A simulator maintenance task scheduling method and system based on fault prediction and capability profiling, an electronic device, and a storage medium

Abstract

The application relates to the technical field of computers and discloses a simulator maintenance task scheduling method and system based on fault prediction and capability profiling, an electronic device and a storage medium. The method comprises the following steps: a subsystem fault propagation model is constructed; future fault probability, time and root cause subsystems are predicted according to real-time operation parameters; a pre-task vector containing a root cause skill label is generated, and a pre-scheduling scheme is cached; if a match is found when an actual fault occurs, the match is quickly output, otherwise, a regular scheduling process is entered; candidate personnel are screened based on a dynamic capability profiling vector and a task vector; a multi-objective optimization model is established to solve a Pareto optimal solution set and output a scheme. The application realizes a paradigm leap from passive response to active prediction, significantly improves response efficiency and resource utilization, and reduces operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a method, system, electronic device, and storage medium for scheduling simulator maintenance tasks based on fault prediction and capability profiling. Background Technology

[0002] Currently, simulator maintenance tasks are typically assigned using a single-site independent management model, with each site's maintenance personnel and technical experts managing tasks independently. Task assignments are mostly done manually by the local supervisor. When a complex fault occurs at a site and the local personnel are insufficient, it is necessary to request support from experts at other sites, which is a cumbersome process with slow response and high costs.

[0003] Meanwhile, the assessment of maintenance personnel's capabilities relies heavily on manual records or subjective impressions, lacking dynamic, quantitative, and multi-dimensional capability profiles, making it difficult to achieve accurate matching of personnel with fault tasks. In addition, scheduling decisions only consider "who can repair," without taking into account multiple factors such as the urgency of the task, the current workload of personnel, travel costs, and skill freshness. This results in senior experts being overworked while ordinary personnel lack training, leading to low overall response efficiency and resource utilization.

[0004] Crucially, existing scheduling methods are generally reactive, "post-event remediation" responses. The system only initiates personnel screening and dispatch processes after a fault actually occurs and an alarm is generated, making it impossible to proactively predict risks using the massive amounts of time-series data from simulator operating parameters. Even if maintenance personnel detect abnormal signs in the system, there is a lack of a technical loop to translate "predictive information" into "scheduling instructions." This results in an incompressible response vacuum in scheduling calculations and personnel coordination during emergency faults, and makes it difficult to identify specialized maintenance experts with specific capabilities in advance to pinpoint the root cause of the fault. Therefore, how to predict risks and pre-generate scheduling plans using propagation models before a fault occurs to compress response time has become a critical technical bottleneck that urgently needs to be addressed in this field.

[0005] Therefore, this application provides a simulator maintenance task scheduling method based on fault prediction and capability profiling to solve the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this invention is to provide a simulator maintenance task scheduling method, system, electronic device and storage medium based on fault prediction and capability profiling, in order to solve the technical problems in the prior art caused by information silos among personnel from multiple bases, ambiguous capability assessment and extensive scheduling decisions, resulting in untimely fault response, misallocation of human resources, high operation and maintenance costs and difficulty in meeting the timeliness requirements for simulator fault repair.

[0007] To address the aforementioned technical problems, this invention provides a simulator maintenance task scheduling method based on fault prediction and capability profiling, comprising:

[0008] Acquire multi-source data of maintenance personnel, and construct a dynamic capability profile vector of the maintenance personnel based on the multi-source data;

[0009] Obtain task information for simulator maintenance tasks, and organize the task information into a task vector;

[0010] Perform fault prediction, construct a subsystem fault propagation model, predict the fault probability and occurrence time within the future time window based on real-time operating parameters, and generate a pre-task vector containing skill tags of fault subsystem and root factor system when the predicted probability exceeds the threshold, and cache the corresponding scheduling scheme. When the actual maintenance task occurs, if it matches the cached scheduling scheme, it will be output directly.

[0011] Otherwise, based on the skill-related and location-related information in the task vector, as well as the preset skill threshold, personnel who meet the skill threshold in the skill dimension and the task requirements in the location dimension are selected from the dynamic ability profile vector to form a candidate set.

[0012] Each candidate in the candidate set and the matching scheme of the maintenance task are treated as an individual. A multi-objective optimization model is constructed with the goal of minimizing the comprehensive objective function value and includes constraints. Based on the optimization objective and constraints, a multi-objective optimization algorithm is used to iteratively solve the problem to obtain the Pareto optimal solution set.

[0013] The individual with the smallest comprehensive objective function value is selected from the Pareto optimal solution set as the optimal scheduling scheme output, and alternative schemes are also output to guide the execution of the simulator maintenance task.

[0014] In some specific embodiments, acquiring multi-source data of maintenance personnel and constructing a dynamic capability profile vector of the maintenance personnel based on the multi-source data further includes:

[0015] Calculate the skill vector score of the maintenance personnel for each machine model and each subsystem;

[0016] The skill freshness is calculated based on the number of times a specific subsystem fault is handled within a preset time period, and the skill vector score is attenuated and corrected according to the skill freshness to obtain an effective skill score.

[0017] Calculate the experience value and load index of the maintenance personnel;

[0018] The location information and cost coefficient of the maintenance personnel are obtained, and the dynamic capability profile vector is composed of the effective skill score, the experience value, the skill freshness, the load index, the location information and the cost coefficient.

[0019] In some specific embodiments, calculating the maintenance personnel's skill vector score for each machine model and each subsystem further includes:

[0020] The training data of the maintenance personnel for specific machine models and subsystems is obtained. A training score is obtained by weighting the number of training sessions and exam scores. The historical records of the maintenance personnel's handling of faults of the machine model and subsystem are obtained. Based on the number of handling sessions and fault complexity, the data is normalized with the maximum value of the corresponding indicators of all maintenance personnel in the system for the machine model and subsystem to obtain a maintenance experience score. The success rate of the maintenance personnel in handling faults of the machine model and subsystem is obtained to obtain a fault repair success rate score.

[0021] The training score, maintenance experience score, and fault repair success rate score are weighted and combined to obtain the maintenance personnel's skill vector score for a specific model and subsystem, with the maintenance experience score having the largest weight.

[0022] The skill freshness is determined based on the number of times the maintenance personnel handle specific subsystem faults within a preset time period. When the number of handlings reaches a preset threshold, the skill freshness is at its maximum value. The skill freshness decreases by a preset step size for each decrease in the number of handlings. When the skill freshness is lower than a preset percentage, the skill vector score is attenuated and corrected to obtain an effective skill score.

[0023] The empirical value is obtained by weighting and summing the maintenance personnel's years of experience, the total number of faults handled, and the average fault complexity.

[0024] In some specific embodiments, acquiring task information for simulator maintenance tasks and organizing the task information into a task vector further includes:

[0025] Obtain the fault code, fault type, and fault model from the fault record, and determine the task type as one of fault repair, planned maintenance, software upgrade, or periodic qualification test;

[0026] Based on the scope of the fault's impact and the simulator's usage plan, the urgency level is determined as urgent, general, or non-urgent.

[0027] The faulty machine model and the subsystems involved in the fault are mapped to the required skill tags, and the skill thresholds are dynamically adjusted according to the urgency of the task and the complexity of the fault.

[0028] Retrieve historical repair records from the historical maintenance database that are identical to the current task's model, subsystem, and fault type. Determine the estimated time for the task based on the statistical values ​​of historical repair time. Organize the task type, urgency level, required skill tags, estimated time, and task location into the task vector.

[0029] In some specific embodiments, each candidate in the candidate set and the matching scheme of the maintenance task are treated as an individual, and a multi-objective optimization model is constructed with the goal of minimizing the comprehensive objective function value. The comprehensive objective function includes at least a weighted combination of three dimensions: response time, comprehensive cost, and skill utilization rate, and further includes:

[0030] The overall objective function is defined as a weighted combination of response time, overall cost, and skill utilization rate, where response time and overall cost are positive terms, and skill utilization rate is a negative term.

[0031] The response time is determined, which consists of local response time and cross-base travel time, wherein the cross-base travel time is determined based on the mode of transportation between the candidate's current base and the mission location;

[0032] The comprehensive cost is determined, which consists of personnel salary cost, travel cost and opportunity cost. The personnel salary cost is determined based on the candidate's hourly salary and the duration of the task. The travel cost is determined based on the candidate's average daily cost of cross-base business trips and the number of business trip days. The opportunity cost is the cost incurred due to the delay of other pending tasks caused by scheduling the candidate.

[0033] The constraints include that the candidate's current task load does not exceed a preset high load threshold, the response time of an emergency task does not exceed a specified fault response time limit, the candidate's skill level is not lower than the skill threshold, the skill level is determined based on the skill information in the dynamic capability profile vector, and the candidate's location dimension meets the requirements for local or remote execution of the task.

[0034] In some specific embodiments, a multi-objective optimization algorithm is used to iteratively solve the problem and obtain a Pareto optimal solution set, further including:

[0035] The multi-objective optimization model is iteratively optimized using a non-dominated sorting genetic algorithm with an elitist strategy. During the iteration process, non-dominated sorting is performed based on the dominance relationship of individuals, and crowding is calculated to maintain the diversity of the solution set. Offspring population is generated through selection, crossover, and mutation operations. The optimal non-dominated solution in the parent generation is retained in the offspring generation. When the preset iteration termination condition is reached, the Pareto optimal solution set is output.

[0036] In some specific embodiments, the individual with the smallest comprehensive objective function value is selected from the Pareto optimal solution set as the optimal scheduling scheme output, and alternative schemes are output to guide the execution of the simulator maintenance task, further including:

[0037] The optimal scheduling scheme and alternative schemes are used to allocate maintenance personnel to perform the simulator maintenance tasks;

[0038] During task execution, when the rescheduling trigger condition is met, the Pareto optimal solution set and solution output are recalculated using the latest status of all personnel at the trigger time and all unfinished tasks as input. The rescheduling trigger condition includes at least one of the following: the selected personnel refuse the task, the original task is canceled, a new higher priority task appears, the urgency or complexity of the current task changes, or the on-duty status of the selected personnel or the current task load changes abruptly.

[0039] Based on the same concept, the present invention also provides a simulator maintenance task scheduling system based on fault prediction and capability profiling, comprising:

[0040] The dynamic capability profile vector construction module is configured to acquire multi-source data of maintenance personnel and construct a dynamic capability profile vector of the maintenance personnel based on the multi-source data.

[0041] The task vector organization module is configured to acquire task information of simulator maintenance tasks and organize the task information into task vectors.

[0042] The fault prediction module is configured to perform fault prediction, build a subsystem fault propagation model, predict the probability of faults and the occurrence time within a future time window based on real-time operating parameters, and generate a pre-task vector containing skill tags of fault subsystems and root factor systems when the predicted probability exceeds a threshold, and cache the corresponding scheduling scheme. When an actual maintenance task occurs, if it matches the cached scheduling scheme, it will be output directly.

[0043] The candidate set construction module is configured to otherwise select personnel from the dynamic ability profile vector whose skill dimension meets the skill threshold and whose position dimension meets the task requirements based on the skill-related information and location-related information in the task vector, as well as a preset skill threshold, to form a candidate set.

[0044] The optimization scheduling module is configured to treat each candidate in the candidate set and the matching scheme of the maintenance task as an individual, construct a multi-objective optimization model with the goal of minimizing the comprehensive objective function value, and include constraints. Based on the optimization objective and constraints, a multi-objective optimization algorithm is used to iteratively solve the problem to obtain the Pareto optimal solution set.

[0045] The output execution module is configured to select the individual with the smallest comprehensive objective function value from the Pareto optimal solution set as the optimal scheduling scheme output, and output alternative schemes to guide the execution of the simulator maintenance task.

[0046] Based on the same concept, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of a simulator maintenance task scheduling method based on fault prediction and capability profiling.

[0047] Based on the same concept, the present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a simulator maintenance task scheduling method based on fault prediction and capability profiling.

[0048] Compared with existing technologies, its advantages are as follows:

[0049] This invention discloses a simulator maintenance task scheduling method, system, electronic device, and storage medium based on fault prediction and capability profiling, which improves fault response efficiency: it breaks down base barriers and automatically matches through algorithms, shortening cross-base response time from the traditional "days" to "hours", effectively reducing unplanned simulator downtime, ensuring that simulator fault repair can meet the prescribed timeliness requirements, and enabling maintenance tasks to be executed in a timely manner.

[0050] Achieve precise allocation of human resources: By creating multi-dimensional quantitative capability profiles, shift from "assigning tasks based on impressions" to "assigning tasks based on data," ensuring that the most suitable personnel perform the most appropriate maintenance tasks, thereby improving the first-time repair rate of complex faults and reliably guaranteeing the quality of simulator maintenance.

[0051] Effectively reduce overall operation and maintenance costs: Through global optimization algorithms, the optimal balance point is found between travel costs, personnel load, and task urgency, reducing unnecessary long-distance travel for experts, balancing personnel load, and enabling limited maintenance resources to serve simulator operation more economically.

[0052] Promote balanced development of personnel capabilities: In non-urgent tasks, a skills enhancement scheduling strategy can be set up to assign tasks with learning potential to less experienced personnel and match them with expert guidance, so as to systematically improve the overall technical level of the team, while dynamically maintaining the freshness of skills and ensuring that the maintenance team has sufficient practical ability in the long term.

[0053] Enhance the digitalization of management decisions: Provide managers with visualized, data-driven scheduling decision support, making operation and maintenance management more scientific and transparent, reducing task delays, and thus ensuring the normal progress of simulator training tasks. Attached Figure Description

[0054] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0055] Figure 1 This is a flowchart illustrating some specific embodiments of the simulator maintenance task scheduling method based on fault prediction and capability profiling of the present invention;

[0056] Figure 2 This is a schematic diagram of the structure of a simulator maintenance task scheduling system based on fault prediction and capability profiling in some specific embodiments of the present invention;

[0057] Figure 3 This is a schematic diagram of the structure of an electronic device according to some specific embodiments of the present invention;

[0058] In the diagram, 710 is the processor; 720 is the memory; 730 is the input device; and 740 is the output device. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0061] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0062] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0063] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0064] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0065] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0066] Reference Figure 1 A simulator maintenance task scheduling method based on fault prediction and capability profiling includes:

[0067] S101, Obtain multi-source data of maintenance personnel, and construct a dynamic capability profile vector of the maintenance personnel based on the multi-source data;

[0068] S102, Obtain task information for simulator maintenance tasks, and organize the task information into a task vector;

[0069] S103, perform fault prediction, construct a subsystem fault propagation model, predict the fault probability and occurrence time within the future time window based on real-time operating parameters, when the predicted probability exceeds the threshold, generate a pre-task vector containing skill tags of the fault subsystem and root factor system, and cache the corresponding scheduling scheme. When the actual maintenance task occurs, if it matches the cached scheduling scheme, it will be output directly.

[0070] S104, otherwise, based on the skill-related information and location-related information in the task vector, and the preset skill threshold, select personnel from the dynamic ability profile vector whose skill dimension meets the skill threshold and whose location dimension meets the task requirements, and form a candidate personnel set.

[0071] S105, taking each candidate in the candidate set and the matching scheme of the maintenance task as an individual, constructing a multi-objective optimization model with the optimization objective of minimizing the comprehensive objective function value, and including constraints, and using a multi-objective optimization algorithm to iteratively solve based on the optimization objective and constraints to obtain the Pareto optimal solution set;

[0072] S106, Select the individual with the smallest comprehensive objective function value from the Pareto optimal solution set as the optimal scheduling scheme output, and output alternative schemes to guide the execution of the simulator maintenance task.

[0073] Specifically, in this embodiment of the invention, multi-source data of maintenance personnel is collected in real time from the maintenance management system, simulator data acquisition system, personnel attendance and business trip system, training management system, and fault record system of each base. This includes historical maintenance records, training records, fault handling results, attendance status, business trip information, and salary cost data. Based on the collected multi-source data, a dynamic capability profile vector for each maintenance personnel is constructed. Specifically, for each maintenance personnel, their skill vector score for each model and subsystem is calculated: by obtaining the personnel's training data for a specific model and subsystem, a training score is obtained by weighting the number of training sessions and exam scores; the personnel's historical records of handling faults of that model and subsystem are obtained, and the maximum value of the corresponding indicator for all maintenance personnel under that model and subsystem is normalized based on the number of times the fault was handled and the complexity coefficient of each fault, to obtain a maintenance experience score; the personnel's success rate in handling faults of that model and subsystem is obtained to obtain a fault repair success rate score; the training score, maintenance experience score, and fault repair success rate score are weighted and combined, with the maintenance experience score having the largest weight, to obtain the skill vector score. Skill freshness is determined based on the number of times a person handles specific subsystem faults within a preset time period: the skill freshness reaches its maximum when the number of handlings reaches a preset threshold, and decreases by a preset step size for each decrease in the number of handlings; when the skill freshness is below a preset percentage, the skill vector score is attenuated to obtain an effective skill score. Simultaneously, the person's years of experience, the total number of faults handled, and the average fault complexity are weighted and summed to obtain an experience value; a real-time load index is calculated based on the number of tasks currently assigned to the person, the estimated remaining time for each task, and the total time scheduled for future shifts; the person's current location, on-duty status, business travel availability, and business travel willingness coefficient are obtained to constitute location information; the person's hourly wage and average daily cost of cross-location business travel are obtained to constitute a cost coefficient. The effective skill score, experience value, skill freshness, load index, location information, and cost coefficient together constitute the dynamic capability profile vector of the maintenance personnel. For simulator maintenance tasks, the fault code, fault type, and fault model are obtained from the fault records, and the task type is determined to be one of the following: fault repair, planned maintenance, software upgrade, or periodic qualification testing. Based on the scope of the fault's impact and the simulator's usage plan, the urgency level is automatically determined to be three levels: urgent, general, or non-urgent. The fault model and the subsystems involved in the fault are mapped to the required skill tags, and the skill thresholds are dynamically adjusted according to the urgency of the task and the complexity of the fault. Historical repair records with the same model, subsystem, and fault type as the current task are retrieved from the historical maintenance database, and the estimated time of the task is determined based on the statistical values ​​of historical repair time. The task type, urgency level, required skill tags, estimated time, and task location are organized into a task vector.Based on the required skill tags and preset skill thresholds in the task vector, personnel with effective skill scores not lower than the skill threshold are selected from the dynamic capability profile vectors of all maintenance personnel; at the same time, based on the location of the task and the personnel's location information, personnel whose on-duty status is on duty and whose business travel status meets the task requirements are selected, and the above selection results constitute the candidate personnel set. A multi-objective optimization model is constructed by treating each candidate in the candidate set and the matching scheme with the current maintenance task as an individual. This model includes constraints and aims to minimize the comprehensive objective function value. The comprehensive objective function value is defined as a weighted combination of response time, comprehensive cost, and skill utilization rate, where response time and comprehensive cost are positive terms, and skill utilization rate is a negative term. Response time consists of local response time and cross-base travel time, with cross-base travel time determined based on the mode of transportation between the candidate's current base and the task location. Comprehensive cost consists of personnel salary cost, travel cost, and opportunity cost. Personnel salary cost is determined based on the candidate's hourly salary and task duration, travel cost is determined based on the candidate's average daily cross-base travel cost and number of travel days, and opportunity cost is the cost incurred due to delays in other pending tasks caused by scheduling the candidate. Constraints include that the candidate's load index does not exceed a preset high load threshold, the response time of emergency tasks does not exceed the specified fault response time limit, the candidate's effective skill score is not lower than the skill threshold, and the candidate's location information meets the requirements for local or remote task execution. Based on the optimization objective and constraints, an iterative optimization using a non-dominated sorting genetic algorithm with an elitist strategy is employed. During iteration, non-dominated sorting is performed based on the dominance relationships of individuals, and crowding is calculated to maintain solution diversity. Offspring populations are generated through selection, crossover, and mutation operations, retaining the optimal non-dominated solutions from the parent generation. When a preset iteration termination condition is met, a Pareto optimal solution set is output. The individual with the smallest comprehensive objective function value is selected from the Pareto optimal solution set as the optimal scheduling scheme, while several alternative schemes are output for assigning maintenance personnel to perform the simulator maintenance task. During task execution, when the rescheduling trigger condition is met, the latest status of all personnel at the trigger time and all unfinished tasks are used as input to re-execute the above optimization scheduling steps and output execution steps.

[0074] In some applications, multi-source data of maintenance personnel is acquired, and a dynamic capability profile vector of the maintenance personnel is constructed based on the multi-source data. This includes calculating the skill vector score of the maintenance personnel for each machine model and each subsystem; calculating the skill freshness based on the number of times a specific subsystem fault is handled within a preset time period, and attenuating and correcting the skill vector score according to the skill freshness to obtain an effective skill score; calculating the experience value and load index of the maintenance personnel; and acquiring the location information and cost coefficient of the maintenance personnel. The dynamic capability profile vector is composed of the effective skill score, the experience value, the skill freshness, the load index, the location information, and the cost coefficient.

[0075] Understandably, for each machine model and subsystem, training data for the maintenance personnel is obtained, and a training score is obtained by weighting the number of training sessions and exam scores. The personnel's historical records of handling faults for that machine model and subsystem are obtained, and the repair experience score is obtained by normalizing the data based on the number of times the fault was handled and the complexity coefficient of each fault, compared with the maximum value of the corresponding indicators for all maintenance personnel under that machine model and subsystem. The success rate of the personnel in handling faults for that machine model and subsystem is obtained, resulting in a fault repair success rate score. The training score, repair experience score, and fault repair success rate score are then weighted and combined, with the repair experience score having the highest weight, to obtain the personnel's skill vector score for a specific machine model and subsystem. Skill freshness is determined based on the number of times the maintenance personnel handle faults in a specific subsystem within a preset time period: skill freshness reaches its maximum value when the number of handling sessions reaches a preset threshold, and decreases by a preset step size for each decrease in the number of handling sessions; when the skill freshness is lower than a preset percentage, the skill vector score is attenuated to obtain an effective skill score. The maintenance personnel's years of experience, total number of faults handled, and average fault complexity are weighted and summed to obtain an experience value. A load index is calculated based on the number of tasks currently assigned to the personnel, the estimated remaining time for each task, and the total scheduled time for future shifts. The personnel's current location, on-duty status, travel availability, and travel willingness coefficient are obtained to form location information. The personnel's hourly wage and average daily cost of cross-location travel are obtained to form a cost coefficient. The effective skill score, experience value, skill freshness, load index, location information, and cost coefficient together constitute the maintenance personnel's dynamic capability profile vector.

[0076] In some applications, the skill vector score of the maintenance personnel for each machine model and each subsystem is calculated. This includes obtaining the training data of the maintenance personnel for specific machine models and subsystems, obtaining a training score based on the number of training sessions and exam scores, obtaining the historical records of the maintenance personnel's handling of faults in the specified machine model and subsystem, normalizing the data based on the number of handling sessions and fault complexity with the maximum value of the corresponding indicators of all maintenance personnel in the system for the specified machine model and subsystem to obtain a maintenance experience score, and obtaining the success rate of the maintenance personnel in handling faults in the specified machine model and subsystem to obtain a fault repair success rate score. The training score, the maintenance experience score, and the... The success rate scores for fault repair are weighted and combined to obtain the skill vector score of the maintenance personnel on a specific model and subsystem, with maintenance experience score having the largest weight. The skill freshness is determined based on the number of times the maintenance personnel handle faults in a specific subsystem within a preset time period. The skill freshness is at its maximum when the number of handling reaches a preset threshold. The skill freshness decreases by a preset step size for each decrease in the number of handling. When the skill freshness is lower than a preset percentage, the skill vector score is attenuated and corrected to obtain an effective skill score. The years of service of the maintenance personnel, the total number of faults handled, and the average fault complexity are weighted and summed to obtain an experience value.

[0077] Understandably, for a specific machine model and subsystem, training data for the maintenance personnel is obtained, including the number of training sessions and exam scores. A training score is derived based on a weighted average of the training sessions and exam scores. Historical repair records of the maintenance personnel handling faults of that machine model and subsystem are obtained. Based on the number of historical repairs and the complexity coefficient of each fault, normalization is performed using the maximum value of the corresponding indicator for all maintenance personnel under that machine model and subsystem, thus obtaining a repair experience score. The success rate of the maintenance personnel handling faults of that machine model and subsystem, i.e., the ratio of successful repairs to total repairs, is obtained as the fault repair success rate score. The training score, repair experience score, and fault repair success rate score are weighted and combined, with the repair experience score having the largest weight, to obtain the maintenance personnel's skill vector score for that specific machine model and subsystem. Based on this, a skill freshness is determined according to the number of times the maintenance personnel handle faults in the specific subsystem within a preset time period: when the number of repairs reaches a preset threshold, the skill freshness is maximized; for each decrease in the number of repairs, the skill freshness decreases by a preset step size. When the skill freshness rate falls below a preset percentage, the skill vector score undergoes attenuation correction to obtain an effective skill score. An empirical value is obtained by multiplying the maintenance personnel's years of experience, the total number of faults handled, and the average fault complexity by their respective weighting coefficients and then summing the results.

[0078] In some applications, task information for simulator maintenance tasks is acquired and organized into task vectors. This includes retrieving fault codes, fault types, and fault models from fault records; determining the task type as one of fault repair, planned maintenance, software upgrade, or periodic qualification testing; classifying the urgency level as urgent, general, or non-urgent based on the fault's impact scope and simulator usage plan; mapping the fault model to the subsystems involved in the fault as required skill tags, and dynamically adjusting skill thresholds based on the task's urgency and fault complexity; retrieving historical repair records from a historical maintenance database that match the current task's model, subsystem, and fault type; determining the estimated time for the task based on statistical values ​​of historical repair times; and organizing the task type, urgency level, required skill tags, estimated time, and task location into the task vector.

[0079] Understandably, when a simulator maintenance task arises, the fault code, fault type, and faulty model are retrieved from the fault log. Based on this, the task type is determined to be one of the following: fault repair, planned maintenance, software upgrade, or periodic qualification testing. The urgency of the task is automatically classified into three levels: urgent, general, or non-urgent, based on the scope of the fault's impact (e.g., whether it causes training interruption) and the simulator's usage plan (e.g., scheduling for the next few hours). The faulty model and the subsystems involved are mapped to skill tags required to complete the task, and skill thresholds are dynamically adjusted based on the task's urgency and fault complexity: for urgent tasks or complex faults, skill thresholds are appropriately increased to ensure that the personnel undertaking the task possess a sufficiently high skill level; for non-urgent tasks or simple faults, skill thresholds can be appropriately decreased to broaden the pool of candidates. Historical repair records with the same model, subsystem, and fault type as the current task are retrieved from the historical maintenance database, and the estimated time for the task is determined based on the statistical values ​​(e.g., arithmetic mean or median) of the actual repair time in these historical records. The five pieces of information—task type, urgency, required skill tags, estimated time, and task location—are organized into a structured task vector for subsequent filtering and optimized scheduling.

[0080] In some applications, model building and fault prediction steps are also included:

[0081] Construct a time sequence graph of fault propagation in the simulator subsystem, where nodes represent subsystems, the weights of directed edges represent the causal strength of historical fault propagation, and each edge is associated with a time delay probability distribution.

[0082] The system collects the operating parameters of each subsystem in real time, calculates the real-time health of each subsystem, and predicts the failure probability and estimated occurrence time of each subsystem within a future time window based on the fault propagation time sequence diagram and dynamic Bayesian network.

[0083] Understandably, a fault propagation time sequence diagram of the simulator's subsystems is constructed. Simulators typically comprise multiple subsystems, including motion systems, visual systems, avionics systems, air conditioning systems, power systems, and data acquisition systems. Each subsystem is defined as a node in the diagram. To characterize the causal direction and intensity of fault propagation between subsystems, fault maintenance records for at least six months are extracted from a historical maintenance database. Each record includes a timestamp of the fault occurrence, fault code, root factor system annotations by maintenance personnel, and a time sequence of abnormal states in each subsystem during fault handling. A time-series association rule mining method is employed, with a time window set at 30 minutes, to statistically analyze the fault propagation within each subsystem. Within the time window following the occurrence of the abnormal state, the subsystem The conditional probability of an abnormal state occurring is determined. When this conditional probability exceeds a preset confidence threshold of 0.6, it is preliminarily determined that an abnormal state exists. arrive Potential causal edges were identified. To further verify the directionality of causality and rule out spurious correlations, Granger causality tests were performed on continuous sensor data from the subsystems: for the time series data of the two subsystems, a vector autoregression model was used to test for potential causal edges. Whether the historical value has increased significantly The prediction accuracy, if the test The statistic corresponds to If the value is less than 0.05, then accept. yes Granger causality. For new models or subsystem combinations lacking sufficient historical data, maintenance experts are allowed to input subjective confidence levels in the form of a priori causal matrix, where each element takes a value between 0 and 1, representing the degree to which the expert believes a causal relationship exists. Finally, the weight of each directed edge is calculated using a Bayesian fusion formula: ,in The number of historical records increases dynamically as the amount of historical data accumulates; for example, when the number of historical records is less than 50... When the value is 0.3, it is greater than 200. A value of 0.9 is used to gradually transition from expert experience to data-driven approaches. Each directed edge is also associated with a time delay probability distribution. The time difference between the occurrence of anomalies in the source subsystem and the target subsystem is collected from all historical propagation events satisfying causal relationships. A Gaussian mixture model is used to fit this time difference data, with two Gaussian components corresponding to short-term propagation (within minutes) and long-term propagation (tens of minutes to half an hour), respectively. The mean, variance, and mixing coefficient of each component are estimated using the expectation-maximization algorithm, thus obtaining the probability density function of the time delay. .

[0084] Based on the completed fault propagation sequence diagram, real-time operational parameters of each subsystem are collected to calculate real-time health. Each subsystem is equipped with several sensors or acquires key parameters through data acquisition interfaces. For example, the motion system collects vibration amplitude, motor current, and position feedback error; the visual system collects frame rate, rendering latency, and image frame drop rate; and the avionics system collects bus communication error rate and power supply voltage fluctuations. A statistical model of the parameters under health conditions is pre-established for each subsystem: sensor data for at least one week during normal operation of the subsystem is collected, and the mean vector of each feature dimension is calculated. Covariance Matrix During real-time execution, the feature vector extracted at the current moment is... Substituting into the Mahalanobis distance formula, calculate its distance from the health status distribution. Distance is converted into health score using an exponential function. The health score ranges from 0 to 1. A value closer to 1 indicates a healthier operating status, while a value closer to 0 indicates a higher degree of abnormality. The health scores of all subsystems are updated every fixed sampling period of 1 minute, and the health score sequence is stored as time-series data.

[0085] To predict the failure probability within a future time window, a dynamic Bayesian network is used as the inference engine. The dynamic Bayesian network extends the aforementioned failure propagation time series graph into a two-time-slice Bayesian network, where each time slice contains the hidden state variables of all subsystems. The hidden state is defined as a binary variable indicating whether the subsystem will fail in the next time slice, while the observed variable is the calculated real-time health score. The transition probability in the network is jointly determined by the edge weights of the failure propagation graph and the time delay distribution: specifically, if there exists a transition probability from the subsystem... To subsystem Given the directed edges of the subsystem at the previous time step, the subsystem... The conditional probability of a failure occurring in the current time slice is proportional to the edge weight. Considering the time delay distribution, the propagation is discretized in time, for example, the delay distribution is discretized into probability masses of several time steps. A particle filter algorithm is used for online inference. Initially, 1000 particles are generated, each representing a possible configuration of the hidden state of all subsystems. Upon receiving a new health observation, the weight of each particle is calculated according to the observation model. The observation model is set so that when the hidden state is failure, the health follows a normal distribution with a mean of 0.2 and a variance of 0.1. Under normal conditions, health status follows a normal distribution with a mean of 0.9 and a variance of 0.05. Resampling is performed to eliminate low-weight particles. Based on the particle distribution, the probability of each subsystem failing within a future time window (e.g., 15 minutes, 30 minutes, 1 hour) is calculated; this represents the proportion of particles whose hidden states are failures. Through backpropagation inference, the root factor system can be traced from subsystems with high predicted failure probabilities. In the dynamic Bayesian network, maximum a posteriori estimation is used to find the hidden state sequence most likely to lead to the current observation sequence. Tracing back along the time direction, the starting node of the failure propagation path is found as the root factor system. A list of predicted failures is output, where each element contains the subsystem identifier predicted to fail, the estimated occurrence time (the median or expected value of the probability distribution of failure times in particle filtering), the predicted probability value, and the set of root factor systems obtained through inference.

[0086] When the predicted failure probability of any subsystem exceeds a preset threshold of 0.7 and the estimated remaining time (estimated occurrence time minus the current time) is greater than the minimum scheduling preparation time of 30 minutes, the pre-scheduling process is triggered. This threshold can be dynamically adjusted according to the importance of the simulator and the training plan. For example, the threshold can be lowered to 0.5 one week before the simulator inspection date stipulated by the Civil Aviation Administration to improve the sensitivity of the early warning. Through the above model construction and failure prediction steps, potential failure risks can be identified in advance, providing accurate time, location, and skill requirement information for the subsequent generation of pre-scheduling plans, thereby shortening the response delay after the actual failure occurs.

[0087] In some applications, scheme generation and switching steps are also included:

[0088] When the predicted failure probability of any subsystem exceeds a preset threshold and the estimated remaining time is greater than the minimum scheduling preparation time, a pre-task vector is generated. The pre-task vector includes at least the predicted failure subsystem and the root factor system skill tags obtained through reasoning, the task location, and the predictive maintenance type.

[0089] Based on the pre-task vector and the current maintenance personnel dynamic capability profile vector, a pre-scheduling scheme is generated and cached. The pre-scheduling scheme includes a set of candidate personnel and a subset of individuals from the Pareto optimal solution set.

[0090] When an actual failure occurs, if the failed subsystem matches the predicted node in the pre-scheduling scheme and the failure time is within the prediction window, the cached pre-scheduling scheme is quickly verified and output, or a lightweight re-optimization is performed before outputting.

[0091] Understandably, in the predictive pre-scheduling step, when the predicted failure probability of any subsystem exceeds a preset threshold and the estimated remaining time is greater than the minimum scheduling preparation time, the scheme generation and switching sub-step is immediately triggered. A pre-task vector is generated, and its data structure is fully compatible with the task vector triggered by real-time failures, ensuring seamless integration with subsequent scheduling processes. Specifically, the task type is set to predictive maintenance, with a coding value of 5, distinguishing it from code 1 for fault repair, code 2 for planned maintenance, code 3 for software upgrades, and code 4 for periodic qualification testing. The urgency level is determined by a combination of the predicted failure probability and the estimated remaining time: if the predicted failure probability is greater than 0.85 and the estimated remaining time is less than 1 hour, the urgency level is set to high alert, with a code value of 2; if the predicted failure probability is between 0.7 and 0.85 and the estimated remaining time is between 1 and 2 hours, it is set to normal alert, with a code value of 1.5. This code value falls between that of general tasks and non-urgent tasks, allowing the weight coefficients in the subsequent multi-objective optimization model to be adjusted accordingly, for example, appropriately increasing the weight of response time for high alert tasks. The required skill tags not only include the skill tags of the predicted failure subsystem itself, but also the skill tags of the root factor system set obtained through backpropagation inference. These root factor system skill tags are included in the skill requirement list of the pre-task vector, forming a skill tag set. For example, if the predicted failure subsystem is a visual system, and the inferred root factor system is a motion system, then the required skill tags for the pre-task vector are {visual system, motion system}. The estimated time consumption is also set based on the 90th percentile of historical repair times for similar faults, using a conservative estimate to ensure sufficient time margin for the pre-scheduling plan. The task location is directly taken from the code of the current simulator's base. After the pre-task vector is generated, it is used as input to call the multi-objective optimization scheduling process, but the optimization objective is slightly adjusted: since it is predictive maintenance, the weight coefficient of skill utilization in the comprehensive objective function is appropriately increased to encourage the allocation of pre-tasks to personnel with higher skill scores on the root factor system, so that personnel with root cause resolution capabilities can be put into standby status before the fault actually occurs. From the dynamic capability profile database of maintenance personnel in all bases across the country, a preliminary screening is performed based on the skill tag set in the pre-task vector to select personnel with effective skill scores not lower than the preset skill threshold and whose position dimension meets the task requirements, forming a candidate set. A non-dominated sorting genetic algorithm with an elitist strategy is used to solve the multi-objective optimization model to generate a Pareto optimal solution set. Unlike real-time fault scheduling, pre-scheduled solving allows for a larger computational time budget, such as setting the maximum number of iterations to 100 generations instead of the 50 generations in real-time scheduling, in order to obtain better solution set quality.The top few individuals in the Pareto optimal solution set, the top 5 individuals with the smallest combined objective function value, along with the corresponding candidate set, intermediate population information during the optimization process, and the hash value of the task vector, are stored in the pre-scheduling cache table. The cache table's key is composed of the predicted fault subsystem identifier, the prediction time window start stamp, and the simulator equipment number for fast subsequent indexing. The pre-scheduling plan does not actually assign tasks to personnel, nor does it modify personnel load indices or current task lists; it is only stored as a "frozen plan" in memory or cache. The pre-scheduling plan's generation timestamp and validity period are recorded, with the validity period set to the end of the prediction time window plus 30 minutes from the generation time.

[0092] When a simulator actually malfunctions and receives a fault code and fault occurrence timestamp from the fault recording system, the fault subsystem identifier is extracted. Using this fault subsystem identifier, the current timestamp, and the device number as a key, a query is performed in the pre-scheduling cache table to check for a matching pre-scheduling scheme. Matching conditions include: the fault subsystem is identical to the predicted node, and the actual fault time falls within the prediction time window of the pre-scheduling scheme; that is, the actual fault time is no earlier than the prediction start time and no later than the prediction end time plus an allowable deviation, such as 15 minutes. If a match is successful, the process proceeds to the rapid verification phase. The main task of the rapid verification phase is to check whether the current state of the candidate personnel set in the cache scheme has changed significantly compared to the time of pre-scheduling generation. Specific verification content includes: whether the load index of each candidate personnel is still no higher than the preset high load threshold of 80%, whether the on-duty status is 1 indicating they are still on duty, whether the travel availability status still meets the task requirements, whether the travel willingness coefficient has not significantly decreased by more than 0.2, and whether the personnel's current base has not changed across bases. The system checks for unconsidered new tasks that are better than those in the cached solution, such as urgent tasks with higher priority that appear after pre-scheduling is generated. However, this check is only a suggestion and does not forcibly reject the cached solution. If the status changes of all candidates are within the preset tolerance range, the optimal individual from the cached pre-scheduling solution is directly output as the scheduling solution. This output includes the suggested executor's identifier, estimated arrival time, estimated completion time, estimated total cost, and skill matching degree. Alternative solutions are also output for management decision-making. The total time from fault confirmation to solution output is usually less than 10 seconds, far lower than the average 90 seconds required for resolving.

[0093] If rapid verification reveals that some candidate personnel no longer meet the constraints—for example, the load index of a candidate in the original optimal solution has risen to 85%, exceeding the threshold, or the personnel are currently on a business trip and unable to take on new tasks—the cached solution will not be discarded directly. Instead, a lightweight re-optimization will be performed. This lightweight re-optimization uses a subset of individuals from the Pareto optimal solution set in the cached solution as seeds for the initial population, rather than randomly initializing the population. Specifically, the first 10 individuals from the Pareto optimal solution set generated during pre-scheduling are read from the cache and encoded as part of the initial population. Simultaneously, several new individuals are randomly added from the current candidate personnel set to maintain the population size. These new individuals are obtained from the current personnel states by re-executing the initial screening step. A non-dominated sorting genetic algorithm is used for iteration, but the maximum number of iterations is reduced from the usual 50 generations to 10 generations. The crossover probability remains at 0.8, and the mutation probability is reduced to 0.02 to minimize disturbance to existing high-quality solutions. The latest states of all personnel at the current moment and all unfinished tasks are then re-introduced into the optimization model as constraints. Since the initial population is already close to the optimal solution, lightweight re-optimization typically converges to a feasible corrective scheme in just 5 to 10 generations, with the total time kept under 30 seconds. The re-optimized optimal scheduling scheme is output, and the corresponding entry in the pre-scheduled cache table is updated synchronously, marking the old cache scheme as invalid.

[0094] After outputting the scheduling plan, the task order is pushed to the selected personnel's mobile terminal, and the predictive maintenance task is marked as dispatched in the task management system. If, during the waiting period after the pre-scheduling plan is generated and before the actual fault occurs, the predicted fault probability is detected to drop below the threshold, or the simulator operator performs preventative maintenance to eliminate the abnormal state, the corresponding pre-scheduling plan is deleted from the cache, and the false alarm event is recorded for subsequent updates to the threshold parameters of the prediction model. If no effective pre-scheduling plan is found when the actual fault occurs, for example, the faulty subsystem is inconsistent with the predicted node or the fault time exceeds the prediction window, the process reverts to the conventional task scheduling process described in claim 1, i.e., the complete optimization solution is executed again from the task information acquisition. Through the above plan generation and switching steps, a leap from passive response to proactive pre-scheduling and from complete recalculation to cache acceleration is achieved, improving the scheduling response efficiency of simulator maintenance tasks in emergency fault scenarios.

[0095] In some applications, each candidate in the candidate set and the matching scheme with the maintenance task are treated as an individual. A multi-objective optimization model is constructed with the goal of minimizing the comprehensive objective function value. The comprehensive objective function includes at least a weighted combination of three dimensions: response time, comprehensive cost, and skill utilization rate. Specifically, the comprehensive objective function is defined as a weighted combination of response time, comprehensive cost, and skill utilization rate, where response time and comprehensive cost are positive terms, and skill utilization rate is a negative term. The response time is determined, consisting of local response time and cross-base travel time, where cross-base travel time is determined based on the mode of transportation between the candidate's current base and the task location. The comprehensive objective function is then determined. The comprehensive cost consists of personnel salary costs, travel costs, and opportunity costs. Personnel salary costs are determined based on the candidate's hourly wage and task duration. Travel costs are determined based on the candidate's average daily cost of cross-base business trips and the number of days of business trip. Opportunity costs are the costs incurred due to delays in other pending tasks caused by scheduling the candidate. The constraints include: the candidate's current task load not exceeding a preset high-load threshold; the response time for emergency tasks not exceeding a specified fault response time limit; the candidate's skill level not lower than the skill threshold, the skill level determined based on skill information in the dynamic capability profile vector; and the candidate's location dimension meeting the requirements for local or remote task execution.

[0096] Understandably, the overall objective function is defined as a weighted combination of response time, overall cost, and skill utilization rate. Response time and overall cost are positive terms (i.e., the larger their values, the less favorable the optimization), while skill utilization rate is a negative term (i.e., the larger their value, the more favorable the optimization). Response time consists of two parts: local response time and cross-base travel time. Local response time is a fixed standard time, while cross-base travel time is determined based on the mode of transportation (e.g., high-speed rail, airplane, car) between the candidate's current base and the task location, and the required travel time. The overall cost consists of personnel salary costs, travel costs, and opportunity costs: Personnel salary costs are calculated based on the candidate's hourly wage and the total time spent performing the task (i.e., the sum of response time and estimated task duration); travel costs are calculated based on the candidate's average daily cost of cross-base travel and the number of travel days, with the number of travel days converted from the sum of round-trip travel time and estimated task duration to a full day; opportunity costs are the costs incurred due to the delay of other tasks originally assigned to the candidate caused by scheduling this task, specifically determined by multiplying the estimated total duration of the delayed tasks by the candidate's hourly wage. Skill utilization rate is calculated based on the proportion of the candidate's effective skill scores for the required machine type and subsystems for the current task among the highest-scoring candidates in the candidate pool. In addition, the model includes several constraints: the candidate's current task load (i.e., load index) must not exceed a preset high load threshold to avoid personnel overload; for emergency tasks, the response time must not exceed the specified fault response time limit; the candidate's skill level (i.e., effective skill score) must not be lower than the skill threshold set for the task, and this skill level is derived from the skill information in the dynamic capability profile vector; the candidate's location dimension (including whether they are on duty, whether they are available for business trips, and their willingness to travel) must meet the task's local or remote execution requirements. Based on the above optimization objectives and constraints, a multi-objective optimization algorithm is used to iteratively solve the problem and obtain the Pareto optimal solution set.

[0097] In some applications, a multi-objective optimization algorithm is used to iteratively solve the problem and obtain a Pareto optimal solution set. This includes using a non-dominated sorting genetic algorithm with an elitist strategy to iteratively optimize the multi-objective optimization model. During the iteration process, non-dominated sorting is performed based on the dominance relationship of individuals, and crowding is calculated to maintain the diversity of the solution set. Offspring populations are generated through selection, crossover, and mutation operations. The optimal non-dominated solution in the parent generation is retained in the offspring generation. When a preset iteration termination condition is reached, the Pareto optimal solution set is output.

[0098] Understandably, each candidate in the candidate set and its matching scheme with the maintenance task is encoded as an individual in the population, with each individual containing three objective values: response time, overall cost, and skill utilization. After initializing the population, the algorithm enters an iterative process: in each iteration, all individuals are non-dominated and sorted based on their dominance relationships, dividing the population into multiple non-dominated layers, with the first layer being the optimal non-dominated solution set. Simultaneously, the crowding degree of each individual within its non-dominated layer is calculated; higher crowding degree indicates a sparser distribution of individuals around that individual, and retaining individuals with high crowding degree helps maintain the diversity of the solution set. Parent individuals are selected from the current population through selection operations (e.g., roulette wheel selection). Crossover and mutation operations are performed on the selected parent individuals to generate the offspring population. Crossover is used to exchange some genes between two parent individuals to produce new individuals, and mutation is used to randomly perturb some genes of individuals to increase population diversity. An elite retention strategy is adopted, directly copying the optimal non-dominated solution from the parent population (i.e., individuals in the first non-dominated layer) to the offspring population to ensure the algorithm's convergence speed. When the number of iterations reaches the preset maximum number of iterations, or when the change in the comprehensive objective function value of the best individual in multiple consecutive generations is less than the preset threshold, the algorithm terminates the iteration and outputs the final non-dominated optimal solution set, i.e., the Pareto optimal solution set.

[0099] In some applications, the individual with the smallest comprehensive objective function value is selected from the Pareto optimal solution set as the optimal scheduling scheme output, and alternative schemes are output to guide the execution of the simulator maintenance task. This includes the optimal scheduling scheme and alternative schemes for allocating maintenance personnel to perform the simulator maintenance task. During task execution, when the rescheduling trigger condition is met, the Pareto optimal solution set and scheme output are re-solved using the latest status of all personnel at the trigger time and all unfinished tasks as input. The rescheduling trigger condition includes at least one of the following: the selected personnel refuse the task, the original task is canceled, a new higher priority task appears, the urgency or complexity of the current task changes, or the on-duty status of the selected personnel or the current task load changes abruptly.

[0100] Understandably, the optimal scheduling scheme is pushed to the management terminal. After confirmation by the manager, the task order is pushed to the selected personnel's mobile terminal, along with historical repair cases and related technical guidance resources to assist the personnel in efficiently completing the maintenance task. During task execution, the status of the selected personnel and the task are monitored in real time. When the rescheduling trigger condition is met, the latest status of all personnel at the trigger time (including location information, load index, on-duty status, business trip availability, etc.) and all incomplete tasks (including the current task and other pending tasks) are immediately used as input to re-execute the optimized scheduling steps and output execution steps. That is, the Pareto optimal solution set is re-solved and a new scheduling scheme is output, thereby achieving rapid response to dynamic changes. The rescheduling trigger condition includes at least one of the following situations: the selected personnel refuse to execute the assigned task; the original task is canceled for some reason; a new, higher-priority simulator maintenance task appears in the system; the urgency level of the current task (e.g., upgraded from general to urgent) or the complexity of the fault changes; the on-duty status of the selected personnel (e.g., sudden absence) or the current task load (e.g., receiving other urgent tasks causing the load to exceed the limit) changes abruptly.

[0101] The following describes another embodiment of the simulator maintenance task scheduling method based on fault prediction and capability profiling according to the present invention:

[0102] This embodiment includes:

[0103] Multi-dimensional capability profile construction:

[0104] Define each maintenance worker's profile as a six-dimensional vector:

[0105] ;

[0106] The quantification methods and core models for each dimension are as follows, ensuring the objectivity, dynamism, and accuracy of the profile:

[0107] (Skill Vector)

[0108] A two-dimensional matrix consisting of "aircraft type (e.g., A320, A330, A350, B737, B787, etc.)" and "subsystems (e.g., motion, vision, avionics, air conditioning, etc.)" is denoted as:

[0109] ;

[0110] in, Indicates the model classification code ( =1,2,..., , (Total number of models)

[0111] Classification and coding of simulator subsystems ( =1,2,..., , (Total number of subsystems)

[0112] This indicates that the person was in the Type 1 The proficiency rating on each subsystem ranges from 0 to 100 points, with higher scores indicating greater skill proficiency.

[0113] Skill vector scoring model:

[0114] This model is built on a weighted Bayesian algorithm, combining multi-source data such as personnel's historical maintenance records, training records, and fault handling results to achieve dynamic quantitative calculation of skill scores. The specific formula is as follows:

[0115] ;

[0116] in,

[0117] The training score (0-100 points) is calculated based on the number of times the personnel complete the relevant training for this model and subsystem, and the exam scores. The more training sessions and the higher the exam scores, the higher the score. For example, completing one theoretical training session earns 20 points, completing one practical training session earns 30 points, and an additional 10 points are awarded for an exam score of 90 or above, for a maximum score of 100 points.

[0118] The maintenance experience score (0-100 points) is calculated based on the number of times the personnel handled faults in this subsystem of this model, weighted by the complexity of the faults. The formula is as follows:

[0119] ;

[0120] in,

[0121] This is the first time in history that this person has been processed. Type 1 The total number of subsystem failures;

[0122] The first case handled for this person Complexity coefficients of faults in the same subsystem of the same type of aircraft (according to the ATA chapter classification, complex faults D=3, general faults D=2, and simple faults D=1).

[0123] The denominator is the total number of maintenance personnel within the system on the corresponding machine model. Subsystem Below The maximum value of the indicator is used to normalize the score to the 0-100 range.

[0124] The formula for calculating the success rate of fault repair (0-100 points) is as follows:

[0125] ;

[0126] in,

[0127] Process the first for this person Type 1 The number of successful repairs for each subsystem failure (the simulator can be put into normal use after the failure is repaired, meeting the relevant standards of the Civil Aviation Administration), if ,but ;

[0128] The weighting coefficients for training score, maintenance experience score, and fault repair success rate score should satisfy the following conditions. It can be adjusted according to actual business needs; the default value is... This means prioritizing repair experience and repair success rate while also considering training.

[0129] (Experience Value)

[0130] Based on their years of experience, the total number of faults handled, and their complexity, the quantitative range is 0-100 points, and the formula is as follows:

[0131] ;

[0132] Among them, 0.2, 0.4, and 0.4 are the weighting coefficients corresponding to years of service, cumulative number of faults handled, and average fault complexity, and the sum of the three is 1, which can be adjusted according to business needs;

[0133] Years of service ( ≤30, and if it exceeds 30, it will be calculated as 30).

[0134] This represents the total number of times this person has handled faults across all models and subsystems since starting their career;

[0135] The sum of the complexity coefficients for handling all faults for this person, if ,but .

[0136] (Frequency Vector):

[0137] The frequency of operations on different subsystems within the past three months is denoted as follows: ;

[0138] in,

[0139] For the person who has handled the first case in the past three months The number of subsystem failures is used to reflect the "freshness" of personnel skills, preventing skills from becoming rusty due to long-term disuse.

[0140] Skill freshness This represents the person's number The recent proficiency level of each subsystem skill is maintained at a certain level, quantified from 0-100%, and calculated using the following formula:

[0141] ;

[0142] That is when At that time, the skill freshness of this subsystem is 100%; For every 1 unit reduced, the freshness decreases by 20%.

[0143] Skill Freshness Correction Coefficient The rules for correcting an individual's effective skill rating are as follows:

[0144] ,when ;

[0145] ,when ;

[0146] Revised Effective Skill Rating This is used for subsequent task matching and scheduling calculations.

[0147] (Workload Index)

[0148] The number of currently assigned tasks, their estimated duration, and their existing schedules for the next 24 hours (such as rest periods and routine inspections) are calculated in real time, with a quantification range of 0-100%. The formula is as follows:

[0149] ;

[0150] in,

[0151] This represents the number of tasks that person is currently handling.

[0152] For the first The estimated remaining time for each currently being processed task;

[0153] This represents the total time already scheduled for this person within the next 24 hours;

[0154] Standard working hours per day (default 8 hours / day);

[0155] If the load is considered too high, it is not recommended to assign new tasks. However, if there are urgent tasks and no other available candidates, this constraint can be broken, but a load warning and task priority adjustment must be triggered simultaneously.

[0156] (Location Vector)

[0157] Current location, on-duty status, availability for business trips, and willingness to travel are recorded as follows:

[0158] ;

[0159] in,

[0160] The current location of the person is coded, using the same coding rules as the location of the task, to match the task's Location dimension;

[0161] The status is "on duty" (1 = on duty, 0 = off duty).

[0162] The status is "Business trip is possible" (1 = Business trip is possible, 0 = Business trip is not possible).

[0163] The business trip willingness coefficient (0-1, where 1 represents the highest willingness and 0 represents unwillingness to travel) is set by the employee, with an initial value set by the employee. The system automatically adjusts it based on historical business trip data, using the following adjustment formula:

[0164] ;

[0165] Among them, the historical business trip acceptance rate = the number of historical business trip tasks accepted / the total number of historical business trip tasks assigned, and the value is always limited to the range of 0-1.

[0166] (Cost Factor)

[0167] Personnel's hourly wage, average daily cost of business trips, and travel expenses for cross-base dispatch are recorded as follows:

[0168] ;

[0169] in, This is the employee's hourly wage (yuan / hour).

[0170] The average daily cost of this employee's cross-base business trip (RMB / day);

[0171] Other costs (such as overtime pay, expert subsidies, etc.) are used for subsequent comprehensive cost calculations.

[0172] Task structured decomposition:

[0173] When a maintenance task (such as "Error in motion system of Unit 2 at Base A, code X123") is created, the system automatically deconstructs it into a task vector:

[0174] ;

[0175] The structured descriptions of each dimension are as follows to ensure accurate extraction of task features:

[0176] Based on task type, it is divided into four categories: "Fault Repair", "Planned Maintenance", "Software Upgrade" and "Periodic Qualification & Testing", corresponding to codes 1-4. Among them, fault repair corresponds to fault information in the simulator fault record system, while planned maintenance and periodic qualification & testing are generated periodically according to the relevant requirements of the Civil Aviation Administration.

[0177] The urgency level is divided into three levels: "Urgent" (affects training, inspection, and other tasks, requiring a response within 4 hours, code 3), "General" (planned maintenance, requiring a response within 24 hours, code 2), and "Non-urgent" (software upgrades, periodic evaluation & testing, requiring a response within 72 hours, code 1). The urgency level is automatically determined by the system based on the scope of the fault's impact and the simulator's usage plan, and can also be manually adjusted.

[0178] For the required skills, automatically map them to the skill tags of the required device model and subsystem, denoted as: ,in This is the device model code corresponding to the task. Code the corresponding subsystem for the task and set skill thresholds. (Default score is 60 points, which can be dynamically adjusted based on the urgency of the task and the complexity of the fault. The threshold can be increased for complex faults or urgent tasks, and decreased for non-urgent tasks.) The revised effective skill score of the candidate is required. .

[0179] To estimate the time consumption, the calculation is automatically performed based on the fault type, historical repair data, and subsystem complexity. The formula is as follows:

[0180] ;

[0181] in

[0182] This represents the total number of times the system has handled this type of fault in this subsystem of this machine model throughout its history.

[0183] For the first time in history Secondary processing of the same model Same subsystem If there is no historical data, the actual repair time for similar faults should be manually set to an initial value.

[0184] The location of the mission is denoted as the base code of the mission location, and is linked to the personnel location vector. In The same set of encoding rules is used, with one-to-one correspondence.

[0185] Intelligent matching and optimized scheduling:

[0186] Initial screening stage:

[0187] based on and From the personnel profile database of all bases nationwide, those whose skills meet the requirements (corrected effective skill scores) are selected. )and The status allows this task to be accepted. , The set of candidates who meet the task requirements is denoted as , The initial screening can quickly narrow down the candidate pool and improve the efficiency of subsequent optimization by limiting the number of candidates.

[0188] Optimization phase:

[0189] The personnel-task matching problem is modeled as a multi-objective optimization model, the core of which is the "multi-objective optimization algorithm", as follows:

[0190] Construction of multi-objective optimization model:

[0191] Objective function: Minimize overall cost while maximizing skill utilization and considering response efficiency. The final objective function is defined as follows:

[0192] ;

[0193] in,

[0194] The comprehensive objective function value for multi-objective optimization is given; a smaller value indicates better overall performance of the scheduling scheme. The parameters are defined and calculated as follows:

[0195] (Response Time): The estimated total time (in hours) between task arrival and personnel starting processing, including local response and travel time to other locations. The formula is:

[0196] ;

[0197] in,

[0198] Local response time refers to the standard time taken for personnel to receive the mission and prepare tools while at the mission base; the default is 0.5 hours.

[0199] This refers to the estimated total travel time for personnel traveling from their current base to the mission base. It is automatically calculated based on the mode of transportation (high-speed rail, airplane, car) between the two bases, combined with real-time traffic data. For example, the high-speed rail journey from Zhuhai base to Guangzhou base takes 1.5 hours, plus time spent at stations and during security checks. Set to 2 hours.

[0200] (Total Costs): Includes personnel salary costs, travel costs, and opportunity costs. The formula is as follows:

[0201] ;

[0202] in,

[0203] , All derived from personnel cost coefficient The corresponding parameters in;

[0204] The formula for calculating the number of business trip days is: That is, the total time of round-trip travel time plus the estimated time of the task, divided by 24 hours and rounded up, with the minimum value being 1 day; Opportunity cost refers to the cost incurred due to the delay of previously assigned tasks caused by scheduling this person to perform this task. The calculation formula is as follows: ;

[0205] in,

[0206] The estimated total time for delayed tasks, when there are no tasks to be executed. .

[0207] (Skill Utilization): The scheduling system is encouraged to prioritize assigning tasks to suitable personnel with high skill matching, provided that urgent tasks are met. The formula is:

[0208] ;

[0209] in

[0210] For candidate set The corrected effective skill rating for all personnel and their assigned tasks based on the required machine type and subsystem. The maximum value, The range is 0-100. The higher the value, the higher the skill matching degree, and the more likely it is to avoid "using a great talent for a small job" or "using a small talent for a big job".

[0211] The weight coefficients of the objective function are respectively the priority weights of response time, overall cost, and skill utilization rate, satisfying the following conditions: It can be dynamically adjusted according to business needs. Default value: Emergency Task ( )hour, General tasks ( )hour, Non-urgent tasks ( )hour, .

[0212] Constraints:

[0213] Load constraints: Candidate load index This is to avoid excessive staff workload leading to task delays or a decline in quality.

[0214] Urgency Constraint: Urgent Tasks ( Priority will be given to local or fastest-reaching destinations. The minimum number of personnel should be assigned to ensure timely response. The response time for emergency tasks must not exceed the fault response time limit stipulated by the Civil Aviation Administration and the company.

[0215] Skill Constraints: Revised Effective Skill Scores of Candidates And the skill freshness rate is ≥60% (if the freshness rate is less than 60%, additional online expert guidance is required).

[0216] Location constraints: personnel The status must meet the task requirements, and for remote tasks, it must be ensured that personnel can travel.

[0217] Implementation of a multi-objective optimization algorithm (improved NSGA-II algorithm):

[0218] An improved non-dominated sorting genetic algorithm with an elite strategy (NSGA-II) is used to solve the above multi-objective optimization model. Compared with traditional genetic algorithms and ant colony algorithms, this algorithm has the advantages of fast convergence speed, uniform solution set distribution, and effective ability to find Pareto optimal solutions. In addition, the algorithm is improved in combination with the business characteristics of simulator maintenance scheduling. The specific steps are as follows:

[0219] Initialize the population:

[0220] The individuals in the candidate set U are used as the initial population, and the population size is... ( (Number of candidates), each individual corresponds to one candidate, and the individual encoding uses real number encoding, with the encoding vector being... This ensures that the encoding corresponds one-to-one with the parameters of the target function.

[0221] Non-dominated sorting:

[0222] For each individual in the population, calculate its dominance relationship. Dominance relationship is defined as follows: for two individuals A and B, if A performs no worse than B on all objective functions, and performs strictly better than B on at least one objective function, then A is said to dominate B. For the three optimization objectives of this model, this means that A dominates B if and only if A... ≤B. A. ≤B. A. ≥B. A dominates B when at least one of the three inequalities is strictly true. Based on the dominance relationship, the population is divided into different non-dominated layers. The first layer is the optimal non-dominated solution, corresponding to the candidate with the best overall performance.

[0223] Crowding calculation:

[0224] Calculate the crowding degree of each individual in its non-dominated layer. The greater the crowding degree, the sparser the solutions around that individual. Retain individuals with high crowding degree to ensure the diversity of the solution set and avoid local optima.

[0225] Selection, crossover, and mutation operations:

[0226] The roulette wheel selection method is used to select parent individuals, with the crossover probability set to 0.8 and the mutation probability set to 0.05. Crossover is performed using a two-point crossover method, and mutation is performed using a Gaussian mutation method to generate the offspring population. At the same time, an elite retention strategy is added to directly retain the best non-dominated solution in the parent generation to the offspring to ensure the convergence speed of the algorithm.

[0227] Iteration termination:

[0228] Set the maximum number of iterations. If the iteration reaches the maximum number of iterations, or if the objective function F value of the best individual does not change significantly (change ≤ 0.01) for 10 consecutive generations, then the iteration stops and the Pareto optimal solution set is output.

[0229] Solution and output:

[0230] From the Pareto optimal solution set, based on the weight coefficients of the objective function, the individual with the highest overall score (smallest F-value) is selected as the optimal solution, and the optimal dispatch suggestion is output: It also outputs 2 to 3 alternative solutions for managers to make decisions.

[0231] Dynamic rescheduling:

[0232] If the selected personnel refuse, the task is canceled, a new higher priority task appears, the task attributes change (such as an increase in urgency or increased fault complexity), or the personnel status changes abruptly (such as sudden absence from duty or overload), the system will automatically trigger a new round of optimization calculation based on the current status (personnel load, location, task urgency) to achieve real-time rescheduling. The rescheduling response time is ≤1 minute, ensuring the flexibility and adaptability of the scheduling scheme.

[0233] Optional supplementary features:

[0234] Learning-based scheduling:

[0235] The system records the execution results of each scheduling session (actual completion time, actual cost, user satisfaction, and first-time fault repair rate), and uses this as feedback to continuously optimize the weights in the objective function using the gradient descent algorithm. The loss function is defined as the deviation between the actual effect after scheduling and the expected result. ;

[0236] in,

[0237] As the weights of the loss terms, the loss is minimized through gradient descent, and the weight coefficients are iteratively updated to make the scheduling strategy continuously closer to the actual business preferences. After training with more than 1,000 scheduling cases, the satisfaction rate of the scheduling scheme can be improved to over 95%.

[0238] Visualized scheduling dashboard: Displays a heat map of maintenance personnel (based on location and load), current task distribution map, optimal scheduling path suggestions, and simulator fault status, providing managers with global decision support. At the same time, it allows real-time viewing of the execution progress and personnel status of each task, realizing visualization and transparency of operation and maintenance management.

[0239] Fault closed-loop management: Connect to the simulator fault recording system to record data of the entire fault repair process, including fault codes, repair steps, tools used, personnel operation records, etc., and automatically archive them to the knowledge base. At the same time, update personnel capability profiles to form a closed loop of "fault record - dispatching and order assignment - repair execution - data feedback - profile update".

[0240] The following describes this embodiment in conjunction with application scenarios:

[0241] Suppose that at 10:00 AM one morning, the motion system (subsystem code YS) of the A320 simulator B-01 at the Guangzhou base (code GZ) suddenly malfunctions, with fault code X123 (a complex fault, ATA chapter classification D=3), causing training to be interrupted. The system operates according to the following procedure:

[0242] Task creation:

[0243] The system automatically retrieves fault information from the simulator data acquisition system and fault recording system, and generates tasks accordingly. Task vector:

[0244] ;

[0245] in,

[0246] (Fault repair);

[0247] (urgent);

[0248] (Model code A320, subsystem code YS);

[0249] Hours (Based on historical data, the average repair time for complex faults in the A320 motion system is 4 hours).

[0250] Skill Threshold point;

[0251] Emergency Task Weighting .

[0252] Initial screening:

[0253] The system searched the national database and found skills that meet the "A320-Motor System" criteria (corrected valid skill score). A total of 15 people (with scores below 10) and whose status allows them to undertake the task, including 3 locals from Guangzhou (Zhang San, Li Si, and Wang Wu) and 12 from other places such as Zhuhai, Hebei, and Xinjiang, forming a candidate group. (Total of 15 people).

[0254] Profile matching (calculated based on a skill vector scoring model):

[0255] The profile data of the core personnel in the candidate pool is as follows:

[0256] Zhang San (Guangzhou Base): ( , (Calculated to 92 points) point, (Freshness 100%, correction coefficient η=1, effective skill score 92 points). (Currently handling a complex fault, estimated remaining time 2 hours). , .

[0257] Li Si (Guangzhou Base): ( (Calculated score: 70 points) point, (Freshness 60%, correction coefficient η=1, effective skill score 70 points). (No current task) , .

[0258] Zhao Liu (Zhuhai Base, Code ZH): ( (Calculated score: 98) point, (Freshness 100%, correction coefficient η=1, effective skill score 98 points). (No current task) , .

[0259] Optimization solution (based on the improved NSGA-II algorithm):

[0260] The system calculates the F-value for each candidate based on the objective function. The core calculation process is as follows:

[0261] Option A (Zhang San): Hours (0.5 hours for local response + 2 hours for current task to complete), Yuan;

[0262] ;

[0263] .

[0264] Option B (Li Si): Hour;

[0265] Yuan;

[0266] ;

[0267] .

[0268] Option C (Zhao Liu): Hours (2 hours on the go + 0.5 hours local response);

[0269] Yuan;

[0270] ;

[0271] .

[0272] Output and execution:

[0273] After the supervisor confirms Solution B, the system automatically pushes a task order to Li Si's mobile device, simultaneously pushing historical repair cases for the fault, online expert contact information, and updating Li Si's load index. .

[0274] Closed-loop feedback:

[0275] Under expert guidance, Li Si successfully repaired the fault within 4 hours (actual time: 4 hours, actual cost: 270 yuan). The system recorded the actual data for this dispatch and updated Li Si's profile data.

[0276] Improved to 73 points ( and promote), Improved to 62 points (80% freshness), and at the same time archive the fault repair case to the knowledge base for subsequent skill scoring and scheduling optimization.

[0277] This embodiment selected five training bases (Guangzhou, Zhuhai, Hebei, Xinjiang, and Chongqing) of a major airline, 36 simulators (covering A320, A330, A350, B737, B787, and other aircraft types), and 100 maintenance personnel for a three-month comparative experiment. The experiment was divided into two groups: the experimental group (using the intelligent scheduling system of this invention) and the control group (using the traditional manual scheduling mode). The experimental data are shown in the table below:

[0278] Cross-base response time (average) 2.5 hours 24 hours An increase of 89.6% First-time repair rate of complex faults 92.3% 75.0% An increase of 17.3% Overall Operation and Maintenance Costs (Monthly Average) 867,000 yuan 1.26 million yuan Reduced by 31.2% Task delay rate 2.1% 18.3% Reduced by 88.5% Personnel skill matching (average) 82.6% 49.0% An increase of 68.4% Skills enhancement for general staff (3-month average) 18.7% 5.2% An increase of 259.6% Scheduling decision efficiency (average time) 1.2 minutes 4.3 minutes An increase of 72.5%

[0279] In addition, the core algorithm was validated separately, and the performance of the improved NSGA-II algorithm was compared with that of the traditional genetic algorithm and ant colony algorithm. The experimental data are shown in the table below:

[0280] Improved NSGA-II algorithm 30s 68.5 excellent Traditional genetic algorithm 55th generation 89.2 generally Traditional ant colony algorithm 62nd generation 95.7 Poor

[0281] For the purpose of simplicity, the method steps disclosed in the above embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0282] like Figure 2 As shown, the present invention also provides a simulator maintenance task scheduling system based on fault prediction and capability profiling, comprising:

[0283] The dynamic capability profile vector construction module 201 is configured to acquire multi-source data of maintenance personnel and construct a dynamic capability profile vector of the maintenance personnel based on the multi-source data.

[0284] The task vector organization module 202 is configured to acquire task information of simulator maintenance tasks and organize the task information into task vectors;

[0285] The fault prediction module 203 is configured to perform fault prediction, construct a subsystem fault propagation model, predict the fault probability and occurrence time within a future time window based on real-time operating parameters, and generate a pre-task vector containing skill tags of the fault subsystem and root factor system when the predicted probability exceeds the threshold, and cache the corresponding scheduling scheme. When the actual maintenance task occurs, if it matches the cached scheduling scheme, it will be output directly.

[0286] The candidate set construction module 204 is configured to otherwise select personnel from the dynamic ability profile vector whose skill dimension meets the skill threshold and whose position dimension meets the task requirements based on the skill-related information and location-related information in the task vector, as well as a preset skill threshold, to form a candidate set.

[0287] The optimization scheduling module 205 is configured to treat each candidate in the candidate set and the matching scheme of the maintenance task as an individual, construct a multi-objective optimization model with the goal of minimizing the comprehensive objective function value, and include constraints. Based on the optimization objective and constraints, a multi-objective optimization algorithm is used to iteratively solve the problem to obtain a Pareto optimal solution set.

[0288] The output execution module 206 is configured to select the individual with the smallest comprehensive objective function value from the Pareto optimal solution set as the optimal scheduling scheme output, and output alternative schemes to guide the execution of the simulator maintenance task.

[0289] It is worth noting that although only some basic functional modules are disclosed in the embodiments of this invention, it does not mean that the composition of this system is limited to the above-mentioned basic functional modules. On the contrary, what this embodiment intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can arbitrarily add one or more functional modules in combination with existing technology to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. The fact that this embodiment only discloses a few basic functional modules should not be considered as the scope of protection of the claims of this invention being limited to the disclosed basic functional modules. At the same time, for the convenience of description, the above device is described separately according to its functions as various units and modules. Of course, in implementing this invention, the functions of each unit and module can be implemented in one or more software and / or hardware.

[0290] like Figure 3 As shown, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of a simulator maintenance task scheduling method based on fault prediction and capability profiling.

[0291] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 3 The structure shown in this embodiment of the invention includes an electronic device comprising one or more processors 710 and a memory 720; the processors 710 in this electronic device may be one or more. Figure 3 Taking a processor 710 as an example; a memory 720 is used to store one or more programs; the one or more programs are executed by the one or more processors 710, so that the one or more processors 710 implement a simulator maintenance task scheduling method based on fault prediction and capability profiling as described in any one of the embodiments of the present invention.

[0292] The electronic device may also include an input device 730 and an output device 740.

[0293] The processor 710, memory 720, input device 730, and output device 740 in this electronic device can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0294] The memory 720 in this electronic device serves as a computer-readable storage medium, capable of storing one or more programs. These programs can be software programs, computer-executable programs, or modules, such as the program instructions / modules corresponding to the simulator maintenance task scheduling method based on fault prediction and capability profiling provided in this embodiment of the invention. The processor 710 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 720, thereby implementing the simulator maintenance task scheduling method based on fault prediction and capability profiling described in the above method embodiment.

[0295] The memory 720 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 720 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 720 may further include memory remotely located relative to the processor 710, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0296] Input device 730 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the electronic device. Output device 740 may include display devices such as a display screen.

[0297] The present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a simulator maintenance task scheduling method based on fault prediction and capability profiling.

[0298] Specifically, the computer storage medium in this embodiment of the invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0299] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A simulator maintenance task scheduling method based on fault prediction and capability profiling, characterized in that, include: Acquire multi-source data of maintenance personnel, and construct a dynamic capability profile vector of the maintenance personnel based on the multi-source data; Obtain task information for simulator maintenance tasks, and organize the task information into a task vector; Perform fault prediction, construct a subsystem fault propagation model, predict the fault probability and occurrence time within the future time window based on real-time operating parameters, and generate a pre-task vector containing skill tags of fault subsystem and root factor system when the predicted probability exceeds the threshold, and cache the corresponding scheduling scheme. When the actual maintenance task occurs, if it matches the cached scheduling scheme, it will be output directly. Otherwise, based on the skill-related and location-related information in the task vector, as well as the preset skill threshold, personnel who meet the skill threshold in the skill dimension and the task requirements in the location dimension are selected from the dynamic ability profile vector to form a candidate set. Each candidate in the candidate set and the matching scheme of the maintenance task are treated as an individual. A multi-objective optimization model is constructed with the goal of minimizing the comprehensive objective function value and includes constraints. Based on the optimization objective and constraints, a multi-objective optimization algorithm is used to iteratively solve the problem to obtain the Pareto optimal solution set. The individual with the smallest comprehensive objective function value is selected from the Pareto optimal solution set as the optimal scheduling scheme output, and alternative schemes are also output to guide the execution of the simulator maintenance task.

2. The simulator maintenance task scheduling method based on fault prediction and capability profiling according to claim 1, characterized in that, Acquiring multi-source data on maintenance personnel, and constructing a dynamic capability profile vector of the maintenance personnel based on the multi-source data, further includes: Calculate the skill vector score of the maintenance personnel for each machine model and each subsystem; The skill freshness is calculated based on the number of times a specific subsystem fault is handled within a preset time period, and the skill vector score is attenuated and corrected according to the skill freshness to obtain an effective skill score. Calculate the experience value and load index of the maintenance personnel; The location information and cost coefficient of the maintenance personnel are obtained, and the dynamic capability profile vector is composed of the effective skill score, the experience value, the skill freshness, the load index, the location information and the cost coefficient.

3. The simulator maintenance task scheduling method based on fault prediction and capability profiling according to claim 2, characterized in that, Calculating the skill vector score of the maintenance personnel for each machine model and each subsystem further includes: The training data of the maintenance personnel for specific machine models and subsystems is obtained. A training score is obtained by weighting the number of training sessions and exam scores. The historical records of the maintenance personnel's handling of faults of the machine model and subsystem are obtained. Based on the number of handling sessions and fault complexity, the data is normalized with the maximum value of the corresponding indicators of all maintenance personnel in the system for the machine model and subsystem to obtain a maintenance experience score. The success rate of the maintenance personnel in handling faults of the machine model and subsystem is obtained to obtain a fault repair success rate score. The training score, maintenance experience score, and fault repair success rate score are weighted and combined to obtain the maintenance personnel's skill vector score for a specific model and subsystem, with the maintenance experience score having the largest weight. The skill freshness is determined based on the number of times the maintenance personnel handle specific subsystem faults within a preset time period. When the number of handlings reaches a preset threshold, the skill freshness is at its maximum value. The skill freshness decreases by a preset step size for each decrease in the number of handlings. When the skill freshness is lower than a preset percentage, the skill vector score is attenuated and corrected to obtain an effective skill score. The empirical value is obtained by weighting and summing the maintenance personnel's years of experience, the total number of faults handled, and the average fault complexity.

4. The simulator maintenance task scheduling method based on fault prediction and capability profiling according to claim 1, characterized in that, Obtain task information for simulator maintenance tasks, organize the task information into a task vector, and further include: Obtain the fault code, fault type, and fault model from the fault record, and determine the task type as one of fault repair, planned maintenance, software upgrade, or periodic qualification test; Based on the scope of the fault's impact and the simulator's usage plan, the urgency level is determined as urgent, general, or non-urgent. The faulty machine model and the subsystems involved in the fault are mapped to the required skill tags, and the skill thresholds are dynamically adjusted according to the urgency of the task and the complexity of the fault. Retrieve historical repair records from the historical maintenance database that are identical to the current task's model, subsystem, and fault type. Determine the estimated time for the task based on the statistical values ​​of historical repair time. Organize the task type, urgency level, required skill tags, estimated time, and task location into the task vector.

5. The simulator maintenance task scheduling method based on fault prediction and capability profiling according to claim 1, characterized in that, Each candidate in the candidate set and its matching scheme with the maintenance task are treated as an individual. A multi-objective optimization model is constructed with the goal of minimizing the comprehensive objective function value. The comprehensive objective function includes at least a weighted combination of three dimensions: response time, comprehensive cost, and skill utilization rate, and further includes: The overall objective function is defined as a weighted combination of response time, overall cost, and skill utilization rate, where response time and overall cost are positive terms, and skill utilization rate is a negative term. The response time is determined, which consists of local response time and cross-base travel time, wherein the cross-base travel time is determined based on the mode of transportation between the candidate's current base and the mission location; The comprehensive cost is determined, which consists of personnel salary cost, travel cost and opportunity cost. The personnel salary cost is determined based on the candidate's hourly salary and the duration of the task. The travel cost is determined based on the candidate's average daily cost of cross-base business trips and the number of business trip days. The opportunity cost is the cost incurred due to the delay of other pending tasks caused by scheduling the candidate. The constraints include that the candidate's current task load does not exceed a preset high load threshold, the response time of an emergency task does not exceed a specified fault response time limit, the candidate's skill level is not lower than the skill threshold, the skill level is determined based on the skill information in the dynamic capability profile vector, and the candidate's location dimension meets the requirements for local or remote execution of the task.

6. The simulator maintenance task scheduling method based on fault prediction and capability profiling according to claim 1, characterized in that, A multi-objective optimization algorithm is used for iterative solution to obtain the Pareto optimal solution set, which further includes: The multi-objective optimization model is iteratively optimized using a non-dominated sorting genetic algorithm with an elitist strategy. During the iteration process, non-dominated sorting is performed based on the dominance relationship of individuals, and crowding is calculated to maintain the diversity of the solution set. Offspring population is generated through selection, crossover, and mutation operations. The optimal non-dominated solution in the parent generation is retained in the offspring generation. When the preset iteration termination condition is reached, the Pareto optimal solution set is output.

7. The simulator maintenance task scheduling method based on fault prediction and capability profiling according to claim 1, characterized in that, The individual with the smallest comprehensive objective function value is selected from the Pareto optimal solution set as the optimal scheduling scheme and output as an alternative scheme to guide the execution of the simulator maintenance task. This further includes: The optimal scheduling scheme and alternative schemes are used to allocate maintenance personnel to perform the simulator maintenance tasks; During task execution, when the rescheduling trigger condition is met, the Pareto optimal solution set and solution output are recalculated using the latest status of all personnel at the trigger time and all unfinished tasks as input. The rescheduling trigger condition includes at least one of the following: the selected personnel refuse the task, the original task is canceled, a new higher priority task appears, the urgency or complexity of the current task changes, or the on-duty status of the selected personnel or the current task load changes abruptly.

8. A simulator maintenance task scheduling system based on fault prediction and capability profiling, characterized in that, include: The dynamic capability profile vector construction module is configured to acquire multi-source data of maintenance personnel and construct a dynamic capability profile vector of the maintenance personnel based on the multi-source data. The task vector organization module is configured to acquire task information of simulator maintenance tasks and organize the task information into task vectors. The fault prediction module is configured to perform fault prediction, build a subsystem fault propagation model, predict the probability of faults and the occurrence time within a future time window based on real-time operating parameters, and generate a pre-task vector containing skill tags of fault subsystems and root factor systems when the predicted probability exceeds a threshold, and cache the corresponding scheduling scheme. When an actual maintenance task occurs, if it matches the cached scheduling scheme, it will be output directly. The candidate set construction module is configured to otherwise select personnel from the dynamic ability profile vector whose skill dimension meets the skill threshold and whose position dimension meets the task requirements based on the skill-related information and location-related information in the task vector, as well as a preset skill threshold, to form a candidate set. The optimization scheduling module is configured to treat each candidate in the candidate set and the matching scheme of the maintenance task as an individual, construct a multi-objective optimization model with the goal of minimizing the comprehensive objective function value, and include constraints. Based on the optimization objective and constraints, a multi-objective optimization algorithm is used to iteratively solve the problem to obtain the Pareto optimal solution set. The output execution module is configured to select the individual with the smallest comprehensive objective function value from the Pareto optimal solution set as the optimal scheduling scheme output, and output alternative schemes to guide the execution of the simulator maintenance task.

9. An electronic device, characterized in that, include: The system includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the method according to any one of claims 1 to 7.

Images