintenance cost, estimated total maintenance time and estimated equipment reliability after maintenance.

Images