Reusable rocket lru repair level dynamic decision method and system

Abstract

The application discloses a reusable rocket LRU repair level dynamic decision method and system, relates to the technical field of spaceflight equipment maintenance management, and comprises the following steps: collecting full life cycle data of the LRU; constructing an association library after preprocessing, matching a loss mode, a maintenance candidate scheme and a key constraint threshold according to the cumulative launch number N; quantitatively scoring the maintenance candidate scheme based on a four-dimensional index system to obtain a four-dimensional quantitative score; dynamically allocating a weight according to N and combining a forced constraint judgment to perform weighted calculation on the four-dimensional quantitative score calculation and select an optimal maintenance scheme. The prior art mainly depends on the LORA method of a traditional spacecraft and an aircraft maintenance level decision method. When these methods are applied to the reusable rocket, due to the fact that the model itself lacks a sensing and response mechanism for the key state variable "cumulative launch number", the decision accuracy is insufficient, and the maintenance scheme optimization demand in a high-frequency reuse scene cannot be met.

Description

Technical Field

[0001] This invention relates to the field of aerospace equipment maintenance management technology, and in particular to a dynamic decision-making method and system for the repair level of reusable rocket LRUs. Background Technology

[0002] Repair-level decisions for reusable rocket LRUs are crucial for ensuring rocket reuse reliability, optimizing operating costs, and improving launch efficiency. Existing technologies primarily rely on two approaches: first, the traditional spacecraft LORA method, which prioritizes economics, employs a simplistic decision-making logic, and fails to consider mission timeliness and equipment reuse requirements; second, aircraft maintenance-level decision-making methods, with fixed indicator weights, lack a dynamic adjustment mechanism for the cumulative wear and tear from multiple rocket reuses. When applied to the high-frequency reuse scenarios of reusable rockets, these methods reveal fundamental technical flaws in their underlying models: the static economic model of traditional LORA cannot respond to changes in mission pace, while the fixed-weight decision-making model for aircraft maintenance cannot perceive the accumulated equipment wear and tear represented by the cumulative number of launches N. Therefore, existing technologies lack a closed-loop optimization mechanism that dynamically couples the equipment state (N) with maintenance decision weights, resulting in insufficient decision accuracy and a lack of continuous evolution capabilities. Summary of the Invention

[0003] This application provides a dynamic decision-making method and system for the repair level of reusable rocket LRUs, which solves the technical problems of existing equipment maintenance decision-making methods lacking a dynamic adjustment mechanism for multiple reuses of equipment, having a single decision logic and lacking continuous optimization capabilities.

[0004] The first aspect of this application provides a dynamic decision-making method for the repair level of a reusable rocket LRU. The method includes: collecting full lifecycle data of the reusable rocket LRU, including basic data, maintenance cost and cycle data, reliability data, and rocket fleet resource data; preprocessing the full lifecycle data, and then querying an association database constructed based on K-means clustering and FP-Growth algorithms to match the LRU's wear pattern, maintenance candidate schemes, and key constraint thresholds based on the cumulative launch count N and real-time performance parameters; quantifying and scoring the maintenance candidate schemes based on a four-dimensional index system to obtain a four-dimensional quantitative score; dynamically allocating the weights of the four-dimensional indexes based on the cumulative launch count N using a recursive least squares online learning model with a forgetting factor, and combining this with mandatory constraint judgment to perform a weighted calculation on the four-dimensional quantitative score to select the optimal maintenance scheme; periodically using the NSGA-II algorithm to perform multi-objective optimization of the decision rules based on a time-series database, and using the Kappa coefficient to verify the stability of the optimization results, updating the decision logic after the verification is passed.

[0005] The second aspect of this application provides a dynamic decision-making system for the repair level of a reusable rocket LRU. The system includes: a full lifecycle data acquisition module for collecting full lifecycle data of the reusable rocket LRU, including basic data, maintenance cost and cycle data, reliability data, and rocket fleet resource data; a correlation library construction module for preprocessing the full lifecycle data and then matching the LRU's wear pattern, maintenance candidate schemes, and key constraint thresholds based on the cumulative launch count N and real-time performance parameters by querying a correlation library constructed based on K-means clustering and FP-Growth algorithms; and four-dimensional quantization. The scoring acquisition module quantifies the candidate maintenance solutions based on a four-dimensional index system to obtain a four-dimensional quantitative score. The optimal maintenance solution acquisition module uses a recursive least squares online learning model with a forgetting factor to dynamically allocate the weights of the four-dimensional indexes based on the cumulative number of launches N, combined with mandatory constraint judgment, to perform a weighted calculation on the four-dimensional quantitative score and select the optimal maintenance solution. The decision logic update module periodically uses the NSGA-II algorithm to perform multi-objective optimization of the decision rules based on a time-series database, and uses the Kappa coefficient to verify the stability of the optimization results. After passing the verification, the decision logic is updated. One or more technical solutions provided in this application have at least the following technical effects or advantages: This application collects full lifecycle data of reusable rocket LRUs, preprocesses it, and queries an association library built based on K-means clustering and FP-Growth algorithms to match wear patterns, candidate maintenance schemes, and key constraint thresholds. A quantitative score is obtained based on a four-dimensional indicator system, and indicator weights are dynamically allocated according to the cumulative number of launches. The optimal maintenance scheme is selected by weighted calculation of the score combined with mandatory constraint judgments. Subsequently, parameters are calibrated, weights are adjusted, and rules are optimized through a time-series database, thereby achieving dynamic adaptation and continuous optimization of the maintenance scheme. This enables reusable rocket LRU repair decisions to accurately adapt to the needs of different launch frequencies, ensuring the reliability of rocket reuse and optimizing operating costs. The technical effect is to achieve dynamic adaptation and continuous optimization of maintenance decisions based on equipment usage frequency, improving the accuracy and adaptability of decisions, and balancing maintenance efficiency and operating cost optimization. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a flowchart illustrating the dynamic decision-making method for the repair level of a reusable rocket LRU provided in this application embodiment.

[0008] Figure 2 This is a schematic diagram of the structure of the reusable rocket LRU repair level dynamic decision system provided in the embodiments of this application.

[0009] Figure labeling: Module 1 for full lifecycle data acquisition, Module 2 for association library construction, Module 3 for four-dimensional quantitative scoring acquisition, Module 4 for optimal maintenance solution acquisition, and Module 5 for decision logic update. Detailed Implementation

[0010] This application provides a dynamic decision-making method and system for the repair level of reusable rocket LRUs, which solves the technical problems of existing equipment maintenance decision-making methods lacking a dynamic adjustment mechanism for multiple reuses of equipment, having a single decision logic and lacking continuous optimization capabilities.

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0012] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] Example 1, as Figure 1 As shown, a dynamic decision-making method for repair levels of reusable rocket LRUs is provided, wherein the method includes: Collect full lifecycle data for reusable rocket LRUs, including basic data, maintenance cost and cycle data, reliability data, and rocket team resource data.

[0014] In this embodiment of the application, LRU, or Line Replaceable Unit, is a core component on a reusable rocket that has independent functions and can be disassembled, replaced, and repaired independently, providing support for the operation of various rocket systems.

[0015] Specifically, the first step is to establish a multi-dimensional data acquisition framework to ensure comprehensive data coverage and compliance with subsequent processing requirements. For LRU basic data, the cumulative launch count N and the planned rocket reuse interval are collected. The cumulative number of launches N is automatically read from the rocket launch log, and the planned rocket reuse interval is... Data is obtained from the rocket mission planning system to ensure the authenticity and accuracy of basic information. For maintenance cost and cycle data, specific costs for on-site, base, and return-to-factory maintenance are summarized. , , This data is retrieved through the operation and maintenance cost management system; simultaneously, the time spent on on-site, at the base, and returning to the factory for repairs is recorded. , , The maintenance work order system records data to form a complete dataset of maintenance costs and cycles. For reliability data collection, equipment satisfaction (E) and personnel qualification (P) are collected, queried through the equipment management system in conjunction with the personnel certification system; process reliability (R) and spare parts reliability are also collected. Data on process compliance, spare parts reliability, equipment adequacy, and maintenance personnel professional qualifications are obtained by combining process compliance records with spare parts quality inspection reports. Regarding the Arrows team's resource data, the base equipment load rate is statistically analyzed. Data is collected in real time by the base equipment monitoring system; spare parts inventory levels are statistically analyzed. By querying the inventory of the spare parts management system, a comprehensive collection of four types of core data is completed, and a data pool for decision-making is built.

[0016] Following the data acquisition steps outlined above, the data preprocessing stage commences. Data cleaning and standardization methods are employed to ensure data validity and consistency, preventing deviations in subsequent quantification calculations. Outliers in the raw data are removed using the 3σ criterion. The mean and standard deviation for each data category are calculated, and data exceeding the mean plus or minus three times the standard deviation are identified as outliers and removed from the dataset. Data processed in this stage contains no outliers and requires no further removal. For potentially missing values, a similarity-based interpolation method is used. Complete data with similar LRU usage scenarios and launch counts are found, and the mean of these similar data is used as the supplementary value for the missing value. Again, data processed in this stage contains no missing values ​​and requires no interpolation.

[0017] Next, the data format was standardized. The unit standardization method was adopted to convert all maintenance cost data into RMB 10,000 as the unit of measurement, and all time-related data into hours as the unit of measurement. For ratio data, the decimal retention rules were used to uniformly retain two decimal places, so that data of different types and units were formed into a unified format to meet the requirements of subsequent quantitative calculations.

[0018] Through the above steps, the collection and preprocessing of reusable rocket LRU full life cycle data were completed, and structured data with clear dimensions, accurate values, and unified format was finally obtained, providing a reliable data foundation for subsequent association library calls and four-dimensional indicator quantification.

[0019] After preprocessing the full lifecycle data, the loss pattern, maintenance candidate scheme and key constraint threshold of the LRU are matched by querying the association library built based on K-means clustering and FP-Growth algorithm, according to the cumulative number of launches N and real-time performance parameters.

[0020] Optionally, after completing the full lifecycle data preprocessing, a correlation database is first constructed. The automatic construction of the correlation database is based on data mining technology. The specific process is as follows: First, the full lifecycle data of historical LRUs is collected, and feature vectors for each LRU instance are constructed. Then, the K-means clustering algorithm is executed on the feature vector set, and the optimal number of clusters is determined based on the silhouette coefficient, automatically classifying the LRU status into low, medium, and high multi-level loss patterns. Subsequently, for all historical maintenance work orders under each loss pattern cluster, the FP-Growth frequent pattern mining algorithm is used to filter out maintenance action-successful result combinations with support greater than 10% and confidence greater than 85%, forming a candidate maintenance solution library for the corresponding loss pattern. Next, based on the successful samples of maintenance solutions within each loss pattern cluster, the lower bound of the 95% confidence interval between the cycle matching degree and the guarantee capability index is calculated and set as a key constraint threshold. Finally, the multi-level loss patterns, the candidate maintenance solution library, and the key constraint threshold are mapped according to their corresponding relationships to complete the generation of the correlation database.

[0021] The four-dimensional indicator system is used to quantitatively score the candidate maintenance solutions, resulting in a four-dimensional quantitative score.

[0022] In one embodiment of this application, for each candidate maintenance solution, the core parameter data of the corresponding four-dimensional indicators are first extracted using a parameter extraction method. Specific values ​​of various parameters corresponding to on-site maintenance base repairs and return-to-factory repairs are retrieved from channels such as the operation and maintenance cost management system and the maintenance work order system to ensure that parameters such as cost, time consumption, equipment satisfaction, and base equipment load rate for each candidate solution are complete and accurate, providing data support for subsequent scoring.

[0023] Next, the scores for each candidate scheme across the four indicators were calculated sequentially using formula substitution. When calculating the economic efficiency score, the maintenance costs of each scheme were substituted into the corresponding formula to first determine the ratio of the target scheme's cost to the baseline return-to-factory cost. This ratio was then subtracted from 1 and multiplied by 100 to obtain a standardized score. When calculating the cycle matching score, the maintenance time of each scheme and the rocket's planned reuse interval were substituted into the formula, and the ratio was calculated using the same logic before determining the score. When calculating the support capability score, sub-parameters such as equipment adequacy and personnel qualifications were multiplied by their corresponding weights, and the weighted results were summed to obtain the final score. When calculating the resource utilization score, the resource utilization index was first calculated based on the base equipment load rate and spare parts inventory, and then substituted into the formula to obtain the standardized score.

[0024] Afterwards, the validity of all scores is confirmed, the calculation process of each indicator is checked one by one to ensure accuracy, the parameters are correctly substituted, and each score is within the standardized range of 0 to 100. If any abnormal scores are found, they are reviewed and adjusted in a timely manner to ensure the accuracy of the single indicator scoring results.

[0025] Finally, the scores of the four aspects of each maintenance candidate solution—economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate—are sorted and collected in sequence, and the scores of the four indicators corresponding to each solution are clarified, so as to obtain the complete four-dimensional quantitative scoring results of the maintenance candidate solutions.

[0026] By using a recursive least squares online learning model with a forgetting factor, the weights of the four-dimensional indicators are dynamically allocated according to the cumulative number of launches N. Combined with the forced constraint judgment, the four-dimensional quantitative score is weighted and the optimal maintenance plan is selected.

[0027] Specifically, firstly, dynamic weighting coefficients are defined for four indicators: economy, cycle matching degree, support capability, and resource utilization. These weighting coefficients are dynamically allocated based on the cumulative number of LRU launches to adapt to maintenance needs under different usage frequencies. Next, a mandatory constraint judgment factor is defined. This factor is a binary factor, set based on the cycle matching degree threshold, and is used to screen solutions that meet the basic maintenance cycle requirements. Finally, a weighted formula is called to integrate the dynamic weighting coefficients and the mandatory constraint judgment factor into the calculation process, performing a weighted calculation on the acquired four-dimensional quantitative scores. The calculation result is used to select the LRU maintenance solution with the best suitability.

[0028] Based on a time-series database, the NSGA-II algorithm is used periodically to perform multi-objective optimization of the decision rules, and the Kappa coefficient is used to verify the stability of the optimization results. Once the verification is passed, the decision logic is updated.

[0029] Specifically, firstly, a time-series database is constructed by storing actual maintenance data. Then, based on this database, basic parameters are calibrated, and cost and time estimation formulas are updated. Dynamic weights are adjusted by calculating the Pearson correlation coefficient, and the loss interval threshold and decision rules are optimized. Subsequently, with economic efficiency, cycle matching degree, and reliability as optimization objectives, the NSGA-II algorithm is used to search for the Pareto optimal solution set of the decision rules. Combining the actual costs and times in the time-series database, robust regression is used to refit the cost and time estimation formulas. Finally, new decision rules are constructed using the Pareto optimal solution set and the updated estimation formulas. Stability testing is performed on a validation set using the Kappa coefficient of the decision results. Once the validation is successful, the decision logic is updated.

[0030] Furthermore, the method provided in this application embodiment includes: Collect historical LRU lifecycle data to construct a feature vector set for each LRU instance; perform K-means clustering on the feature vector set, determine the optimal number of clusters based on the silhouette coefficient, and automatically classify the LRU status into low, medium, and high loss modes to generate multi-level loss modes; for all historical maintenance work orders under each loss mode cluster in the multi-level loss modes, use the FP-Growth frequent pattern mining algorithm to find maintenance action-success result combinations with support greater than 10% and confidence greater than 85%, forming a candidate maintenance solution library for each loss mode; based on the successful samples of maintenance solutions within each loss mode cluster, calculate the lower bound of the 95% confidence interval between the cycle matching degree and the guarantee capability index, and set it as a key constraint threshold; map the multi-level loss modes, the candidate maintenance solution library, and the key constraint threshold according to the corresponding relationship to generate the association library.

[0031] Specifically, the first step is to collect historical LRU lifecycle data, extracting key information from the rocket operation and maintenance management system, component sensor records, and past maintenance work order files. This information includes the cumulative number of LRU launches, as well as key performance parameters such as vibration spectrum entropy, valve response time standard deviation, and thermal protection layer thickness attenuation rate. The vibration spectrum entropy is obtained by analyzing the vibration signals collected by sensors and calculating the entropy value of the spectrum distribution after Fourier transform; the valve response time standard deviation is a dispersion index calculated statistically after multiple tests of the valve's time from receiving a command to completing an action; the thermal protection layer thickness attenuation rate is obtained by comparing the initial thickness of the LRU with the measured thickness after use, calculating the difference between the two as a percentage of the initial thickness. This data is then organized in a unified format to construct a feature vector set for each LRU instance, for example, X = [N, F1, F2, ..., Fk], where N is the cumulative number of launches and F1~Fk are key performance parameters, ensuring that the vector comprehensively reflects the LRU's usage status and performance.

[0032] Next, the K-means clustering algorithm is executed on the constructed feature vector set. First, several feature vectors are randomly selected as initial cluster centers. Then, the Euclidean distance from each feature vector to each cluster center is calculated, and each vector is assigned to the nearest cluster. The cluster centers are then recalculated based on the vectors within each cluster, and the steps of distance calculation, vector assignment, and center update are repeated until the cluster centers no longer change significantly or the preset number of iterations is reached. Afterward, silhouette coefficients are calculated for different numbers of clusters, typically from 2 to 5. The silhouette coefficient measures the compactness and separation of clusters. The silhouette coefficient of each sample is equal to the difference between its average distance to other samples in the same cluster and its average distance to the nearest dissimilar sample, divided by the larger of the two. The cluster with the largest average silhouette coefficient is selected as the optimal number of clusters. This automatically divides the LRU state into three loss modes: low, medium, and high. Each mode corresponds to a set of LRUs with similar performance degradation and failure probability.

[0033] For all historical maintenance work orders under each loss pattern cluster, the FP-Growth frequent pattern mining algorithm is employed. First, the maintenance work orders are preprocessed, breaking each work order down into two parts: maintenance actions and maintenance results. Maintenance actions include specific operations such as component replacement, parameter calibration, cleaning and maintenance, and sealing checks. Maintenance results are categorized into successful reuse and unsuccessful reuse, forming a structured transaction set. Then, an FP-tree is constructed, sorting the maintenance actions in the transaction set by frequency and inserting them into the tree to form nodes and paths. The FP-tree is then traversed to mine frequent itemsets. Maintenance action-successful result combinations with support greater than 10% and confidence greater than 85% are selected. Support represents the proportion of this combination appearing in the transaction set of the same loss pattern, and confidence represents the proportion of successful reuse after executing the maintenance action. These qualified combinations together constitute the candidate maintenance solution library for the corresponding loss pattern.

[0034] Subsequently, based on successful maintenance schemes within each loss pattern cluster, a normal distribution confidence interval calculation method was used to determine the key constraint thresholds. Cycle matching degree and support capability data for successful samples were collected. Cycle matching degree is the ratio of maintenance time to the rocket's planned reuse interval, and support capability is the weighted sum of equipment satisfaction, personnel qualifications, process reliability, and spare parts reliability calculated according to preset weights. After performing a normality test on these data, a 95% confidence interval was calculated. This interval reflects the statistical distribution range of the data, and the lower bound of the interval was taken as the key constraint threshold. This threshold setting ensures that, in the vast majority of cases, maintenance schemes can meet cycle and reliability requirements, avoiding excessively high or low constraint standards due to extreme samples.

[0035] Finally, the low, medium, and high multi-level loss modes, their corresponding candidate maintenance scheme libraries, and the cycle matching degree and key constraint thresholds for each mode are structurally organized according to a one-to-one correspondence. This related information is then organized using database storage technology to form a quickly searchable association library. Those skilled in the art can then quickly match the corresponding loss mode, available maintenance candidate schemes, and necessary constraints by simply inputting the cumulative number of LRU launches and key performance parameters from the association library.

[0036] Through a series of steps including data aggregation, cluster analysis, frequent pattern mining, confidence interval calculation, and association mapping, the association library is automatically constructed, enabling it to accurately adapt to different wear states of LRUs and providing a reliable data-driven basis for maintenance decisions of reusable rocket LRUs.

[0037] Furthermore, the method provided in this application embodiment includes: The four-dimensional indicators include economic efficiency, cycle matching degree, guarantee capacity, and resource utilization rate.

[0038] Optionally, economic indicators are expressed using formulas. = Calculation and acquisition, where For return-to-factory costs, Corresponding to on-site repair costs Base maintenance costs Return-to-factory repair costs In the calculation, first find the ratio of the target repair solution cost to the baseline return-to-factory cost, then subtract the ratio from 1 and multiply by 100 to obtain a standardized score for economic efficiency. The higher the score, the higher the cost-effectiveness of the repair solution.

[0039] The period matching index is expressed by the formula. = Calculation and acquisition, where For the planned reuse interval of the rocket, Corresponding to on-site repair time Base maintenance time Time spent on returning to the factory for repair In the calculation, the ratio of maintenance time to the planned rocket reuse interval is first calculated, then the ratio is subtracted from 1 and multiplied by 100 to obtain the standardized score of cycle matching. The higher the score, the better the maintenance cycle matches the rocket reuse rhythm.

[0040] The guarantee capacity indicator is expressed by the formula. The values ​​were calculated, where α=0.2, β=0.2, γ=0.3, δ=0.3, and the core parameters E represent equipment satisfaction, P represents personnel qualifications, and R represents process reliability. For spare parts reliability, the sub-parameters are weighted and summed according to their corresponding weights to obtain a standardized score of support capability. The higher the score, the more sufficient the comprehensive support conditions of the maintenance plan.

[0041] Resource utilization rate index is expressed by formula = Calculation and acquisition, where Where θ=0.6, demand threshold=20, core parameters For the base equipment load rate, This represents the spare parts inventory level. The resource utilization index is calculated first during the calculation. Then subtract the index from 1 and multiply by 100 to get the standardized score of resource utilization. The lower the score, the less resource the maintenance plan uses.

[0042] Furthermore, the method provided in this application embodiment includes: Define dynamic weight coefficients , ,in, , , , These correspond to the weights of economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate, respectively; mandatory constraint judgment is defined. The mandatory constraint determination The periodic matching degree threshold is set as a binary factor; based on the dynamic weight coefficient. The aforementioned mandatory constraint determination The weighting formula is called to weight the four-dimensional quantitative score calculation and output the optimal maintenance plan.

[0043] Specifically, firstly, dynamic weight coefficients are defined using a recursive least squares online learning model with a forgetting factor. Dynamic weighting coefficient The allocation of (N) is achieved through an online learning model, specifically including: State awareness: The model input is a real-time state vector S(t)=[N(t),D(t)], where D(t) is the Mahalanobis distance between the current LRU performance parameter and its cluster center, used to quantify its degree of anomaly relative to the same type of LRU.

[0044] Utility feedback: The supervision signal for model learning comes from the ex-post utility evaluation U(t) of historical decisions. The utility function is defined as: Where C_actual is the actual cost, T_delay is the periodic delay, and R_success is the binary result of successful reuse after repair.

[0045] Online learning algorithm: A recursive least squares algorithm with an exponential forgetting factor (FFRLS) is used to update the parameter θ of the weight prediction model W=f(S(t);θ) in real time. The forgetting factor λ is set to 0.95, enabling the model to gradually forget old data and quickly adapt to changes in rocket reuse strategies, thereby achieving adaptive and personalized allocation of weights W_i(N).

[0046] Model inputs: cumulative number of launches N, and the standardized Euclidean distance D between the real-time performance parameters and the cluster centers, which are used as quantitative indicators of the degree of loss.

[0047] Model output: Optimal weight vector of four-dimensional indicators [ , , , Among them, economic efficiency Periodic matching degree , support capability Resource utilization rate .

[0048] Training mechanism: Based on the actual comprehensive utility value of historical decisions (defined as: To optimize the objective, a recursive least squares method with a forgetting factor is used to update the parameters of the weight allocation model W=f(N,D) online. This forgetting factor ensures that the model can quickly adapt to changes in rocket reuse strategies.

[0049] Next, the threshold binary decision method is used to define the mandatory constraint decision. First, determine the acceptable threshold for cycle matching, i.e. ≤24 hours ≥80%, when the cycle matching score of the maintenance plan meets this threshold. A value of 1 indicates that the solution passes the mandatory constraint; otherwise, ... A value of 0 indicates that the solution is excluded. For example, a cycle matching score of ≥60 is considered qualified, and this standard is used to judge each maintenance candidate solution, thus implementing the setting of binary factors.

[0050] Subsequently, a weighted summation method based on dynamic weight coefficients was adopted. and mandatory constraint determination Perform weighted calculations. For each maintenance candidate solution, multiply its four-dimensional quantitative score by the corresponding dynamic weight coefficient, and then sum the products; simultaneously, if the solution's... =0, then this option is directly excluded; if If the value is 1, then calculate its weighted total score and final decision value, and finally select the scheme with the highest final decision value as the optimal maintenance scheme.

[0051] By using the interval weight allocation method, the threshold binary judgment method, and the weighted summation method, the process of dynamic weight allocation, forced constraint judgment, and weighted calculation was completed, and the optimal maintenance scheme that adapts to the cumulative number of LRU launches and passes the forced constraint of period matching degree was finally output.

[0052] Furthermore, the method provided in this application embodiment includes: The expression for the weighting formula is: ;in, For weighted scores, according to The value represents a quantitative score of economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate. For quantitative scoring of economic efficiency, A quantitative score for the degree of periodic matching. To ensure the quantitative scoring of capabilities, A quantitative score for resource utilization.

[0053] Specifically, firstly, a four-dimensional quantitative score is obtained for each repair candidate solution. ,in Quantitative scoring of economic efficiency Quantitative scoring for periodic matching To ensure quantitative scoring of abilities, Resource utilization is quantified and scored using standardized scores obtained through the previous four-dimensional index quantification process.

[0054] Next, the dynamic weighting coefficient is determined based on the cumulative number of launches N. ,in , , , These correspond to the weights of economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate, respectively. For example, when N≤3 is the low-loss range, the weight of economic efficiency is moderately increased, the weight of cycle matching degree is moderate, and the weights of guarantee capability and resource utilization rate are relatively reduced; when N≥11 is the high-loss range, the weight of cycle matching degree is significantly increased, the weight of guarantee capability is increased, and the weights of economic efficiency and resource utilization rate are appropriately reduced, thereby achieving dynamic allocation of weights according to N.

[0055] Then, using the weighted product summation method, each... With the corresponding Multiply them, then add the four products together, that is, use the formula... Calculate the weighted score for each maintenance candidate. Then, for all cases that pass the mandatory constraint judgment, i.e., the periodic matching degree meets the threshold, For each of the maintenance candidate solutions with a value of 1, calculate its value one by one. .

[0056] By combining the weighted product summation method and the score comparison method with dynamic weight coefficients and four-dimensional quantitative scoring, the optimal maintenance scheme adapted to the cumulative number of LRU launches is output.

[0057] Furthermore, the method provided in this application embodiment includes: After applying the weighting formula to the four-dimensional quantitative score calculation, the result is combined with the mandatory constraint judgment. Determine the final decision value The expression is as follows: .

[0058] Specifically, first, calculate the weighted score for each maintenance candidate solution. Through formula Implementation. Among them, The weights of the four-dimensional indicators are dynamically allocated based on the cumulative number of launches N. This is a four-dimensional quantitative score for each scheme, calculated using a weighted summation method.

[0059] Next, determine the mandatory constraint criteria. A threshold-based determination method is used, based on a passing threshold for the cycle matching degree. If the cycle matching degree of the maintenance plan meets the threshold, Assign a value of 1; if not satisfied, It is assigned a value of 0, thus setting it as a binary factor.

[0060] Then, calculate the final decision value. Through formula Implementation. Multiplication is used to determine the mandatory constraints. With weighted score Multiply, if =0, then =0, this option is excluded; if =1, then = .

[0061] Finally, a numerical comparison method is used to determine the final decision value for all maintenance candidate solutions. Compare and select The largest possible solution is the optimal maintenance solution.

[0062] By employing weighted summation, threshold determination, multiplication, and numerical comparison methods, combined with dynamic weighting coefficients and mandatory constraint determination, a reusable rocket LRU maintenance scheme that meets the cycle matching requirements and achieves the best overall score was output.

[0063] Furthermore, the method provided in this application embodiment includes: The process involves storing actual maintenance data to construct a time-series database; calibrating basic parameters and updating cost and time estimation formulas based on the time-series database; adjusting dynamic weights by calculating the Pearson correlation coefficient based on the time-series database; optimizing loss interval thresholds and decision rules based on the time-series database and performing stability verification; using economic efficiency, cycle matching degree, and guarantee capability as optimization objectives, searching for Pareto optimal solutions using a second-generation non-dominated sorting genetic algorithm; refitting cost and time estimation formulas using the actual costs and times in the time-series database through robust regression; constructing updated decision rules based on the Pareto optimal solution set and the updated cost and time estimation formulas, and performing stability testing on a validation set by testing the Kappa coefficient of the decision results.

[0064] In one embodiment, after the decision is executed, a comprehensive collection of core data from the actual maintenance process is performed. This data includes maintenance level, actual maintenance cost, actual maintenance time, equipment satisfaction, base equipment load rate, and the cumulative number of launches for the corresponding LRU. The timestamp of each data item is also recorded. Following the construction specifications of a time-series database, core maintenance data are set as indicator fields, cumulative launches as category labels, and timestamps as index fields. The standardized data is stored in the database through batch writing, and data compression is enabled to reduce storage resource consumption. This ultimately constructs an LRU maintenance time-series database containing complete time-series dimensions, providing comprehensive and orderly data support for subsequent optimization work.

[0065] Based on the constructed time-series database, the RANSAC algorithm in robust regression is used to calibrate the basic parameters and update the cost and time estimation formulas. The cost estimation formula is C_L=f(N,L), and the time estimation formula is T_L=f(N,L), where C_L represents the estimated maintenance cost, T_L represents the estimated maintenance time, N is the cumulative number of LRU launches, and L is the maintenance level. The actual cost C_actual and actual time T_actual corresponding to different maintenance levels and cumulative launches are extracted from the time-series database. These actual data are substituted into the initial estimation formulas. The RANSAC algorithm automatically identifies and removes outliers in the data to avoid interference with formula fitting. Through refitting, the optimized cost and time estimation formulas are obtained, completing the calibration of the basic parameters and improving the accuracy of the formulas in predicting actual maintenance costs and times.

[0066] Next, leveraging the data resources in the time-series database, dynamic weights are adjusted by calculating the Pearson correlation coefficient. First, historical data for four indicators—economic efficiency, cycle matching degree, support capability, and resource utilization rate—are selected from the time-series database. Simultaneously, corresponding maintenance effectiveness data is extracted, with the post-maintenance LRU reuse success rate and failure rate as the core evaluation criteria. The Pearson correlation coefficient measures the degree of linear correlation between two variables. By calculating the Pearson correlation coefficients between the four indicators and the maintenance effectiveness data, the influence strength of each indicator on maintenance effectiveness can be clearly defined. Based on the calculated correlation coefficients, the dynamic weight coefficients are adjusted. Indicators with a strong positive correlation to maintenance effectiveness and high correlation coefficients are appropriately weighted, while indicators with low correlation coefficients and small impact are weighted, making the weight allocation more aligned with the actual needs of maintenance effectiveness.

[0067] Then, based on historical data in the time-series database, the loss interval threshold and decision rules were optimized, and stability verification was conducted. Actual loss data of LRUs under different cumulative launch counts were extracted from the time-series database. By statistically analyzing the failure frequency of LRUs within each cumulative launch count interval and the adaptability of existing maintenance schemes, the boundary thresholds for low, medium, and high loss intervals were iteratively adjusted to ensure that the loss mode division more accurately reflects the actual situation. Simultaneously, the execution effect of historical decisions was combined to revise relevant decision rules such as the mandatory constraint judgment criteria and weighted calculation rules. Stability verification employed a one-way ANOVA method. Multiple sets of historical data from different time periods were randomly selected from the time-series database as test samples. The optimized loss interval threshold and decision rules were applied to these samples for multiple decision tests. By analyzing the dispersion of the test results, if there were no significant differences in the results, it was determined that the optimized loss interval threshold and decision rules possessed stability.

[0068] After accumulating maintenance data for every six months, a multi-objective optimization cycle for the decision-making rules is automatically initiated. With economy, cycle matching, and support capability as the core optimization objectives, the second-generation non-dominated sorting genetic algorithm, NSGA-II, is used to optimize and search for the decision-making rules. These rules include key elements such as weighted model parameters and constraint thresholds. Through non-dominated sorting and crowding calculation, the NSGA-II algorithm can find trade-offs among multiple objectives and search for a Pareto-optimal solution set. This solution set contains multiple optimal candidate strategies, allowing decision-makers to select the most suitable strategy from the solution set based on the key requirements of the current rocket launch mission.

[0069] The actual cost (C_actual) and actual time (T_actual) data from the time-series database are retrieved again, and the RANSAC algorithm from robust regression is used to refit the cost and time prediction formulas. This process leverages the algorithm's resistance to outlier interference and, combined with the latest accumulated actual data, performs a second optimization on the cost prediction formula C_L=f(N,L) and the time prediction formula T_L=f(N,L), further improving the prediction accuracy of the formulas and ensuring that they accurately reflect the relationship between the cumulative number of launches (N) and maintenance level (L) and the actual cost and time.

[0070] Finally, based on the Pareto optimal solution set obtained by the NSGA-II algorithm and the cost and time estimation formulas refitted through robust regression, an updated decision rule was constructed. To verify the stability of the new decision rule, 100 historical state snapshots at different time points were randomly selected from the time-series database. These snapshots contained complete state information such as the cumulative number of LRU launches and performance parameters at the corresponding time points. Maintenance decisions were made on these snapshots using both the old and new decision rules, and the consistency index, i.e., the Kappa coefficient, of the two decision results was calculated. When the Kappa coefficient was greater than 0.8, it indicated that the new decision rule had high consistency with the original stable rule, and the optimization results had good stability. At this point, the new decision rule was deployed online, completing the update of the decision logic.

[0071] Through a series of interconnected steps, including time-series database construction, basic parameter calibration, dynamic weight adjustment, loss range and decision rule optimization, multi-objective algorithm search and stability verification, the decision rules are continuously optimized, ensuring that the decision logic can be dynamically adjusted according to actual operation and maintenance data, thereby improving the accuracy and stability of LRU repair-level decisions for reusable rockets.

[0072] In summary, the reusable rocket LRU repair level dynamic decision-making method provided in this application has the following technical effects: This application collects full lifecycle data of reusable rocket LRUs, preprocesses it, calls related libraries to match loss patterns and other information, quantifies and scores it using four-dimensional indicators, dynamically allocates weights based on cumulative launch counts, and calculates weighted scores using mandatory constraints. After selecting the optimal solution, it iterative optimization is achieved using a time-series database, enabling precise decision-making on maintenance plans, improving maintenance efficiency and reliability, and optimizing operating costs. This achieves the technical effect of dynamically adapting and continuously optimizing maintenance decisions based on equipment usage frequency, improving the accuracy and adaptability of decisions, and balancing maintenance efficiency and operating cost optimization.

[0073] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a reusable rocket LRU repair level dynamic decision-making system, the system comprising: The full life cycle data acquisition module 1 is used to collect full life cycle data of reusable rocket LRUs, including basic data, maintenance cost and cycle data, reliability data and rocket team resource data.

[0074] The association library construction module 2 is used to preprocess the full life cycle data and then match the loss mode, maintenance candidate scheme and key constraint threshold of LRU according to the cumulative number of launches N and real-time performance parameters by querying the association library constructed based on K-means clustering and FP-Growth algorithm.

[0075] The four-dimensional quantitative scoring acquisition module 3 performs quantitative scoring on the maintenance candidate scheme based on a four-dimensional indicator system to obtain a four-dimensional quantitative score.

[0076] The optimal maintenance solution acquisition module 4 is used to select the optimal maintenance solution by dynamically allocating the weights of the four-dimensional indicators according to the cumulative number of launches N based on the recursive least squares online learning model with forgetting factor, and combining the forced constraint judgment to perform weighted calculation on the four-dimensional quantitative score calculation.

[0077] The decision logic update module 5 is based on a time-series database. It periodically uses the NSGA-II algorithm to perform multi-objective optimization on the decision rules and uses the Kappa coefficient to verify the stability of the optimization results. After the verification is passed, the decision logic is updated.

[0078] Furthermore, the associated library construction module 2 is used to perform the following steps: Collect historical LRU lifecycle data to construct a feature vector set for each LRU instance; perform K-means clustering on the feature vector set, determine the optimal number of clusters based on the silhouette coefficient, and automatically classify the LRU status into low, medium, and high loss modes to generate multi-level loss modes; for all historical maintenance work orders under each loss mode cluster in the multi-level loss modes, use the FP-Growth frequent pattern mining algorithm to find maintenance action-success result combinations with support greater than 10% and confidence greater than 85%, forming a candidate maintenance solution library for each loss mode; based on the successful samples of maintenance solutions within each loss mode cluster, calculate the lower bound of the 95% confidence interval between the cycle matching degree and the guarantee capability index, and set it as a key constraint threshold; map the multi-level loss modes, the candidate maintenance solution library, and the key constraint threshold according to the corresponding relationship to generate the association library.

[0079] Furthermore, the four-dimensional quantitative scoring acquisition module 3 is used to perform the following steps: The four-dimensional indicators include economic efficiency, cycle matching degree, guarantee capacity, and resource utilization rate.

[0080] Furthermore, the optimal maintenance solution acquisition module 4 is used to perform the following steps: Define dynamic weight coefficients , ,in, , , , These correspond to the weights of economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate, respectively; mandatory constraint judgment is defined. The mandatory constraint determination The periodic matching degree threshold is set as a binary factor; based on the dynamic weight coefficient. The aforementioned mandatory constraint determination The weighting formula is called to weight the four-dimensional quantitative score calculation and output the optimal maintenance plan.

[0081] Furthermore, the optimal maintenance solution acquisition module 4 is used to perform the following steps: The expression for the weighting formula is: ;in, For weighted scores, according to The value represents a quantitative score of economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate. For quantitative scoring of economic efficiency, A quantitative score for the degree of periodic matching. To ensure the quantitative scoring of capabilities, A quantitative score for resource utilization.

[0082] Furthermore, the optimal maintenance solution acquisition module 4 is used to perform the following steps: After applying the weighting formula to the four-dimensional quantitative score calculation, the result is combined with the mandatory constraint judgment. Determine the final decision value The expression is as follows: .

[0083] Furthermore, the decision logic update module 5 is used to perform the following steps: The process involves storing actual maintenance data to construct a time-series database; calibrating basic parameters and updating cost and time estimation formulas based on the time-series database; adjusting dynamic weights by calculating the Pearson correlation coefficient based on the time-series database; optimizing loss interval thresholds and decision rules based on the time-series database and performing stability verification; using economic efficiency, cycle matching degree, and guarantee capability as optimization objectives, searching for Pareto optimal solutions using a second-generation non-dominated sorting genetic algorithm; refitting cost and time estimation formulas using the actual costs and times in the time-series database through robust regression; constructing updated decision rules based on the Pareto optimal solution set and the updated cost and time estimation formulas, and performing stability testing on a validation set by testing the Kappa coefficient of the decision results.

[0084] The reusable rocket LRU repair level dynamic decision system provided in this embodiment of the invention can execute the reusable rocket LRU repair level dynamic decision method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0085] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0086] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A dynamic decision-making method for repair level of reusable rocket LRUs, characterized in that, include: Collect full lifecycle data for reusable rocket LRUs, including basic data, maintenance cost and cycle data, reliability data, and rocket fleet resource data; After preprocessing the full lifecycle data, the loss pattern, maintenance candidate scheme and key constraint threshold of the LRU are matched by querying the association library built based on K-means clustering and FP-Growth algorithm, according to the cumulative number of launches N and real-time performance parameters. The four-dimensional indicator system is used to quantitatively score the candidate maintenance solutions, resulting in a four-dimensional quantitative score. By employing a recursive least squares online learning model with a forgetting factor, the weights of four-dimensional indicators are dynamically allocated based on the cumulative number of launches N. Combined with mandatory constraint judgment, the four-dimensional quantitative score is weighted and calculated to select the optimal maintenance scheme. The model input is a real-time state vector S(t) = [N(t), D(t)], where D(t) is the Mahalanobis distance between the current LRU performance parameter and its cluster center, used to quantify its degree of anomaly relative to similar LRUs. The supervision signal for model learning comes from the ex-post utility evaluation U(t) of historical decisions, and the utility function is defined as: ， C_actual represents the actual cost, T_delay represents the cycle delay, R_success represents the binary result of successful reuse after repair, and T0 represents the planned reuse interval of the rocket. A recursive least squares algorithm with an exponential forgetting factor is employed to update the parameter θ of the weight prediction model W=f(S(t);θ) in real time. The forgetting factor λ is set to 0.95, enabling the model to gradually forget outdated data and quickly adapt to changes in rocket reuse strategies, thereby achieving dynamic weight coefficients. Adaptive and personalized allocation When N≤3 is the low-loss range, the weight of economic efficiency is moderately increased, the weight of cycle matching degree is moderate, and the weights of guarantee capability and resource utilization rate are relatively reduced. When N≥11 is the high-loss range, the weight of cycle matching degree is significantly increased, the weight of guarantee capability is increased, and the weights of economic efficiency and resource utilization rate are appropriately reduced. Based on a time-series database, the NSGA-II algorithm is used periodically to perform multi-objective optimization of the decision rules, and the Kappa coefficient is used to verify the stability of the optimization results. Once the verification is passed, the decision logic is updated.

2. The dynamic decision-making method for repair level of reusable rocket LRU as described in claim 1, characterized in that, The association database is automatically constructed based on data mining, and specifically includes the following steps: Collect all lifecycle data of historical LRU instances and construct a feature vector set for each LRU instance; The K-means clustering algorithm is executed on the feature vector set to determine the optimal number of clusters based on the silhouette coefficient, and the LRU state is automatically divided into low, medium and high loss modes to generate multi-level loss modes. For all historical maintenance work orders under each loss pattern cluster in the multi-level loss pattern, the FP-Growth frequent pattern mining algorithm is used to find maintenance action-success result combinations with support greater than 10% and confidence greater than 85%, forming a candidate maintenance solution library for each loss pattern. Based on the successful samples of maintenance schemes within each loss pattern cluster, the lower bound of the 95% confidence interval of the cycle matching degree and the guarantee capability index is calculated and set as the key constraint threshold. The multi-level loss patterns, the candidate maintenance scheme library, and the key constraint thresholds are mapped according to their corresponding relationships to generate the association library.

3. The dynamic decision-making method for repair level of reusable rocket LRU as described in claim 1, characterized in that, The four-dimensional indicators include economic efficiency, cycle matching degree, guarantee capacity, and resource utilization rate.

4. The dynamic decision-making method for repair level of reusable rocket LRU as described in claim 3, characterized in that, Based on the cumulative number of launches N, the weights of the four-dimensional indicators are dynamically allocated. Combined with mandatory constraint judgments, the four-dimensional quantitative scoring is weighted and calculated to select the optimal maintenance plan, including: Define dynamic weight coefficients , ,in, , , , These correspond to the weights of economic efficiency, cycle matching degree, guarantee capacity, and resource utilization rate, respectively. Define mandatory constraint judgment The mandatory constraint determination Set the binary factor according to the periodic matching degree threshold; Based on the dynamic weighting coefficient The aforementioned mandatory constraint determination The weighting formula is called to weight the four-dimensional quantitative score calculation and output the optimal maintenance plan.

5. The dynamic decision-making method for repair level of reusable rocket LRU as described in claim 4, characterized in that, The expression for the weighting formula is: ； in, For weighted scores, according to The value represents a quantitative score of economic efficiency, cycle matching degree, guarantee capability, and resource utilization rate. For quantitative scoring of economic efficiency, A quantitative score for the degree of periodic matching. To ensure the quantitative scoring of capabilities, A quantitative score for resource utilization.

6. The dynamic decision-making method for repair level of reusable rocket LRU as described in claim 5, characterized in that, Based on the dynamic weighting coefficient The aforementioned mandatory constraint determination The four-dimensional quantitative score is weighted using a weighting formula, and the optimal repair solution is output, including: After applying the weighting formula to the four-dimensional quantitative score calculation, the result is combined with the mandatory constraint judgment. Determine the final decision value The expression is as follows: 。 7. The dynamic decision-making method for repair level of reusable rocket LRU as described in claim 1, characterized in that, Based on a time-series database, the NSGA-II algorithm is periodically used to perform multi-objective optimization of the decision rules, and the Kappa coefficient is used to verify the stability of the optimization results. After the verification is passed, the decision logic is updated, including: Store actual maintenance data and build a time-series database; Based on the calibration parameters of the time series database, update the cost and time estimation formulas; Based on the time-series database, dynamic weights are adjusted by calculating the Pearson correlation coefficient; Based on the aforementioned time-series database, the loss interval threshold and decision rules are optimized, and stability is verified. With economic efficiency, cycle matching degree, and guarantee capability as optimization objectives, a second-generation non-dominated sorting genetic algorithm is used to search for decision rules and find the Pareto optimal solution set. Using the actual costs and time consumption data from the time series database, robust regression was employed to refit the cost and time consumption prediction formulas. An update decision rule is constructed using the Pareto optimal solution set and the updated results of the cost and time estimation formulas, and stability is tested on the validation set by testing the Kappa coefficient of the decision results.

8. A reusable rocket LRU repair-level dynamic decision-making system, characterized in that, The system is used to implement the dynamic decision-making method for repair level of reusable rocket LRU as described in any one of claims 1-7, the system comprising: The full lifecycle data acquisition module is used to collect full lifecycle data of reusable rocket LRUs, including basic data, maintenance cost and cycle data, reliability data, and rocket team resource data. The association library construction module is used to preprocess the full life cycle data and then match the LRU's loss mode, maintenance candidate scheme and key constraint thresholds by querying the association library built based on K-means clustering and FP-Growth algorithm, according to the cumulative number of launches N and real-time performance parameters. The four-dimensional quantitative scoring module performs quantitative scoring on the maintenance candidate solutions based on a four-dimensional indicator system to obtain a four-dimensional quantitative score. The optimal maintenance scheme acquisition module is used to select the optimal maintenance scheme by dynamically allocating the weights of the four-dimensional indicators according to the cumulative number of launches N based on the recursive least squares online learning model with forgetting factor, combined with the forced constraint judgment, and performing weighted calculation on the four-dimensional quantitative score calculation. The decision logic update module, based on a time-series database, periodically uses the NSGA-II algorithm to perform multi-objective optimization of the decision rules, and uses the Kappa coefficient to verify the stability of the optimization results. Once the verification is passed, the decision logic is updated.

Images