Flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction

Abstract

The present application belongs to the field of simulation machine maintenance, and specifically relates to a flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction. It aims to solve the problems of excessive maintenance, insufficient maintenance and sudden failure risk in the prior art. The present application comprises: collecting data in real time through multiple sensors, extracting features and mapping them into health scores, and obtaining a comprehensive health index through dynamic weighted fusion; using historical and current data to construct a degradation model to predict the remaining life and health trend; constructing a multi-dimensional cost function by integrating multiple costs, and calculating the optimal maintenance time under the constraints of health, time and resources; finally, automatically generating a maintenance work order and coordinating with the training plan system, the present application realizes precise predictive maintenance and intelligent resource allocation, significantly improving maintenance efficiency and equipment availability.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of simulation machine maintenance, and particularly relates to a flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction. BACKGROUND

[0002] The flight simulator is a key high-precision device for pilot training, and its reliability and availability are directly related to training safety and efficiency. At present, the maintenance of the flight simulator mainly relies on the traditional planned maintenance mode, which includes the following ways: first, regular maintenance based on calendar time, that is, maintenance inspection is performed strictly according to the fixed time period specified by the equipment manufacturer, regardless of the actual running state of the equipment; second, maintenance based on usage time, which triggers a maintenance work order when the actual running hours of the flight simulator reach a preset threshold; third, a simple scheduling optimization system is used, and the core function of which only focuses on matching and scheduling the time window under limited maintenance resources.

[0003] However, the above existing maintenance strategies have many significant defects and limitations in practice. First, there is a widespread problem of over-maintenance, that is, disassembly, inspection and replacement are still carried out according to the fixed period even if the health state of the equipment is still good, resulting in a huge waste of manpower, material resources and spare parts resources, and possibly introducing new failure risks due to unnecessary intervention. Second, the existing methods lack the ability to predict potential equipment failures and are difficult to identify abnormal degradation trends of the equipment, resulting in insufficient maintenance and being unable to effectively avoid unplanned interruptions caused by sudden failures during training, which seriously affects the training progress. Third, the existing maintenance scheduling system is rigid and cannot dynamically and flexibly adjust the maintenance time according to the real-time health state and degradation rate of the equipment, so it cannot achieve optimal allocation of resources.

[0004] In addition, with the development of sensing technology, modern flight simulators generate a large amount of multi-source monitoring data (such as vibration, pressure, temperature, current, etc.) during operation, but the existing maintenance system makes very insufficient use of these data, and the value of the data cannot be effectively mined to support intelligent decision-making. Most of the data are only used for post-fault diagnosis and historical state playback, and cannot serve the real-time evaluation of the health state and the prediction of future trends of the equipment.

[0005] Therefore, the field of flight simulator maintenance urgently needs an intelligent method that can deeply integrate multi-source data, accurately evaluate the current health state, predictively judge the remaining useful life, and dynamically optimize maintenance decisions under multi-dimensional constraints, in order to overcome the shortcomings of the existing technology, realize the paradigm shift from "planned maintenance" to "predictive maintenance", and ultimately ensure training safety, improve equipment availability and reduce the life cycle maintenance cost.

[0006] Based on this, the application proposes a flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction. SUMMARY

[0007] In order to solve the above problems in the prior art, that is, the problems of excessive maintenance, insufficient maintenance and sudden failure risk in the prior art, the application provides a flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction, which comprises:

[0008] Through the deployment of multiple sensors on the flight simulator, multi-source monitoring data is collected in real time, the multi-source monitoring data is processed and the characteristic information corresponding to the health state is extracted, and each characteristic information is mapped to the corresponding health score based on a preset health mapping function;

[0009] According to the influence degree of the monitored parameters of the sensor on the function of the equipment and the historical data fluctuation level, the weight of each health score is dynamically adjusted, and a comprehensive health index is obtained through weighted fusion;

[0010] Based on the historical health data and the comprehensive health index, an equipment degradation trajectory model is constructed to predict the remaining useful life of the equipment and the future health state change trend;

[0011] A multi-dimensional cost function is constructed by combining the preventive maintenance cost, the failure risk cost, the training interruption cost and the spare parts preparation cost, and the optimal maintenance opportunity is dynamically calculated under the conditions of meeting the health constraint, the time window constraint and the resource constraint;

[0012] According to the optimal maintenance opportunity and the comprehensive health index, a predictive maintenance work order is automatically generated, the maintenance work order is cooperatively scheduled with the training plan system, and predictive maintenance and intelligent scheduling are realized.

[0013] Further, the method for processing the multi-source monitoring data and extracting the characteristic information corresponding to the health state is:

[0014] The multi-source monitoring data is subjected to data cleaning and quality evaluation to eliminate outliers and noise;

[0015] The cleaned data is subjected to multi-sampling rate data synchronous processing to unify the monitoring data of different sources to the same time reference;

[0016] The time domain features, frequency domain features and time-frequency domain features are extracted from the synchronized data to comprehensively represent the health state of the equipment.

[0017] Further, the method for mapping each characteristic information to the corresponding health score based on the preset health mapping function is:

[0018] For each feature information extracted from the monitoring data, a Gaussian function is adopted as a health mapping function to map a current value of the feature information to an initial health score between 0 and 1, wherein when the current value is equal to a normal baseline value of the feature information obtained in advance through historical data statistics, the initial health score obtained by mapping is 1, indicating a completely healthy state;

[0019] A data sequence of the feature information in a recent historical time window is obtained, a deterioration trend strength of the feature information is calculated through a trend analysis algorithm, and a trend penalty factor between 0 and 1 is generated;

[0020] The initial health score is multiplied by a difference between 1 and the trend penalty factor to obtain a modified final health score, wherein when there is a deterioration trend, the trend penalty factor is greater than zero, so that the final health score is lower than the initial health score.

[0021] Further, according to the influence degree of the sensor monitored parameter on the equipment function and the historical data fluctuation level, the weights of the health scores are dynamically adjusted, and a comprehensive health index is obtained through weighted fusion, and the method is:

[0022] Based on the correlation degree between the monitoring parameters and the equipment historical failure records, the influence degree of the monitoring parameters on the equipment function is determined;

[0023] According to the fluctuation amplitude of the sensor historical data deviating from the normal baseline value of the sensor, the fluctuation level of the sensor data is evaluated;

[0024] According to the determined influence degree and the evaluated fluctuation level, the weight coefficients corresponding to the health scores are dynamically calculated and updated;

[0025] After multiplying each health score by the weight coefficient corresponding to the health score, the results are superimposed, and the superimposed results are normalized to obtain the comprehensive health index.

[0026] Further, based on the correlation degree between the monitoring parameters and the equipment historical failure records, the influence degree of the monitoring parameters on the equipment function is determined, and the method is:

[0027] Obtain the equipment historical failure records containing historical failure events and their occurrence time;

[0028] Extract data sequences of various monitoring parameters within a preset time window before each historical failure event occurs;

[0029] Using statistical correlation analysis method, the correlation strength between the data sequence of each type of monitoring parameter and the occurrence of each type of historical failure event is calculated;

[0030] The calculated correlation strength is normalized and mapped to an influence degree value of the corresponding monitoring parameter on the device function.

[0031] Further, the fluctuation level of the sensor data is evaluated according to the fluctuation amplitude of the sensor historical data deviating from the normal reference value of the sensor, and the method is as follows:

[0032] Obtain the data sequence output by the sensor within a preset historical period;

[0033] Read the normal working interval reference value of the sensor obtained by statistical analysis of historical data of the normal running state of the device;

[0034] Calculate the amplitude of each data point in the data sequence deviating from the normal working interval reference value, and calculate the overall fluctuation quantitative index based on the amplitude;

[0035] Compare the calculated overall fluctuation quantitative index with the preset fluctuation threshold, and determine the fluctuation level grade of the sensor data according to the comparison result.

[0036] Further, according to the determined influence degree and the evaluated fluctuation level, the weight coefficient corresponding to each health score is dynamically calculated and updated, and the method is as follows:

[0037] Convert the influence degree value corresponding to each monitoring parameter into a correlation coefficient;

[0038] Convert the fluctuation level grade corresponding to each sensor data into a stability coefficient;

[0039] Linearly combine the correlation coefficient and the stability coefficient according to a preset ratio to generate an initial weight factor;

[0040] Normalize the initial weight factor so that the sum of the weight coefficients of all monitoring parameters is 1, and use the normalized result as the latest weight coefficient corresponding to each health score in weighted fusion.

[0041] Further, based on the historical health data and the current comprehensive health index, a device degradation trajectory model is constructed to predict the remaining useful life of the device and the future health state change trend, and the method is as follows:

[0042] Obtain a historical health data sequence containing historical timestamps and corresponding comprehensive health indexes;

[0043] Using a regression analysis method, take time as the independent variable and the comprehensive health index as the dependent variable, and fit to obtain a degradation trajectory model representing the overall performance degradation law of the device, the degradation trajectory model including linear and nonlinear degradation terms related to time;

[0044] identify and extract external stress factors affecting the degradation of the equipment, and introduce the external stress factors as covariates into the degradation trajectory model to quantify the influence of different stress conditions on the degradation rate of the equipment;

[0045] Based on the latest fitted degradation trajectory model, extrapolate the predicted time required for the comprehensive health index to be below the preset critical health threshold, and determine the predicted time as the remaining service life of the equipment;

[0046] Calculate the predicted health index at a future specific time point through the degradation trajectory model, and generate a prediction curve for representing the future health state change trend of the equipment.

[0047] Further, a multi-dimensional cost function is constructed by combining preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost, and the method is as follows:

[0048] A preventive maintenance cost model is constructed to represent the basic cost required for performing preventive maintenance activities, and includes an incremental cost item that increases with the delay of maintenance timing, for reflecting the increase of maintenance complexity caused by potential deterioration of the equipment state;

[0049] A failure risk cost model is constructed by multiplying the total loss cost caused by a single failure event with the probability of failure occurring before a future maintenance timing predicted based on the health trajectory of the equipment;

[0050] A training interruption cost model is constructed by multiplying the loss cost caused by each hour of training interruption, the probability of conflict between the planned maintenance time window and the intended training plan, and the predicted maintenance duration;

[0051] A spare parts preparation cost model is constructed by aggregating the unit price, quantity of various required spare parts, and a preparation time factor function related to the lead time of spare parts procurement or allocation;

[0052] The outputs of the preventive maintenance cost model, the failure risk cost model, the training interruption cost model and the spare parts preparation cost model are summed up to generate a total cost function for evaluating the economy of different maintenance timings as a multi-dimensional cost function.

[0053] Further, under the conditions of meeting health constraints, time window constraints and resource constraints, the optimal maintenance timing is dynamically calculated, and the method is as follows:

[0054] The health constraint is defined as that the predicted health index at the maintenance timing should not be lower than the preset health index safety threshold;

[0055] define the time window constraint as the selected maintenance opportunity must belong to the set of available time windows allowed by the system, which is determined by the training plan gaps and the maintenance team availability;

[0056] define the resource constraint as the total amount of various resources required to perform the maintenance task, which must not exceed the total amount of real-time available resources within the time window in which the maintenance opportunity is located;

[0057] among all feasible maintenance opportunities that simultaneously satisfy the health constraint, the time window constraint and the resource constraint, calculate the total cost function value corresponding to each feasible maintenance opportunity;

[0058] select the feasible maintenance opportunity with the minimum total cost function value and determine it as the final optimal maintenance opportunity.

[0059] Advantages of the present application:

[0060] The present application can comprehensively and accurately perceive the real-time running state of the flight simulator by fusing and processing multi-source monitoring data and extracting health state features, thereby providing reliable data basis for subsequent maintenance decision.

[0061] The present application realizes precise quantitative evaluation of the overall health state of the equipment by dynamically adjusting the weight and fusing calculation to obtain the comprehensive health index, significantly improves the early identification ability of potential faults, and effectively avoids the problems of insufficient maintenance or excessive maintenance.

[0062] The present application realizes advanced prediction of future performance changes of the equipment by constructing the equipment degradation trajectory model and predicting the remaining useful life, so that the maintenance strategy changes from passive response to active intervention, thereby greatly reducing the risk of sudden failure in the training process.

[0063] The present application realizes the best balance between maintenance cost and equipment reliability by constructing a multi-dimensional cost function and dynamically optimizing the maintenance opportunity under multiple constraints, thereby significantly improving the utilization efficiency of maintenance resources and economic benefits.

[0064] The present application realizes seamless connection of maintenance activities and training tasks by automatically generating predictive maintenance work orders and coordinating scheduling with the training plan system, thereby minimizing the interference of maintenance work on normal training arrangement, improving the overall availability of the equipment and the training efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0065] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0066] Figure 1 is a flowchart of a flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction. DETAILED DESCRIPTION

[0067] The application will be described in further detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. In addition, it should be noted that only parts related to the application are shown in the drawings for ease of description.

[0068] It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict. The application will be described in further detail below with reference to the drawings and embodiments.

[0069] The first embodiment of the application provides a flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction, which comprises:

[0070] Step S10, real-time collection of multi-source monitoring data by a plurality of sensors deployed on the flight simulator, processing of the multi-source monitoring data and extraction of feature information corresponding to the health state, mapping of each feature information to a corresponding health score based on a preset health mapping function;

[0071] Step S20, dynamic adjustment of the weight of each health score according to the influence degree of the monitored parameters of the sensor on the device function and the historical data fluctuation level, and obtaining of a comprehensive health index through weighted fusion;

[0072] Step S30, construction of a device degradation trajectory model based on historical health data and the comprehensive health index, prediction of the remaining useful life of the device and the future health state change trend;

[0073] Step S40, construction of a multi-dimensional cost function in combination with preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost, and dynamic calculation of the optimal maintenance opportunity under the conditions of meeting health constraints, time window constraints and resource constraints;

[0074] Step S50, automatic generation of a predictive maintenance work order according to the optimal maintenance opportunity and the comprehensive health index, coordination of the maintenance work order and the training plan system for scheduling, and realization of predictive maintenance and intelligent scheduling.

[0075] In order to more clearly describe the flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction, the following will be described in combination with Figure 1 The steps in the embodiments of the application will be described in detail, and the detailed description of each step is as follows:

[0076] Step S10, real-time collection of multi-source monitoring data through a plurality of sensors deployed on the flight simulator, processing of the multi-source monitoring data and extraction of feature information corresponding to the health state, mapping of each of the feature information to a corresponding health score based on a preset health mapping function;

[0077] Real-time collection of multi-source monitoring data through a plurality of sensors deployed on the flight simulator, wherein the sensors include vibration sensors, pressure sensors, temperature sensors, current sensors and position encoders; the vibration sensors are mainly used for monitoring vibration characteristics of the control system and the motion platform, the pressure sensors are used for monitoring pressure fluctuation of the hydraulic system, the temperature sensors are used for monitoring temperature distribution state of key components, the current sensors are used for collecting working current data of servo motors, and the position encoders are used for detecting motion accuracy deviation; through the cooperative work of these multiple types of sensors, comprehensive and real-time perception and data collection of running states of each key subsystem of the flight simulator are realized, thereby providing multi-dimensional and high-frequency raw data support for subsequent health state evaluation and predictive maintenance.

[0078] In this embodiment, the method for processing the multi-source monitoring data and extracting feature information corresponding to the health state is as follows:

[0079] Step S11, data cleaning and quality evaluation of the multi-source monitoring data to eliminate abnormal values and noise;

[0080] Step S12, multi-sampling rate data synchronization processing of the cleaned data to unify monitoring data of different sources to the same time reference;

[0081] Step S13, extraction of time domain features, frequency domain features and time-frequency domain features from the synchronized data to comprehensively represent the health state of the equipment.

[0082] In this embodiment, the data cleaning and quality evaluation of the multi-source monitoring data are performed by setting reasonable threshold ranges and data rationality rules to automatically identify and eliminate abnormal values and noise points caused by instantaneous sensor failure and signal transmission interference, and by using a sliding window average filtering algorithm to smooth the original data to suppress random noise and improve data quality and reliability;

[0083] Subsequently, multi-sampling rate data synchronization processing of the cleaned data is performed, time stamp alignment and interpolation algorithms are used to unify vibration, pressure, temperature, current and position signals to the same time reference, and multi-source data streams with consistent time sequences are formed to provide a time sequence consistent data basis for subsequent feature extraction and fusion analysis;

[0084] Finally, time domain features, frequency domain features and time-frequency domain features are extracted from the synchronized multi-source data stream, wherein the time domain features include mean, root mean square, variance, peak factor and waveform factor, the frequency domain features are obtained by fast Fourier transform of the signal to obtain the amplitude, frequency and power spectral density of the main frequency component, and the time-frequency domain features are obtained by wavelet transform or short-time Fourier transform to obtain the energy distribution characteristics of the signal in the time-frequency joint domain, so as to comprehensively and multi-angelly represent the health state and performance degradation trend of the equipment.

[0085] In this embodiment, each feature information is mapped to a corresponding health score based on a preset health mapping function, and the method is as follows:

[0086] In step S14, for each feature information extracted from the monitoring data, a Gaussian function is used as a health mapping function to map the current value of the feature information to an initial health score between 0 and 1, wherein when the current value is equal to the normal reference value of the feature information obtained by historical data statistics in advance, the initial health score obtained by mapping is 1, indicating a completely healthy state.

[0087] In step S15, a data sequence of the feature information in a recent historical time window is obtained, a deterioration trend strength of the feature information is calculated by a trend analysis algorithm, and a trend penalty factor between 0 and 1 is generated.

[0088] In step S16, the initial health score is multiplied by the difference between 1 and the trend penalty factor to obtain a final health score after correction, wherein when there is a deterioration trend, the trend penalty factor is greater than zero, so that the final health score is lower than the initial health score.

[0089] For the feature values extracted from each sensor data, a health mapping mechanism based on a Gaussian function is used to calculate the health score. Specifically, for the i-th sensor, first, the key feature value is extracted from its monitoring data, such as the root mean square value of the i-th sensor The specific form of the Gaussian function is , wherein is a normal working reference value obtained by analyzing a large amount of historical data of the sensor in a normal running state of the equipment, is a standard deviation threshold value reflecting the normal fluctuation range of the parameter determined according to the historical data distribution, which is usually 3 times the standard deviation. The characteristic of the mapping function is that when the feature value is equal to the normal reference value , the health score is 1, indicating a completely healthy state; when the feature value deviates from the normal value, the health score decreases according to the Gaussian distribution law, the greater the deviation, the closer the score to 0, and the decline rate is controlled by ;

[0090] After that, the system analyzes the data sequence of the sensor feature value in the recent history time window, evaluates its trend by time series analysis algorithm (such as linear fitting slope calculation or exponential weighted moving average analysis), and calculates a trend penalty factor with a value range of [0, 1] according to the trend , which quantifies the degree and rate of parameter deterioration. Finally, the health score of the sensor is calculated by the formula ; this formula introduces a trend correction mechanism. When the parameter has a deterioration trend > 0, the final health score will be further reduced based on the base score, so as to more sensitively and prospectively reflect the performance degradation of the device; if the parameter is stable or presents an improvement trend = 0, the final health score is equal to the base score and is not affected.

[0091] Step S20, dynamically adjusting the weight of each health score according to the influence degree of the monitored parameter on the device function and the fluctuation level of the historical data, and obtaining a comprehensive health index by weighted fusion, specifically including:

[0092] Step S21, determining the influence degree of the monitored parameter on the device function based on the correlation degree between the monitored parameter and the historical failure record of the device;

[0093] Step S22, evaluating the fluctuation level of the sensor data according to the fluctuation amplitude of the historical data of the sensor deviating from the normal reference value of the sensor;

[0094] Step S23, dynamically calculating and updating the weight coefficient corresponding to each health score according to the determined influence degree and the evaluated fluctuation level;

[0095] Step S24, multiplying each health score by the weight coefficient corresponding to each health score, then superimposing, and normalizing the superimposed result to obtain the comprehensive health index.

[0096] More specifically, step S21, determining the influence degree of the monitored parameter on the device function based on the correlation degree between the monitored parameter and the historical failure record of the device, the method being:

[0097] Step S211, obtaining the historical failure record of the device containing historical failure events and their occurrence time;

[0098] Step S212, extracting the data sequence of each type of monitored parameter within a preset time window before each historical failure event;

[0099] Step S213, using statistical correlation analysis method to calculate the correlation strength between the data sequence of each type of monitored parameter and the occurrence of each type of historical failure event;

[0100] Step S214, the calculated correlation strength is normalized and mapped to the impact degree value of the corresponding monitoring parameter on the equipment function.

[0101] In this embodiment, the device historical failure records containing the historical failure event type, occurrence time, affected subsystem and failure severity are obtained, which are usually derived from the work order log and event report database in the maintenance management system. For each historical failure record, the system extracts the high-frequency or characteristic data sequence of various monitoring parameters, such as vibration, pressure, temperature, etc., within a predetermined time window, e.g. 24 hours or 72 hours, before the failure occurs. The specific window length can be pre-set according to the equipment operation characteristics and failure mode.

[0102] Statistical correlation analysis methods, such as calculating Pearson correlation coefficient, Spearman rank correlation coefficient or mutual information-based correlation measure, are used to quantify the correlation strength between the data sequence of each type of monitoring parameter, such as the change sequence of vibration RMS value, and the occurrence of a specific type of historical failure event. This process usually needs to be performed separately for different failure types to capture the specific relationship between the parameters and the failure. Finally, the calculated correlation strength values between each type of monitoring parameter and each type of failure are normalized, e.g. using min-max normalization or Softmax function, and mapped to a scalar value between 0 and 1. This value is used as a quantitative indicator of the impact degree of the monitoring parameter on the overall function of the equipment. The higher the impact degree value, the stronger the correlation between the parameter and the equipment failure, and the more important it is in the subsequent health index fusion.

[0103] More specifically, step S22, according to the fluctuation amplitude of the sensor historical data deviating from the normal reference value of the sensor, the fluctuation level of the sensor data is evaluated, and the method is as follows:

[0104] Step S221, obtain the data sequence output by the sensor within a predetermined historical period;

[0105] Step S222, read the normal working interval reference value of the sensor pre-calculated by the historical data of the normal operation state of the equipment;

[0106] Step S223, calculate the amplitude of each data point in the data sequence deviating from the normal working interval reference value, and calculate the overall fluctuation quantitative indicator based on the amplitude;

[0107] Step S224, compare the calculated overall fluctuation quantitative indicator with the pre-set fluctuation threshold, and determine the fluctuation level of the sensor data according to the comparison result.

[0108] In this embodiment, a monitoring data sequence continuously output by the sensor in a preset historical period (e.g., the last 30 natural days) is obtained, which should contain enough data points to reflect its long-term fluctuation characteristics. A normal working interval reference value is read, which is obtained by previously statistically analyzing historical data of the sensor in a normal running state of the device verified by the device, and the reference value usually includes a central tendency indicator (such as mean μ) and a dispersion indicator (such as standard deviation σ). The magnitude of each data point in the data sequence deviating from the normal working interval reference value is calculated, which can be calculated as the absolute value |X i - μ| of the difference between each data point and the reference mean value; then, an overall fluctuation quantification indicator is calculated based on all these magnitude values, which can be the average of these magnitudes (mean absolute deviation), standard deviation, or coefficient of variation (ratio of standard deviation to mean value), to comprehensively measure the dispersion and volatility of the data sequence. Finally, the overall fluctuation quantification indicator calculated is compared with a preset fluctuation threshold (for example, a multiple of the standard deviation σ calculated based on historical normal data can be set as the threshold), and the fluctuation level grade of the sensor data is determined according to the comparison result, for example, the fluctuation indicator below 1σ can be rated as "low fluctuation level", between 1σ and 2σ as "moderate fluctuation level", and above 2σ as "high fluctuation level".

[0109] More specifically, in step S23, the weight coefficients corresponding to each health score are dynamically calculated and updated according to the determined influence degree and the evaluated fluctuation level, and the method is as follows:

[0110] In step S231, the influence degree value corresponding to each monitoring parameter is converted into a relevance coefficient;

[0111] In step S232, the fluctuation level grade corresponding to each sensor data is converted into a stability coefficient;

[0112] In step S233, the relevance coefficient and the stability coefficient are linearly combined in a preset proportion to generate an initial weight factor;

[0113] In step S234, the initial weight factor is normalized so that the sum of the weight coefficients of all monitoring parameters is 1, and the normalized result is taken as the latest weight coefficient corresponding to each health score in the weighted fusion.

[0114] In this embodiment, the influence degree value (usually ranging from 0 to 1) of each monitoring parameter determined according to its relevance to the failure of the device function is converted into a relevance coefficient by a preset mapping function, which is usually designed as a linear or nonlinear (such as exponential) positive correlation to ensure that a higher influence degree value is converted into a larger relevance coefficient, highlighting the importance of key parameters.

[0115] Meanwhile, each sensor data is converted into a quantified stability coefficient according to its fluctuation level evaluation result (such as "high", "medium", "low" levels), which can be based on a preset mapping rule, for example, "low fluctuation level" is mapped to a higher stability coefficient (such as 0.9), "medium fluctuation level" is mapped to a medium coefficient (such as 0.6), and "high fluctuation level" is mapped to a lower coefficient (such as 0.3), to reflect the negative impact of data reliability on weight allocation. Then, the correlation coefficient and the stability coefficient are linearly combined according to a preset ratio (for example, 7:3 or other ratios adjusted according to domain knowledge) to generate the initial weight factor of the monitoring parameter, and the calculation formula can be expressed as: initial weight factor = a x correlation coefficient + b x stability coefficient, where a and b are preset weighting coefficients, and a + b = 1. Finally, the initial weight factors calculated for all monitoring parameters are normalized, usually by dividing each initial weight factor by the sum of all initial weight factors, so that the sum of the weight coefficients of all monitoring parameters is strictly equal to 1, and this normalized result is used as the latest weight coefficient of each sensor health score when calculating the comprehensive health index by weighted fusion, and this process can be dynamically executed according to the set period or trigger condition to realize adaptive update of the weight coefficient.

[0116] Step S24, multiply each health score by the weight coefficient corresponding to each health score, then superimpose and normalize the superimposed result to obtain the comprehensive health index.

[0117] In this embodiment, the health scores S i corresponding to each sensor are multiplied by the latest dynamic weight coefficient w i corresponding to each health score to obtain the weighted health score contribution value; then, the weighted contribution values of all sensors are superimposed and summed to obtain an unnormalized weighted sum; finally, the weighted sum is divided by the sum of all weight coefficients, since the weight coefficients have been normalized to make their sum equal to 1, this step can be mathematically simplified to directly use the weighted sum, but to maintain the robustness and universality of the algorithm, it is good practice to explicitly normalize, that is, to calculate the final device comprehensive health index HI by the formula HI = (Σ(w i x S i )) / (Σw i ), which is a continuous value between 0 and 1, intuitively reflecting the overall health status of the device, with a value of 1 indicating complete health and a value of 0 indicating complete failure.

[0118] Specific examples are as follows:

[0119] Step S21, determine the impact degree of monitoring parameters: assume that the historical failure record analysis shows that pressure abnormalities are strongly associated with main pump failures, with a correlation strength of 0.85; temperature abnormalities are moderately associated with seal aging, with a correlation strength of 0.60; and vibration abnormalities are weakly associated with bearing wear, with a correlation strength of 0.40. Through normalization processing (for example, using the Softmax function), the correlation strength is converted into an impact degree value:

[0120] The impact degree value of the pressure sensor: 0.85 / (0.85+0.60+0.40)≈0.46;

[0121] The impact degree value of the temperature sensor: 0.60 / (0.85+0.60+0.40)≈0.32;

[0122] The impact degree value of the vibration sensor: 0.40 / (0.85+0.60+0.40)≈0.22;

[0123] Step S22: evaluate the fluctuation level of sensor data: assume that the fluctuation quantification index (for example, using the coefficient of variation) is calculated according to recent historical data and compared with the threshold value:

[0124] The fluctuation index of the pressure sensor data is 0.08 (<0.1), rated as "low fluctuation level";

[0125] The fluctuation index of the temperature sensor data is 0.15 (between 0.1 and 0.2), rated as "moderate fluctuation level";

[0126] The fluctuation index of the vibration sensor data is 0.25 (0.2), rated as "high fluctuation level".

[0127] Convert to stability coefficient (preset mapping: low=0.9, medium=0.6, high=0.3):

[0128] The stability coefficient of the pressure sensor: 0.9, the stability coefficient of the temperature sensor: 0.6, the stability coefficient of the vibration sensor: 0.3.

[0129] Step S23: dynamically calculate the weight coefficient: convert the impact degree value into the correlation coefficient (here, the impact degree value is directly used as the coefficient), linearly combine according to the preset proportion (correlation: stability=7:3) to generate the initial weight factor:

[0130] The initial weight factor of the pressure sensor = 0.7x0.46+0.3x0.9=0.322+0.27=0.592;

[0131] The initial weight factor of the temperature sensor = 0.7x0.32+0.3x0.6=0.224+0.18=0.404;

[0132] Vibration sensor initial weight factor = 0.7 x 0.22 + 0.3 x 0.3 = 0.154 + 0.09 = 0.244;

[0133] Normalization of initial weight factors (sum = 0.592 + 0.404 + 0.244 = 1.24):

[0134] Pressure sensor latest weight w1 = 0.592 / 1.24 ≈ 0.477;

[0135] Temperature sensor latest weight w2 = 0.404 / 1.24 ≈ 0.326;

[0136] Vibration sensor latest weight w3 = 0.244 / 1.24 ≈ 0.197;

[0137] Step S24: Calculate the comprehensive health index: use the trend-corrected health scores (S1 = 0.095, S2 = 0.331, S3 = 0.095) and the latest weight coefficients to weight and fuse:

[0138] HI = (0.477 x 0.095 + 0.326 x 0.331 + 0.197 x 0.095) / (0.477 + 0.326 + 0.197) = (0.0453 + 0.1079 + 0.0187) / 1.0 = 0.1719.

[0139] Step S30, based on historical health data and the current comprehensive health index, construct a device degradation trajectory model to predict the remaining useful life of the device and the future health state change trend, the method is:

[0140] Step S31, obtain a historical health data sequence containing historical timestamps and their corresponding comprehensive health indexes;

[0141] Step S32, use regression analysis method, take time as independent variable, take comprehensive health index as dependent variable, fit to obtain a degradation trajectory model representing the overall performance degradation law of the device, the degradation trajectory model contains linear and nonlinear degradation terms related to time;

[0142] Step S33, identify and extract external stress factors affecting device degradation, introduce the external stress factors as covariates into the degradation trajectory model to quantify the influence of different stress conditions on the device degradation rate;

[0143] Step S34, based on the latest fitted degradation trajectory model, extrapolate to calculate the predicted time required for the comprehensive health index to be below the preset critical health threshold, and determine the predicted time as the remaining useful life of the device;

[0144] Step S35, calculate the predicted health index of the future specific time point through the degradation trajectory model, and generate a prediction curve for characterizing the future health state change trend of the equipment.

[0145] In this embodiment, the historical health data sequence containing historical timestamps and corresponding comprehensive health indexes is obtained, which is usually stored in the system database in the form of time series and covers a long enough running period to capture the degradation mode of the equipment. A regression analysis method is adopted to take time as the independent variable and the comprehensive health index as the dependent variable, and a degradation trajectory model containing linear, quadratic or higher order nonlinear terms is fitted to characterize the degradation law of the overall performance of the equipment. The general form of the model can be expressed as: HI(t)=β0+β1×t+β2×t 2 +…+ε(t), where β0, β1, β2, etc. are model coefficients fitted by least squares or maximum likelihood estimation, and ε(t) is a random error term for capturing unexplained fluctuations in the model.

[0146] Identify and extract external stress factors that affect the degradation of the equipment, such as environmental temperature, operating load, vibration intensity, etc. These stress factors are introduced into the above degradation trajectory model as covariates in a linear or nonlinear form, for example, the model is expanded to: HI(t)=β0+β1×t+β2×t 2 +γ1×S1(t)+γ2×S2(t)+…+ε(t), where S1(t), S2(t) represent the sequences of different stress factors changing with time, and γ1, γ1, etc. are their corresponding influence coefficients, thereby quantifying the acceleration or deceleration effect of different stress conditions on the degradation rate of the equipment. Based on the latest fitted degradation trajectory model containing stress factors, the predicted time T critical required for the comprehensive health index HI(t) to drop to the preset critical health threshold (for example, HI failure =0.2) starting from the current time is calculated in an extrapolation manner, and this predicted time is determined as the remaining useful life (RUL) of the equipment, i.e. RUL=T failure -t current In addition, by substituting the future time points into the degradation trajectory model, a series of predicted health indexes of the future specific time points can be calculated, and then a continuous prediction curve for intuitively characterizing the future health state change trend of the equipment is generated, which can provide a forward-looking basis for maintenance decisions.

[0147] Step S40, combine preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost to build a multi-dimensional cost function, and dynamically calculate the optimal maintenance time under the conditions of meeting health constraints, time window constraints and resource constraints;

[0148] In this embodiment, a multi-dimensional cost function is constructed by combining preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost, and the method is as follows:

[0149] Step S41, a preventive maintenance cost model is constructed to represent the basic cost required for performing preventive maintenance activities, and contains an incremental cost item that grows with the delay of maintenance opportunity, which is used to reflect the increase in maintenance complexity caused by the potential deterioration of the device state;

[0150] Step S42, a failure risk cost model is constructed by multiplying the total loss cost caused by a single failure event with the probability of failure before a certain maintenance opportunity predicted based on the device health trajectory;

[0151] Step S43, a training interruption cost model is constructed by multiplying the loss cost caused by each hour of training interruption, the probability of conflict between the planned maintenance time window and the intended training plan, and the expected maintenance duration;

[0152] Step S44, a spare parts preparation cost model is constructed by aggregating the unit price, quantity of each type of required spare parts, and a preparation time factor function related to the lead time of spare parts procurement or allocation;

[0153] Step S45, the outputs of the preventive maintenance cost model, the failure risk cost model, the training interruption cost model and the spare parts preparation cost model are summed up to generate a total cost function for evaluating the economy of different maintenance opportunities as a multi-dimensional cost function.

[0154] For step S41, the construction of the preventive maintenance cost model is based on statistical analysis of historical maintenance data and cost composition analysis of actual maintenance jobs. The model aims to quantify the total cost generated by performing preventive maintenance at different time points, which consists of two parts: one is the relatively fixed basic maintenance cost, and the other is the incremental cost that grows dynamically with the delay of maintenance opportunity. The basic preventive maintenance cost The cost data of the same type of maintenance work order in the historical database is statistically averaged, covering the labor time cost, conventional consumable cost and basic management cost required under the standard operation process. The incremental cost item is used to depict the additional cost caused by the further deterioration of the device state due to the delay of maintenance, which is modeled by a linear function proportional to the delay time, specifically represented as 0.2×(Δt / 365)× where At represents the delay days from the current time to the scheduled maintenance time; the coefficient 0.2 means that if the maintenance is delayed for a whole year (365 days), the incremental cost will reach 20% of the base cost, which reflects that with the potential degradation of the equipment health state, the maintenance operation can require more complex troubleshooting, longer operation time, more expensive replacement parts, or more stringent test calibration, thus leading to cost increase. Therefore, the complete preventive maintenance cost model is defined as: This model can dynamically reflect the impact of maintenance timing selection on economic cost, providing key economic inputs for subsequent maintenance decision optimization.

[0155] For step S42, the failure risk cost model is used to quantify the expected economic loss due to potential equipment failure, which is based on the combination of the total loss cost caused by a single failure event and the probability of the failure occurring before a certain future maintenance timing. The total loss cost caused by a single failure event is first defined as This cost is a comprehensive economic indicator obtained by statistical analysis of the direct and indirect costs caused by historical failure events; the direct costs include the labor cost of emergency repair, the spare parts cost of failure replacement, and the additional material cost that can be generated, and the indirect costs include the penalty for cancellation of training tasks due to failure, the opportunity cost of equipment unplanned downtime, and the economic value converted from the overall negative impact on the airline training plan schedule.

[0156] Subsequently, based on the equipment degradation trajectory model constructed in step S30 and its prediction results, the probability of equipment failure before a certain preset maintenance timing t from the current time is calculated as . The calculation of this probability relies on the predicted trajectory of the equipment health index HI(t), which is specifically defined as the probability that the equipment health index first falls below a preset critical health threshold within the time interval [t0, t]. This probability value can be obtained by Monte Carlo simulation of the health index prediction trajectory, or by solving the probability distribution of its first crossing of the failure threshold through an analytical method.

[0157] Finally, the failure risk cost at the maintenance timing t is obtained by multiplying the total cost of a single failure and the calculated failure probability , that is: This model dynamically converts the prediction uncertainty of the equipment health state into an economic risk measure, enabling the maintenance decision optimization process to explicitly weigh the pros and cons between early maintenance to avoid risks and delayed maintenance to save costs.

[0158] For step S43, the training interruption cost model is obtained by multiplying the loss cost per hour of training interruption , the probability of the planned maintenance time window conflicting with the scheduled training plan , and the expected maintenance duration , whose mathematical model is . Wherein, the loss cost per hour of training interruption is a constant derived from historical operation data, covering the comprehensive economic cost of idle students, standby instructors, simulation resource occupation, and possible subsequent course delays. The conflict probability is calculated by comparing the proposed maintenance time window t with the pre-scheduled flight training calendar, and its value is between 0 and 1;

[0159] If the maintenance window completely overlaps with any training task, if it partially overlaps, it is calculated according to the proportion of time overlap, and if there is no overlap, it is 0. The expected maintenance duration is represented by the maintenance urgency score , which is determined by a piecewise function mapping, which defines the basic maintenance duration and its growth slope corresponding to different urgency levels.

[0160] The maintenance urgency score itself is a comprehensive index that quantifies the degree of performance degradation of a specific subsystem (such as the hydraulic system), and its calculation relies on three core performance parameters and their benchmark values and weights. The first parameter is the pressure response delay , which refers to the time delay from the issuance of the control instruction signal x ( n ) to the response of the hydraulic system pressure signal y ( n ) reaching the expected value. This value is calculated by the cross-correlation function , where y (n+τ) represents the pressure value of the pressure response signal y at the sampling point with time index n+τ. Its physical meaning is to find the time offset that maximizes the correlation between the control signal and the response signal τ , and then multiply it by the sampling interval T s to get the actual time delay, i.e. . The second parameter is the pressure stability index , which is defined as the ratio of the standard deviation of the pressure value to the mean value , i.e. , which reflects the volatility of the system pressure, and the larger the value, the more unstable the system, where represents the pressure value at the i-th sampling point. and The two parameters have the same meaning; both represent the average value of the pressure signal across N sampling points. The third parameter is the leakage rate. It is the rate of pressure decrease per unit time when the system is in steady state, and the calculation formula is: ,in and These are the pressure values ​​at the beginning and end of the steady-state observation phase, respectively. and These are the times corresponding to the pressure values ​​at the beginning and end of the steady-state observation phase.

[0161] These three performance parameters are used in the calculation At that time, it will be compared with a predefined benchmark value and normalized. The benchmark value for response latency... Set to 0.5 seconds, pressure volatility benchmark value S ref Set to 0.05 (i.e., 5%), the baseline value for leakage rate. R ref The pressure is set to 0.1 bar / min. The normalized results of each parameter are then multiplied by their respective weighting coefficients (response delay weight w1 = 0.4, pressure fluctuation weight w2 = 0.3, leakage rate weight w3 = 0.3) according to their importance, and then summed to obtain the final comprehensive urgency score. This score is ultimately used in the aforementioned piecewise function to dynamically determine the estimated duration of the maintenance task. This provides accurate input to the cost model, ensuring that the maintenance timing optimization algorithm can comprehensively consider equipment health status and operational impact to make the optimal decision.

[0162] get Then, the maintenance duration can be determined using the following piecewise function. :

[0163]

[0164] The cost model is ultimately integrated into the total cost function, which uses optimization algorithms to find the optimal maintenance timing that minimizes the total cost, thereby minimizing the disruption of maintenance activities to training operations while ensuring equipment reliability.

[0165] In step S44, the spare parts preparation cost model The model is constructed by summarizing the unit price of various required spare parts, their predicted quantities, and a nonlinear preparation time factor function related to the lead time of spare parts procurement or allocation. Its core mathematical model is... In this model, Indicates the first iThe unit procurement cost or inventory holding cost of a spare part, which is derived from the enterprise's spare part master data and procurement contract; representing the predicted maintenance time window t in which the maintenance task is performed, the predicted consumption quantity of the i spare part, which is generated by the maintenance decision optimizer in the intelligent decision layer based on the equipment health state evaluation result, the spare part replacement records of the same type of task in the historical maintenance data, and the specific task content specified by the predictive maintenance work order; is a key time variable, representing the lead time required for the i spare part to initiate a procurement application or be allocated from the central warehouse to the actual arrival at the maintenance site from the current time. The lead time is a dynamic value, the length of which depends on various factors such as the inventory status of the spare part (whether it is within the local safety inventory range), the current supply capability of the supplier, the logistics transportation mode, etc. is a preparation time factor function, which is a coefficient function used to quantify the additional cost impact due to the urgency of the spare part preparation. Its function form is usually determined based on historical logistics and procurement data, for example, it can be defined as a piecewise function , where is the standard planned procurement lead time of this type of spare part. When the actual required lead time is shorter than the standard period, it means that urgent fees need to be paid or higher-cost logistics methods need to be used, so a coefficient k greater than 1 is used to amplify the spare part cost, so as to truly reflect the spare part supply cost due to the urgency of the maintenance decision in the total cost. This spare part preparation cost is finally an important part of the total cost function, which is evaluated by the optimization algorithm together with the preventive maintenance cost, the failure risk cost and the training interruption cost, to ensure that the selected optimal maintenance time Topt is not only technically feasible, but also economically efficient at the level of the logistics supply chain.

[0166] Step S45, summing the outputs of the preventive maintenance cost model, the failure risk cost model, the training interruption cost model and the spare part preparation cost model to generate a total cost function for evaluating the economy of different maintenance times, as a multi-dimensional cost function.

[0167] In this embodiment, the total cost function is defined as:

[0168] .

[0169] Under the conditions of meeting the health constraints, time window constraints and resource constraints, the optimal maintenance time is dynamically calculated, and the method is:

[0170] The health constraint is defined as the predicted health index not falling below a preset health index safety threshold at the time of maintenance.

[0171] Define the time window constraint as follows: the selected maintenance timing must belong to the set of available time windows allowed by the system, which is jointly determined by the training plan gap and the availability of the maintenance team.

[0172] The resource constraint is defined as the total amount of all types of resources required to perform the maintenance task, which shall not exceed the total amount of real-time available resources within the time window in which the maintenance occurs;

[0173] Among all feasible maintenance opportunities that simultaneously satisfy the health constraint, the time window constraint, and the resource constraint, calculate the total cost function value corresponding to each feasible maintenance opportunity.

[0174] The feasible maintenance time that minimizes the total cost function value is selected as the final optimal maintenance time.

[0175] In this embodiment, the predicted health index for each time point in the future, generated by the degradation prediction module, is first obtained from the maintenance decision optimizer. HI ( t The sequence is compared with a preset health index safety threshold. HI critical The comparison is performed, and the health constraint is defined as the maintenance timing of the candidate. t Its predicted health index must not be lower than this safety threshold, that is HI ( t )≥ HI critical This constraint ensures that the selected timing is not later than the deadline allowed by the device's health status.

[0176] Meanwhile, the system obtains future training schedules and calendars in real time from the training system interface in the application service layer, and retrieves the maintenance team's work schedule from the resource management database. By calculating the intersection of training gaps and the available time of maintenance personnel, it dynamically generates a set of available time windows allowed by the system. Available _ windows The time window constraint is defined as the selected maintenance timing. t Must belong to this set, that is t ∈ Available _ windows This ensures that maintenance work will not conflict with important training tasks and that there is sufficient manpower to carry out the work;

[0177] In addition, the system also needs to verify resource constraints, namely the total amount of all resources required to execute this predictive maintenance task (including the number of technicians for specific job types). r personnel Special tools and equipmentr tool and various spare parts r part , not exceeding the maintenance opportunity t total amount of real-time available resources within the time window R available(t) , the data is dynamically obtained from the enterprise's resource management system, and the constraint condition is expressed as ∑ r i ≤ R available ( t ), wherein, r i represents the i-th type of resource required to perform the maintenance task; after defining the above three constraints, the system initializes an empty feasible solution set, and traverses all candidate time windows, for each window t , checks whether it meets the health constraint, time window constraint and resource constraint in turn, and adds the window t that passes all constraint checks to the feasible maintenance opportunity set W ;

[0178] Subsequently, for each feasible maintenance opportunity W in the set t , the system calls the total cost function to calculate the preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost corresponding to it respectively, and sums them up to obtain the total cost value under the opportunity;

[0179] Finally, the system performs an optimization search among all feasible solutions, selects the feasible maintenance opportunity with the minimum total cost function value, and mathematically expresses it as Topt =argmin t∈W C total ( t ), and determines the opportunity Topt as the final optimal maintenance opportunity output to the application service layer, for automatically generating predictive maintenance work orders and scheduling plans, thereby realizing the automation of maintenance decision-making under multiple real constraints.

[0180] Step S50, automatically generating predictive maintenance work orders according to the optimal maintenance opportunity and the comprehensive health index, coordinating the maintenance work orders with the training plan system to realize predictive maintenance and intelligent scheduling, the method being:

[0181] Step S51, determining the date and time window of the planned execution based on the optimal maintenance opportunity;

[0182] Step S52: Based on the comprehensive health index and the predicted trend of health status change, identify the target subsystem that needs to be maintained, and generate a specific maintenance task description for the target subsystem.

[0183] Step S53: Based on the maintenance requirements of the target subsystem, automatically generate the required spare parts list, tool and equipment requirements, and technical personnel skill requirements;

[0184] Step S54: Based on the planned execution time, maintenance task description and resource requirements, automatically assemble and generate a structured predictive maintenance work order;

[0185] Step S55: The predictive maintenance work order is sent to the training plan system to perform conflict detection and automatic negotiation with the established training plan, and finally a specific execution time window is determined near the optimal maintenance time without affecting the normal training task.

[0186] Step S56: Update the final execution time window after determination to the predictive maintenance work order to complete the collaborative scheduling of the work order and the training plan.

[0187] During implementation, the system first determines the optimal maintenance timing based on the output of the optimization algorithm. Topt Determine the planned execution date and specific time window, which is selected from the set of available time windows. Available _ windows Selected from, and Topt The system identifies the closest consecutive time period that satisfies all constraints. Subsequently, the system bases its analysis on the current comprehensive health index. HI current and the future health status change trend generated by the degradation prediction module HI future ( t ), perform health assessments and failure risk predictions for all subsystems; for each subsystem j When its health index HI j Below a preset threshold (e.g., 0.7) or its failure risk probability P failure,j When the risk level exceeds a preset risk threshold (e.g., 0.3), the system will identify it as a target subsystem requiring maintenance and generate operational instructions containing specific problem descriptions and maintenance measures based on its specific performance anomalies (e.g., excessive pressure fluctuation in hydraulic systems, excessive response delay, etc.). Next, based on the identified target subsystem and its maintenance requirements, the system automatically queries the knowledge base and historical work order data to generate a list of required spare parts (including spare part models and quantities). Q i ( t) and the skill level requirement of the required technical personnel. Then, the system synthesizes all the above information, including the planned execution time, the maintenance task description, the resource requirement, and the health improvement expectation HI expected assembles and generates a complete predictive maintenance work order automatically according to a predefined structured template, the work order number follows the rule of "PM-{simulator ID}-{timestamp}" and is labeled as "PREDICTIVE" and the calculated priority. After the work order is generated, the system sends the work order to the training plan system through the application service layer and the standardized interface (such as API) of the training plan system, triggering an automatic negotiation process; the training plan system detects conflicts based on the received work order time window information with the established training plan, and if there is a conflict, it will search for an idle and resource available time window within a preset time range (for example, 3 days before and after) and feedback a suggested new time window through negotiation logic. Finally, the maintenance decision system receives the feedback, updates the final execution time window to the predictive maintenance work order, and reconfirms the resource availability, thereby completing the collaborative scheduling of the work order and the training plan and ensuring that the maintenance task is executed under the premise of minimizing the training interference. Topt

[0188] Although the above embodiment describes each step in the above order, those skilled in the art can understand that, in order to achieve the effect of the embodiment, the different steps do not have to be executed in such an order, and they can be executed simultaneously (in parallel) or in a reversed order, and these simple changes are within the protection scope of the present application.

[0189] The second embodiment of the present application provides a flight simulator predictive maintenance scheduling system based on multi-source data fusion and degradation prediction, which is used to implement the flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction in the first embodiment. The system comprises:

[0190] a health score calculation module configured to collect multi-source monitoring data in real time through a plurality of sensors deployed on the flight simulator, process the multi-source monitoring data, extract feature information corresponding to the health state, and map each feature information to a corresponding health score based on a preset health mapping function;

[0191] a comprehensive health index calculation module configured to dynamically adjust the weight of each health score according to the influence of the monitored parameters on the device function and the historical data fluctuation level, and obtain a comprehensive health index through weighted fusion;

[0192] ​a prediction module configured to construct a device degradation trajectory model based on historical health data and the current comprehensive health index, to predict the remaining useful life of the device and the future health state change trend;

[0193] a maintenance opportunity calculation module configured to construct a multi-dimensional cost function by combining preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost, and to dynamically calculate an optimal maintenance opportunity under the condition of meeting health constraints, time window constraints and resource constraints;

[0194] a maintenance module configured to automatically generate a predictive maintenance work order according to the optimal maintenance opportunity and the comprehensive health index, to coordinate and schedule the maintenance work order and the training plan system, and to realize predictive maintenance and intelligent scheduling.

[0195] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related description of the system described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0196] It should be noted that the flight simulator predictive maintenance scheduling system based on multi-source data fusion and degradation prediction provided in the above embodiments is only exemplified by the division of the above functional modules, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the modules or steps in the embodiments of the present application can be further decomposed or combined, for example, the modules of the above embodiments can be combined into one module, or can be further split into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present application are only for distinguishing the modules and steps, and should not be considered as an improper limitation of the present application.

[0197] The electronic device of the third embodiment of the present application comprises:

[0198] at least one processor; and

[0199] a memory in communication connection with the at least one processor; wherein

[0200] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to realize the flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction.

[0201] The computer readable storage medium of the fourth embodiment of the present application stores computer instructions, and the computer instructions are used to be executed by the computer to realize the flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction.

[0202] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the storage device and the processing device and the related descriptions described above can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0203] Those skilled in the art will realize that the modules, method steps of the examples described in connection with the embodiments disclosed herein can be realized in electronic hardware, computer software, or any combination thereof. The software modules, method steps corresponding to the software modules can be stored in memory devices such as random access memory (RAM), read-only memory (ROM), a hard disk drive, a solid state drive, an optical disk, or any other form of storage medium known in the art. For clarity, the above description has generally been given in terms of the functions performed by the various components and steps. Whether the functions are performed in hardware or software depends on the particular application and design constraints imposed on the overall system. Those skilled in the art can use various methods to implement the functions described above for each particular application, but such implementation should not be considered to be beyond the scope of the present application.

[0204] The terms "first", "second", and the like, are used to distinguish between similar objects, rather than to denote a particular order or sequence.

[0205] The term "comprising" or any other similar term is intended to encompass the inclusion of non-exclusive elements, so that a process, method, article, or device / apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to the process, method, article, or device / apparatus.

[0206] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will all fall within the protection scope of the present application.

Claims

1. A flight simulator predictive maintenance scheduling method based on multi-source data fusion and degradation prediction, characterized in that, The method comprises: Real-time acquisition of multi-source monitoring data through a plurality of sensors deployed on the flight simulator, processing of the multi-source monitoring data, and extraction of feature information corresponding to the health state, mapping of each of the feature information to a corresponding health score based on a preset health mapping function; Dynamic adjustment of the weight of each health score according to the influence degree of the monitored parameter of the sensor on the function of the equipment and the historical data fluctuation level, and obtaining of a comprehensive health index through weighted fusion; Based on historical health data and the comprehensive health index, a device degradation trajectory model is constructed to predict the remaining useful life of the device and the future health state change trend; A multi-dimensional cost function is constructed by combining preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost, and the optimal maintenance opportunity is dynamically calculated under the conditions of meeting health constraints, time window constraints and resource constraints; According to the optimal maintenance opportunity and the comprehensive health index, a predictive maintenance work order is automatically generated, and the maintenance work order is coordinated and scheduled with a training plan system to realize predictive maintenance and intelligent scheduling.

2. The method of claim 1, wherein, The method for processing the multi-source monitoring data and extracting the feature information corresponding to the health state is: Data cleaning and quality evaluation are performed on the multi-source monitoring data to eliminate abnormal values and noise; Multi-sampling rate data synchronization processing is performed on the cleaned data to unify the monitoring data of different sources to the same time reference; Time domain features, frequency domain features and time-frequency domain features are extracted from the synchronized data to comprehensively represent the health state of the equipment.

3. The method of claim 1, wherein, The method for mapping each of the feature information to a corresponding health score based on a preset health mapping function is: For each feature information extracted from the monitoring data, a Gaussian function is used as the health mapping function to map the current value of the feature information to an initial health score between 0 and 1, wherein when the current value is equal to the normal reference value of the feature information obtained through historical data statistics in advance, the initial health score obtained by mapping is 1, indicating a completely healthy state; A data sequence of the feature information within a recent historical time window is obtained, the deterioration trend strength of the feature information is calculated through a trend analysis algorithm, and a trend penalty factor between 0 and 1 is generated; The initial health score is multiplied by the difference between 1 and the trend penalty factor to obtain a final health score after correction, wherein when there is a deterioration trend, the trend penalty factor is greater than zero, so that the final health score is lower than the initial health score.

4. The method of claim 1, wherein, The method for dynamically adjusting the weight of each health score according to the influence degree of the monitored parameter of the sensor on the function of the equipment and the historical data fluctuation level, and obtaining the comprehensive health index through weighted fusion is: Based on the correlation degree between the monitoring parameters and the historical failure records of the equipment, the influence degree of the monitoring parameters on the function of the equipment is determined; According to the fluctuation amplitude of the historical data of the sensor deviating from the normal reference value of the sensor, the fluctuation level of the sensor data is evaluated; According to the determined influence degree and the evaluated fluctuation level, the weight coefficient corresponding to each health score is dynamically calculated and updated; The health scores are multiplied by the weight coefficients corresponding to the health scores, superimposed, and normalized to obtain the comprehensive health index.

5. The method of claim 4, wherein, Based on the correlation between the monitoring parameters and the historical failure records of the equipment, the influence degree of the monitoring parameters on the equipment function is determined, and the method is as follows: Obtain the historical failure records of the equipment containing historical failure events and their occurrence time; Extract the data sequence of each type of monitoring parameter within a preset time window before each historical failure event occurs; Using statistical correlation analysis method, the correlation strength between the data sequence of each type of monitoring parameter and the occurrence of each type of historical failure event is calculated respectively; The calculated correlation strength is normalized and mapped to the influence degree value of the corresponding monitoring parameter on the equipment function.

6. The method of claim 5, wherein, According to the fluctuation amplitude of the sensor historical data deviating from the normal reference value of the sensor, the fluctuation level of the sensor data is evaluated, and the method is as follows: Obtain the data sequence output by the sensor within a preset historical period; Read the normal working interval reference value of the sensor obtained by statistical analysis of the historical data of the normal running state of the equipment; Calculate the amplitude of each data point in the data sequence deviating from the normal working interval reference value, and calculate the overall fluctuation quantitative index based on the amplitude; Compare the calculated overall fluctuation quantitative index with the preset fluctuation threshold, and determine the fluctuation level grade of the sensor data according to the comparison result.

7. The method of claim 6, wherein, According to the determined influence degree and the evaluated fluctuation level, the weight coefficients corresponding to each health score are dynamically calculated and updated, and the method is as follows: Convert the influence degree value corresponding to each monitoring parameter into a correlation coefficient; Convert the fluctuation level grade corresponding to each sensor data into a stability coefficient; Linearly combine the correlation coefficient and the stability coefficient according to a preset proportion to generate an initial weight factor; The initial weight factor is normalized so that the sum of the weight coefficients of all monitoring parameters is 1, and the normalized result is used as the latest weight coefficient corresponding to each health score in weighted fusion.

8. The method of claim 1, wherein, Based on the historical health data and the current comprehensive health index, a device degradation trajectory model is constructed to predict the remaining useful life of the device and the future health state change trend, and the method is as follows: Obtain the historical health data sequence containing historical time stamps and their corresponding comprehensive health indexes; Using regression analysis method, take time as independent variable and comprehensive health index as dependent variable, fit to obtain the degradation trajectory model representing the overall performance degradation law of the equipment, which contains linear and nonlinear degradation terms related to time; Identify and extract external stress factors affecting the degradation of the equipment, and introduce the external stress factors as covariates into the degradation trajectory model to quantify the influence of different stress conditions on the degradation rate of the equipment; Based on the latest fitted degradation trajectory model, extrapolate to calculate the predicted time required for the comprehensive health index to be lower than the preset critical health threshold, and determine the predicted time as the remaining useful life of the equipment; Through the degradation trajectory model, the predicted health index at a future specific time point is calculated to generate a prediction curve for representing the future health state change trend of the equipment.

9. The method of claim 1, wherein, The method comprises the following steps: A multi-dimensional cost function is constructed by combining preventive maintenance cost, failure risk cost, training interruption cost and spare parts preparation cost, which comprises the following steps: A preventive maintenance cost model is constructed to represent the basic cost required for performing preventive maintenance activities, and an incremental cost item is added to reflect the increase in maintenance complexity caused by potential deterioration of the equipment state; A failure risk cost model is constructed by multiplying the total loss cost caused by a single failure event with the probability of failure before a certain maintenance opportunity predicted based on the equipment health trajectory; A training interruption cost model is constructed by multiplying the loss cost caused by each hour of training interruption, the probability of conflict between the planned maintenance time window and the scheduled training plan, and the expected maintenance duration; A spare parts preparation cost model is constructed by aggregating the unit price, quantity of various required spare parts, and a preparation time factor function related to the lead time of spare parts procurement or allocation; 10. The method of claim 1, wherein, The outputs of the preventive maintenance cost model, the failure risk cost model, the training interruption cost model and the spare parts preparation cost model are summed up to generate a total cost function for evaluating the economy of different maintenance opportunities, which is a multi-dimensional cost function. The method for dynamically calculating the optimal maintenance opportunity under the conditions of meeting health constraints, time window constraints and resource constraints comprises the following steps: The health constraint is defined as the predicted health index at the maintenance opportunity being higher than the preset health index safety threshold; The time window constraint is defined as the selected maintenance opportunity must belong to the set of allowed time windows determined by the training plan gaps and the maintenance team availability; The resource constraint is defined as the total amount of various resources required for performing maintenance tasks must not exceed the total amount of real-time available resources within the time window where the maintenance opportunity is located; The total cost function value corresponding to each feasible maintenance opportunity is calculated among all feasible maintenance opportunities that meet the health constraint, the time window constraint and the resource constraint; The feasible maintenance opportunity with the minimum total cost function value is selected as the final optimal maintenance opportunity.

Images