A method and system for managing and optimizing inventory of flight simulation equipment

Abstract

The present application belongs to the field of aviation flight training, modern aviation logistics, general aviation production service, aviation training equipment management and intelligent inventory optimization, and discloses a kind of management and inventory optimization method and system of flight simulation equipment, solve the technical problems existing in the management of current flight simulation equipment management, such as scattered management, inaccurate life prediction, unreasonable inventory scheduling and maintenance plan lag.The present application collects multi-source operation data of flight simulation equipment and constructs a life prediction model that integrates physical degradation law and data-driven learning, and realizes accurate calculation of equipment residual life by combining with the health coefficient after repair;Based on the residual life failure probability, an inventory decision model is constructed with the goal of minimizing total cost, and particle swarm algorithm is used to output inventory strategies such as dynamic reorder point and order quantity;Relying on the prediction error, a hierarchical adaptive correction mechanism is established.The present application realizes the closed-loop intelligentization of equipment management, improves the prediction and scheduling level, reduces the risk of missing parts, and guarantees flight training.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of aviation flight training, modern aviation logistics, general aviation production service, flight simulation equipment management and inventory optimization, and particularly relates to a flight simulation equipment management and inventory optimization method and system. BACKGROUND

[0002] In the coordinated operation of aviation flight training system, modern aviation logistics support system and general aviation production service system, stable supply and scientific maintenance of flight simulation equipment (including core components and supporting aviation materials) and efficient management and reasonable inventory configuration are the key to ensuring the continuity of training tasks. There are many technical defects in the current industry's management and inventory control scheme of flight simulation equipment, as follows:

[0003] Inventory management mode is rigid: the existing technology generally adopts a fixed safety inventory and reordering point control mode, that is, the inventory threshold is set according to historical average consumption data and manual experience, and procurement is triggered when the inventory is below the threshold. This mode does not take into account supply cycle fluctuations, environmental differences and component consumption randomness, which can easily cause two extreme problems: high inventory can cause serious capital occupation, and insufficient inventory can cause parts shortage (AOG), directly interrupting flight training tasks.

[0004] Life and maintenance decision lag: maintenance and component replacement of simulation equipment mostly rely on manual scheduling and periodic maintenance, that is, replacement / repair operations are performed according to preset usage period or training frequency. This method ignores the influence of multiple sources of environment and operation data such as temperature, humidity and load strength, often resulting in waste of resources due to "premature replacement" of equipment, or equipment failure due to "delayed replacement", significantly increasing maintenance costs and operation risks.

[0005] Demand prediction method is single: some systems try to predict component demand through simple statistical methods such as moving average and single variable regression, but such methods cannot handle complex scenarios with multiple factors such as fleet age, training frequency and environmental load, and have insufficient prediction accuracy when faced with high demand volatility and randomness, resulting in a lack of scientificity in procurement and inventory strategies.

[0006] System modules are isolated and have no closed loop: traditional management systems usually only include inventory database, manual maintenance schedule, and basic statistical prediction module, with low data correlation between modules, lagging updates, and maintenance and replacement results unable to feed back to the prediction model in real time. The model remains in the initial setting state for a long time, cannot achieve precision iteration, and has low overall decision efficiency.

[0007] In summary, the current flight simulation equipment management has problems such as decentralized management, inaccurate life prediction, unreasonable inventory scheduling and lagging maintenance planning, which restricts the support capability and safety of flight training tasks. SUMMARY

[0008] In order to solve the above problems in the prior art, the present application provides a flight simulation equipment management and inventory optimization method to solve the problems of scattered management, inaccurate life prediction, unreasonable inventory scheduling and lagging maintenance planning.

[0009] Collecting multi-source operation data of the flight simulation equipment, the multi-source operation data including operation state data and environment data;

[0010] Based on the multi-source operation data, a life prediction model combining physical degradation rules and data-driven learning is used to calculate the prediction results of the flight simulation equipment, the prediction results including life prediction values and probability distribution intervals of residual life;

[0011] Based on the prediction results, an inventory decision model aiming to minimize total cost is constructed and solved to determine a dynamic inventory strategy, the inventory strategy including reorder timing, order quantity and risk level;

[0012] Obtaining maintenance results of the flight simulation equipment in actual operation and calculating prediction errors between the prediction results and the maintenance results;

[0013] According to the prediction errors, the life prediction model and the inventory decision model are adaptively corrected.

[0014] In some preferred embodiments, the life prediction model includes a physical model and a data-driven model, the life prediction model The expression is:

[0015] ;

[0016] Wherein, is a dynamic weight coefficient; is a life prediction value calculated by the physical model; is a life prediction value calculated by the data-driven model.

[0017] In some preferred embodiments, the dynamic weight coefficient is obtained by:

[0018] Obtaining the first average error in the last W times of prediction;

[0019] Obtaining the second average error in the last W times of prediction;

[0020] Based on the inverse relationship between the first average error and the second average error, the dynamic weight coefficient is determined.

[0021] In some preferred embodiments, for the repairable components in the flight simulation equipment, the method further comprises: calculating a post-repair health coefficient based on the multi-dimensional evaluation results of the current repair, and using the post-repair health coefficient to calibrate the prediction results of the repairable components;

[0022] The multi-dimensional evaluation at least includes: a post-repair function test score, a replacement component value ratio, and a repair work order qualification rate.

[0023] In some preferred embodiments, the method of using the post-repair health coefficient to calibrate the prediction results of the repairable components is:

[0024] The post-repair health coefficient is used as a multiplication factor to act on the life prediction value output by the life prediction model to obtain a calibrated life prediction value.

[0025] In some preferred embodiments, the method of constructing and solving an inventory decision model with the objective of minimizing total cost based on the prediction results to determine a dynamic inventory strategy is:

[0026] The probability distribution interval of the remaining life is mapped to the cumulative failure probability of the component within the future order lead time;

[0027] When the cumulative failure probability exceeds a dynamic risk tolerance threshold, the solving of the inventory decision model is triggered;

[0028] The total cost of the inventory decision model includes: inventory holding cost, procurement processing cost, and shortage risk cost.

[0029] In some preferred embodiments, the shortage risk cost is associated with the impact level of the flight training task, and when the length of the shortage exceeds a preset threshold, the shortage risk cost increases in a non-linear manner.

[0030] In some preferred embodiments, the solving of the inventory decision model uses a particle swarm optimization algorithm.

[0031] In some preferred embodiments, the life prediction model and the inventory decision model are adaptively corrected according to the prediction error, and the method is:

[0032] When the prediction error shows a directional bias in statistics, the dynamic weight coefficient of the physical model and the data-driven model in the life prediction model is fine-tuned;

[0033] When the sample proportion of a specific feature sub-domain in the prediction error exceeds a preset concentration threshold, the life prediction model is locally re-calibrated or locally re-trained.

[0034] When the proportion of newly added labeled samples since the last training to the original training set exceeds a preset proportion, or the overall error of the model continuously deteriorates, the life prediction model is retrained in full.

[0035] In the second aspect of the present application, a management and inventory optimization system for flight simulation equipment is provided, based on the management and inventory optimization method for flight simulation equipment, the system comprises:

[0036] A data acquisition module configured to acquire multi-source operation data of the flight simulation equipment, wherein the multi-source operation data comprises operation state data and environment data;

[0037] A life prediction module configured to calculate a prediction result of the flight simulation equipment based on the multi-source operation data by using a life prediction model that fuses physical degradation rules and data-driven learning, wherein the prediction result comprises a life prediction value and a probability distribution interval of the remaining life;

[0038] An inventory decision module configured to construct and solve an inventory decision model aiming to minimize the total cost based on the prediction result, so as to determine a dynamic inventory strategy, comprising: determining the cumulative failure probability of the equipment component within a preset future support time window based on the probability distribution interval of the remaining life;

[0039] Comparing the cumulative failure probability with a dynamic risk tolerance threshold, and triggering the inventory decision module when the cumulative failure probability exceeds the dynamic risk tolerance threshold, wherein the dynamic risk tolerance threshold is automatically adjusted according to the type of the equipment component and the criticality level of the associated flight training task plan;

[0040] The inventory strategy comprises reorder point, order quantity and risk level;

[0041] An error calculation module configured to obtain the maintenance result of the flight simulation equipment in actual operation, and calculate the prediction error between the prediction result and the maintenance result;

[0042] A model correction module configured to adaptively correct the life prediction model and the inventory decision model according to the prediction error.

[0043] The present application has the following advantages:

[0044] 1) The Internet of Things sensors replace manual recording to complete the automatic and real-time collection of equipment operation and environment data, improve management transparency and timeliness, and eliminate data lag problems.

[0045] 2) Hybrid model combining physical mechanism and data-driven, combined with the calibration of the modified health coefficient and the classification of component demand, solves the "too early / too late replacement" problem of traditional manual estimation, reduces maintenance cost and downtime risk.

[0046] 3) Reordering logic based on residual life failure probability and total cost optimization target, dynamically adjusts safety stock and procurement plan, reduces capital occupation, and avoids training interruption caused by lack of parts.

[0047] 4) Real-time feedback of maintenance results to life prediction and inventory model, forming a dynamic closed loop of "prediction - execution - feedback - re-prediction", continuously improving model prediction accuracy and system adaptive ability.

[0048] 5) Through whole life cycle intelligent management, ensure the required equipment for flight training, effectively reduce the lack of parts rate, improve the continuity and safety of training tasks. BRIEF DESCRIPTION OF DRAWINGS

[0049] 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:

[0050] Figure 1 is a step flow chart of a flight simulator equipment management and inventory optimization method of the present application.

[0051] Figure 2 is a flight simulator equipment parts classification flow chart of the present application;

[0052] Figure 3 is a flight simulator equipment parts demand mode classification quadrant diagram of the present application.

[0053] Figure 4 is a flight simulator equipment parts replacement and inventory support simulation flow chart of the present application. DETAILED DESCRIPTION

[0054] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are intended to explain the relevant application, but 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.

[0055] It should be noted that the embodiments and features in the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and examples.

[0056] In order to more clearly describe the flight simulator equipment management and inventory optimization method of the present application, the following will be combined with Figures 1 to 4The steps in the embodiments of the present invention will be described in detail below.

[0057] This invention proposes a management and inventory optimization method for flight simulator equipment. Before carrying out specific management and optimization processes, it is first necessary to classify and define the components of the flight simulator equipment and clarify their demand characteristics and inventory guarantee logic:

[0058] See Figure 2 Based on whether components can be repaired, they are divided into repairable and non-repairable components (also known as repairable parts and non-repairable parts). Based on the number of times they can be repaired, they can be divided into two main categories: reusable parts and consumable parts. Reusable parts refer to those components that are relatively high in value and can be restored to their usability after repair when they malfunction, such as engines and APUs. Although there are not many types of these, they occupy a lot of funds. Consumable parts refer to those components that are relatively low in value and cannot be repaired or have no repair value after malfunction, i.e., single-use parts. The types and quantities of these components are enormous.

[0059] The demand for simulator components is influenced by factors such as fleet age, number of training sessions, and training duration. However, significant randomness exists in aspects like component replacement and repair, making it difficult to predict the frequency of component replacements using a universally effective method. Nevertheless, this paper attempts to classify replacement data to improve efficiency in selecting prediction methods. (See [link to relevant documentation]). Figure 3 The requirements for components of flight simulator equipment are categorized as follows:

[0060] Based on historical replacement data, calculate the average demand interval of components and the square of the coefficient of variation of the replacement data;

[0061] Based on the average demand interval and the magnitude of the squared coefficient of variation, the component demand patterns are divided into at least four categories, and differentiated lifetime prediction model parameters or inventory management strategies are configured for different categories of components.

[0062] Specifically: the average demand interval (ADI) is the average of two consecutive demand periods, and the square of the coefficient of variation is the variance of the quotient of the demand period and the average demand period. Based on these two indicators, the number of replacements is divided into four categories: smooth, intermittent, irregular, and blocky. The smaller the average demand interval, the smaller the square of the coefficient of variation, the more stable the data and the easier it is to predict. Blocky demand is the most difficult to predict. For the above different demand patterns, differentiated configurations are adopted in the subsequent life prediction model parameter initialization, inventory decision model risk tolerance threshold setting, and adaptive correction strategy to improve overall management efficiency. For example, for blocky demand components, the system will automatically adopt a more conservative inventory strategy and a stronger model correction mechanism.

[0063] Based on this, seeFigure 1 A management and inventory optimization method of flight simulation equipment, the method comprising the following steps:

[0064] Collecting multi-source operation data of the flight simulation equipment, the multi-source operation data comprising operation state data and environment data;

[0065] In this embodiment, the operation state data (start-up times, use time length, load power, vibration amplitude, etc.) and the environment data (temperature, humidity, voltage fluctuation, air pressure, etc.) of the flight simulation equipment are automatically collected and real-time transmitted through the Internet of Things sensors, embedded collection terminals and data interface modules. In addition to the above core data, maintenance and replacement data, inventory and procurement data are also synchronously collected, wherein the maintenance and replacement data includes replacement date, fault type, repair result and accessory batch information, and the inventory and procurement data includes inventory quantity, in-transit materials and supplier delivery cycle information. All collected data is uploaded to a cloud database or a local server through a communication gateway to form a unified data access layer. This mode replaces the traditional manual recording mode, realizes the automation and real-time of data collection, and greatly improves the transparency and timeliness of management. At the same time, the features input into the subsequent life prediction model include at least five types of quantifiable physical parameters: component cumulative operating hours (h), average vibration frequency (Hz), shell surface temperature (℃), current working current (A) and environmental humidity (% RH). The above features are real-time acquired by embedded sensors and environment collection modules, laying a data foundation for accurate life prediction.

[0066] In this embodiment, the collected multi-source operation data is also preprocessed and feature extracted;

[0067] In this embodiment, due to the heterogeneity and noise problems of the collected multi-source data, targeted cleaning and standardization operations need to be carried out, specifically including outlier identification and elimination, time alignment and missing value interpolation, unit conversion and data normalization, and establishing a unified feature template according to the type of aviation materials. After completing the basic preprocessing, key influencing factors are extracted through feature engineering, including environmental load index, operation intensity coefficient and average training time length. These factors can effectively represent the correlation between the equipment operation state and the environment, provide more targeted input features for subsequent life prediction, and ensure the accuracy and stability of subsequent model calculation;

[0068] Based on the multi-source operation data, a life prediction model that fuses physical degradation rules and data-driven learning is used to calculate the prediction results of the flight simulation equipment. The prediction results include life prediction values and probability distribution intervals of residual life;

[0069] In this embodiment, the life prediction model includes a physical model and a data-driven model, and the life prediction model The expression is:

[0070] ;

[0071] wherein, is a dynamic weight coefficient; is a life prediction value calculated by a physical model; is a life prediction value calculated by a data-driven model;

[0072] Specifically, the physical model is to calculate a physical prediction value based on a degradation law (such as a Weibull distribution or a lognormal life model), and the specific implementation is as follows:

[0073] Based on the component degradation mechanism, a cumulative physical degradation amount is defined The theoretical expression is:

[0074] ;

[0075] wherein, is a multi-source operating parameter directly related to the degradation mechanism of the component (which is selected according to the structural characteristics and main failure modes of different flight simulator components, for example: for electromechanical execution components, the operating parameters at least include cumulative operating time, load level, current, vibration and temperature; for hydraulic components, the operating parameters at least include working pressure, flow fluctuation, oil temperature and start-stop times; for optical display consumables, the operating parameters at least include cumulative lighting time, working temperature and on-off frequency), is a degradation rate function determined by the degradation mechanism (used to describe the change law of the degradation speed of the component under different operating conditions and environmental conditions, the input is a multi-source operating parameter related to the degradation mechanism of the component, and the specific form of the degradation rate function is designed and parameterized according to the structural characteristics and main failure modes of different flight simulator components; for example: for electromechanical execution components, it can be represented as a weighted linear or segmented linear function of cumulative operating time, load level, working current, vibration amplitude and temperature; for hydraulic components, it can be represented as a nonlinear function of working pressure, flow fluctuation, oil temperature and start-stop times; for optical display consumables, it can be represented as an exponential or logarithmic function of cumulative lighting time and working temperature; the parameters in the degradation rate function are obtained by regression calibration through historical operating data and failure samples, and are dynamically updated according to new failure data during system operation to continuously fit the actual degradation law);

[0076] In actual engineering implementation, the cumulative physical degradation amount is calculated in a discrete time approximation manner, which converts continuous time integration into accumulation of the degradation rate in the sampling period, and the expression is:

[0077] ;

[0078] wherein: is the cumulative physical degradation amount of component i at current time ; is the multi-source operating parameter vector collected at historical sampling time tm; is the time interval between adjacent two samplings; in subsequent calculations, this discrete value is used as the theoretical value is the engineering approximation value at time tk;

[0079] When the cumulative physical degradation amount of component i reaches the preset failure threshold (the failure threshold is set according to the component type, and the initial value is obtained by statistical analysis of historical failure samples, preferably using the high quantile of the corresponding life distribution as the failure criterion, and allowing dynamic correction based on new failure samples during system operation to adapt to changes in actual operating conditions), the component is considered to fail. In actual application, considering that the operating conditions change relatively gently within the short-term prediction window, the expected value of the degradation rate at the current time is used as the equivalent approximation of the degradation rate in the future period of time, and the life prediction value output by the physical model , specifically, the (remaining) life prediction value at the current time is:

[0080] ;

[0081] The data-driven model is a machine learning prediction value trained from historical operating data, and the specific implementation is as follows:

[0082] Based on historical operating data training, the input feature vector is defined as:

[0083] ;

[0084] The life prediction value output by the data-driven model , specifically, the (remaining) life prediction value at the current time is:

[0085] ;

[0086] wherein, , are the cumulative degradation amount and degradation rate of component i at the current time calculated by the physical model (the degradation rate is the calculation result of the degradation rate function of the component under the current operating parameter condition, used to represent the actual degradation level of the component at the current time); For the current moment The collected multi-source operational parameter vector; For historical sequences within a sliding time window The statistical feature vectors (such as mean, standard deviation, maximum value, etc.) calculated by the sequence used for discrete calculation of the physical model are described in the following context. The length of the time window is set according to the frequency of component use and the data sampling period, and is updated in a rolling manner with a fixed step size as new data arrives. The nonlinear regression function, trained using historical life samples, is used to characterize the mapping relationship between life and degradation characteristics under complex working conditions. It can be any of the following models: random forest regression model, support vector regression model, gradient boosting tree model, or multilayer neural network model. Its training objective is to minimize the regression error between the predicted remaining life and the actual life. After the model is trained, the corresponding data-driven life prediction value is output in the online prediction stage based on the real-time collected running data.

[0087] Through dynamic weighting coefficients The prediction results from the two models are weighted and fused to obtain a unified lifetime prediction value:

[0088] ;

[0089] This hybrid model balances mechanistic interpretability and data adaptability, considering the actual lifespan variations of different equipment types (consumable parts and disposable parts). The model operates in three phases: First, an offline phase, fitting the physical lifespan distribution based on historical failure data and training data-driven models separately for consumable parts and disposable parts. The training data includes at least 1000 known lifespan samples, with stratified sampling maintaining a balanced lifespan distribution. Second, an online phase, inputting sensor data in real time and outputting both physical and data predictions. Third, a dynamic fusion phase, based on dynamic weighting coefficients... The model calculates the final lifetime prediction results and uses a combination of leave-one-out validation (LOOCV) and rolling window validation to ensure temporal consistency. The output remaining lifetime probability distribution interval (90% confidence interval) can also provide key evidence for inventory risk assessment, solving the problem of traditional lifetime prediction results being singular and lacking in reference.

[0090] The probability distribution interval of the remaining lifespan is determined by the following method:

[0091] Model uncertainty modeling: Based on historical prediction errors during training (such as residual data corresponding to MAE and RMSE), statistical error distribution characteristics are analyzed within a fixed sliding window (such as containing 500 consecutive historical prediction samples), and a probability distribution model of the error (such as lognormal or Weibull distribution) is fitted to construct an initial confidence interval for the point prediction value. The choice of distribution model depends on the distribution characteristics of the error data (such as lognormal distribution when the error is right-skewed, and Weibull distribution when it has a 'bathtub curve' characteristic).

[0092] Dual-model interval fusion: First, lifetime distribution functions for the two types of models are constructed separately: Physical model side: The lifetime distribution function of the physical model is obtained by fitting based on degradation laws (such as Weibull distribution, log-normal distribution). This represents the cumulative probability that the remaining lifetime, as predicted by the physical model, will not exceed time t; On the data-driven model side: the lifetime distribution function output by the machine learning model trained from historical operating data. , representing the cumulative probability that the remaining lifetime will not exceed time t under the prediction of the data-driven model; subsequently, through the dynamic weight coefficient wt (which is fused with the model's principal coefficients) Consistent, based on the dynamic calculation of the proportion of real-time prediction error variance of the two models (with a value range of [0,1]), the distribution functions of the two models are weighted and fused to obtain the comprehensive lifetime distribution function. The formula is: Finally, based on Construct a comprehensive confidence interval (e.g., 90% confidence level);

[0093] Probability distribution generation and risk quantification: Based on the comprehensive confidence interval, random sampling is performed using the Monte Carlo method (sampling times N=10000 times) to generate an empirical probability distribution of remaining lifespan. This allows for the output of a lifespan interval with a specified confidence level (e.g., 90%), and the calculation of the cumulative failure probability P (L≤T) within any future time window. Here, T is a preset lifespan threshold (e.g., the rated lifespan of a reusable part, or the upper limit of the lifespan corresponding to the safety stock of consumable parts), and L is the remaining lifespan of the component. Ultimately, this cumulative failure probability serves as the direct input to the inventory decision model, triggering reordering decisions.

[0094] The dynamic weighting coefficient is obtained by the following method:

[0095] Obtain the The first average error in the most recent W predictions ;

[0096] Obtain the The second average error in the most recent W predictions ;

[0097] wherein W needs to satisfy the following constraints and adaptive adjustment rules:

[0098] Time span constraint: W needs to cover one complete typical usage cycle of the corresponding equipment component - the typical usage cycle of the rotating part is 2 weeks (corresponding to its complete operation-maintenance cycle), and the typical usage cycle of the consumable part is 1 week (corresponding to its rapid consumption cycle), so as to ensure the representativeness of the error estimation;

[0099] Data volume constraint: the number of valid prediction points of each window is bound to the usage frequency of the component: 200 valid prediction points for high-frequency consumable parts (daily usage ≥ 10 times), and 50 valid prediction points for low-frequency rotating parts (daily usage < 10 times) to avoid unstable statistics; adaptive strategy: the system defaults to an initial value (W = 100), and subsequently automatically adjusts based on the fusion error variance of the physical model and the data-driven model; adjustment rules: ;

[0100] Upper and lower limits of W adjustment: not lower than the lower limit of the data volume constraint, not higher than the upper limit of the data volume constraint;

[0101] determining the dynamic weight coefficient based on the inverse relationship between the first average error and the second average error ; the formula expression is:

[0102] ;

[0103] When the physical model error is small, the value automatically increases, increasing the weight of the physical model prediction result; when the data model performs better, the value automatically decreases, increasing the influence of the data-driven model, and if there is a systematic deviation for a continuous number of times (such as 5 consecutive times of overestimation), the system will automatically fine-tune (such as reducing 0.05) to ensure that the fusion result remains stable and accurate in the long term. This dynamic adjustment mechanism allows the model to always adapt to changes in the equipment operating state, improving the accuracy of life prediction and avoiding the prediction deviation problem that is prone to occur in traditional fixed weight models;

[0104] In this embodiment, for the repairable components (i.e., rotating parts) of the flight simulator equipment that are put into use after repair, the method further comprises: calculating a post-repair health coefficient based on the multi-dimensional evaluation results of this repair, and using the post-repair health coefficient to calibrate the prediction results of the repairable components; the consumable parts are one-time use parts and do not require a repair process, so there is no need for post-repair health coefficient calibration;

[0105] The multi-dimensional evaluation includes at least: post-repair functional test score, replacement component value ratio, and repair work order pass rate;

[0106] In this embodiment, the "post-repair status" is quantified into a single value, namely the post-repair health coefficient. The weighted composite score, based on multi-dimensional scoring, is expressed by the following formula:

[0107]

[0108] Each item It is a normalized sub-score, weighted and ;

[0109] Commonly used items (Selectable based on your system's collected data) Includes: Post-repair functional test score (For example, a perfect score of 100, normalized to [0,1]), preferably using automated test results and performing normalization processing; the proportion of replacement component value. This refers to the ratio of the value of the replaced or repaired parts to the value of the corresponding new parts, used to reflect the depth of repair; repair quality / work order pass rate (e.g., whether it passed multiple retests). It is used to characterize the reliability of maintenance processes and results; it can also include a historical failure rate correction item (if the same component fails repeatedly in a short period of time, the score will be reduced). Sub-items such as maintenance personnel / maintenance location rating Example initial weight settings: engineering experience value, functional test score weight 0.5, replacement part value ratio 0.3, repair quality / work order pass rate 0.2. These can be adjusted through historical data regression. Specifically, the weights of each sub-item in the health coefficient are... During system operation, based on historical maintenance samples, least squares regression with the actual lifespan recovery rate after maintenance as the supervision variable is used for statistical correction, and the system is periodically updated under non-negativity and normalization constraints to ensure that the health coefficient is consistent with the actual maintenance effect. The formula expression for the weight of each sub-item in the health coefficient is as follows:

[0110] ;

[0111] ;

[0112] In the formula, The normalized score of the i-th sub-item in the k-th repair; is the corresponding actual life recovery ratio after repair; w = [w1, …, wm] is a weight vector of the health coefficient after repair; N is the number of repair samples available for regression, and the system automatically triggers weight update once N≥50 effective repair records are accumulated, to ensure that the health coefficient evaluation gradually approaches the actual repair effectiveness;

[0113] After obtaining the health coefficient after repair, it is introduced into the life prediction process as a state calibration factor to correct the initial health level of the repairable component after repair. The method for califying the prediction result of the repairable component using the health coefficient after repair is:

[0114] The health coefficient after repair is used as a multiplication factor to act on the life prediction value output by the life prediction model to obtain the calibrated life prediction value The specific expression is: In this way, the influence of repair quality and repair depth on the actual operation ability of the component can be explicitly introduced into the life prediction process, and systematic prediction deviation caused by ignoring the difference in state after repair can be avoided.

[0115] Based on the prediction result, an inventory decision model is constructed and solved to minimize the total cost, to determine a dynamic inventory strategy, which includes reorder time, order quantity, and risk level.

[0116] In this embodiment, the method for constructing and solving an inventory decision model based on the prediction result to minimize the total cost and determine a dynamic inventory strategy is:

[0117] The probability distribution interval of the remaining life is mapped to the cumulative failure probability of the component within the future order lead time.

[0118] When the cumulative failure probability exceeds a dynamic risk tolerance threshold, the solution of the inventory decision model is triggered. The dynamic risk tolerance threshold is automatically adjusted according to the type of the equipment component and the criticality level of the associated flight training task plan.

[0119] The total cost of the inventory decision model includes inventory holding cost, procurement processing cost, and shortage risk cost.

[0120] Specifically, for each critical component, the system calculates the “failure probability” of the future period (namely, the cumulative failure probability P) according to the life prediction distribution, and compares the cumulative failure probability with a dynamic risk tolerance threshold When , the solution of the inventory decision model is automatically triggered to determine the optimal reorder time and order quantity.

[0121] The dynamic risk tolerance threshold is mainly set according to the associated flight training task level to ensure that high-level tasks enjoy a higher equipment support level. In this framework, different types of components can be adaptively fine-tuned to the threshold based on their importance (such as turnover parts or consumable parts). For example, the threshold of turnover parts is usually set more conservatively (i.e. lower). The baseline correspondence between task level and risk tolerance threshold is shown in Table 1:

[0122] Table 1

[0123]

[0124] The system automatically reads the task level from the flight training task plan and queries the corresponding risk tolerance threshold according to the task level. If the predicted cumulative failure probability is to be calculated, the inventory decision module is automatically triggered to make replenishment decisions and optimization solutions.

[0125] The goal of inventory optimization is to minimize the total cost, which is calculated as follows:

[0126]

[0127] Where, is the average inventory level, is the expected number of missing parts, M is the number of purchases, is the inventory holding cost, is the missing part risk cost (the risk cost of turnover parts is much higher than that of consumable parts), is the purchase and transportation cost; this model strongly correlates the life prediction results with the inventory cost, enabling the scientific formulation of inventory strategies.

[0128] In this embodiment, the missing part risk cost is associated with the impact level of the flight training task, and the missing part risk cost increases in a non-linear manner when the missing part duration exceeds the preset threshold.

[0129] Specifically, the missing part risk cost is used to measure the economic loss or business impact caused by the delay, cancellation or rescheduling of flight training tasks due to the absence of key components. The calculation of this cost is based on the task impact level, and its expression is as follows:

[0130]

[0131] Where, is the basic risk coefficient corresponding to the task level, reflecting the sensitivity of different flight training tasks to missing part events; is the non-linear weighting coefficient corresponding to the task level, determining the speed of risk cost growth when the missing part duration exceeds the threshold; ​​The duration of the missing part for the current component can be measured in hours or days, indicating the length of time from the insufficient inventory to the replenishment of the component; The duration threshold of the missing part, within which the risk cost of the missing part increases linearly; beyond the threshold, the risk cost increases exponentially;

[0132] In application, if the component (such as engine and other rotating parts) is missing, it will cause the core flight training task to stop, and the corresponding basic risk coefficient And the nonlinear weighting coefficient Will be much higher than that of ordinary components (such as operation panel keys and other consumables); when the missing duration exceeds the set threshold (such as 3 days), the cost will increase by the square, which reflects the nonlinear impact of task delay, and this setting can force the system to prioritize the supply of equipment for high-priority training tasks and reduce the loss of major tasks due to missing parts;

[0133] The solving inventory decision model adopts a particle swarm optimization algorithm to find the optimal safety stock and reorder point to minimize the total cost. Specifically, the particle swarm optimization algorithm (PSO) is used to find the optimal safety stock S and order quantity N to minimize the total cost C, and the output results are the optimal reorder time, quantity and risk level of each component. The system can automatically generate a purchase task list. Compared with the traditional manual procurement plan, this algorithm can quickly traverse multiple inventory strategy combinations to find the optimal balance point of cost and risk, avoiding the problem of high inventory leading to capital occupation, reducing the risk of missing parts caused by insufficient inventory, and improving the intelligence and refinement level of inventory management;

[0134] After determining the dynamic inventory decision, see Figure 4 It also includes simulating the replacement of aircraft parts, on the one hand, the removed aircraft parts are repaired and returned to the warehouse, and the corresponding inventory is increased by 1; on the other hand, it checks whether there is inventory of the aircraft parts. If there is inventory, it means that the aircraft parts have been successfully secured once, and the corresponding inventory should be reduced by 1; if there is no inventory, it means that it has not been successfully secured once, and the corresponding number of missing parts is increased by 1. Finally, it checks whether the simulation end condition is met. If the condition is met, the simulation ends, and the result is the optimal result. Otherwise, it recycles until the optimal result appears;

[0135] The demand of spare parts is random, in order to reduce AOG, a safety stock should be set, the safety stock refers to the minimum stock of spare parts, the safety stock is not a fixed value, depends on the regularity of spare parts consumption, the length of supply cycle and the management decision of spare parts guarantee reliability and so on, for example, the repair site becomes closer, the safety stock can be lower, the guarantee rate includes the planned guarantee rate and the actual guarantee rate, the planned guarantee rate is determined by the price of the spare parts and the annual consumption quantity, the value is not disturbed by any organization, the planned number of safety stock is based on this guarantee rate, the actual guarantee rate is the actual value of guarantee rate reached through ordering, borrowing, local manufacturing, simulation active string parts and other ways, usually expressed as: guarantee rate = number of times of issuing materials / total number of times of issuing materials x 100%;

[0136] Obtaining a maintenance result of the flight simulation equipment in actual operation, and calculating a prediction error between the prediction result and the maintenance result;

[0137] According to the prediction error, adaptively correcting the life prediction model and the inventory decision model;

[0138] In this embodiment, according to the prediction error, the life prediction model and the inventory decision model are adaptively corrected, and the method is as follows:

[0139] When the prediction error presents a directional deviation in statistics, the dynamic weight coefficient of the physical model and the data-driven model in the life prediction model is fine-tuned;

[0140] When the sample proportion of a specific feature sub-domain in the prediction error exceeds a preset concentration threshold, the life prediction model is locally re-calibrated or locally re-trained;

[0141] When the number of newly added labeled samples after the last training accounts for more than a preset proportion of the original training set, or the overall error of the model continuously deteriorates, the life prediction model is fully re-trained;

[0142] Specifically, the maintenance personnel will record the actual replacement time, fault type, repair result and other information in the system, and the system automatically calculates the life prediction error rate , wherein Lpred is the predicted lifetime value given by the model at the moment before the actual failure or replacement of the component, with units of hours (h) or number of uses, depending on the component type; Lreal is the real lifetime of the component recorded by the maintenance personnel, i.e. the actual length of time / number of uses from the last installation into use to the current failure or replacement, with the same units as Lpred; |Lpred-Lreal| represents the absolute error between the predicted lifetime and the actual lifetime, used to measure the model bias without distinguishing between positive and negative; E is the lifetime prediction error rate (Error Rate), used to characterize the relative deviation between the model prediction and the real lifetime, E ∈ [0, +∞), and generally the lower the better; when E≤10%, it indicates that the prediction is relatively accurate; the system will first define online statistics (based on the latest window W), including directional bias , overall error , local error concentration FD, and the proportion of new data , the expressions are as follows:

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] wherein D is the feature sub-domain, is the error threshold is the number of new samples, is the number of original training set samples;

[0148] First level: dynamic weight fine-tuning

[0149] Triggering condition: there is a significant systematic bias (i.e., the absolute value of the bias ), but the overall accuracy remains stable (i.e., the relative change in MAE ); (typical parameter settings: bias threshold of 50 hours, deterioration tolerance of 10%);

[0150] Fine-tuning fusion weight coefficient: , the weight range is limited to the interval [0, 1] 0,1] interval;

[0151] Second level: local model re-calibration:

[0152] Triggering condition: the error is highly concentrated in a specific feature sub-domain (i.e., ), and the sub-domain data is sufficient (i.e., ), (typical parameter settings: concentration threshold of 60%, minimum sample number threshold of 50);

[0153] Execution action: only the model corresponding to the feature sub-domain is locally retrained (i.e., recalibrated), and the other parts remain unchanged. Specifically, first, all samples in the feature sub-domain D are selected from the full amount of samples and the number is confirmed to meet the threshold requirement. Then, only the parameter module associated with the sub-domain in the model is locked, and the remaining parameters are fixed. Subsequently, the sub-domain samples are input, and the original loss function and optimizer are used to perform small batch iteration training on the exclusive module. Through back propagation, only the sub-domain associated parameters are updated to the sub-domain MAE convergence. Finally, the trained sub-domain parameters replace the corresponding module parameters of the original model, and the local recalibration is completed. This is a very common and mature method in industrial-level adaptive model optimization, and will not be described in detail here.

[0154] Third level: full model retraining

[0155] Triggering conditions (any of them can be met): 1, significant change in data distribution: the proportion of newly labeled data exceeds the threshold (i.e., η> ηth, typical value ηth=20%); 2, performance continues to deteriorate: the relative change in MAE exceeds the deterioration threshold for consecutive evaluation windows (i.e., ΔMAE%> R2, deterioration threshold R2=15%, M=3 for consecutive MM windows); 3, local correction is invalid: after local recalibration, the overall error rate is still higher than the acceptable level (e.g., E> 15%);

[0156]

[0157] 3, local correction is invalid: after local recalibration, the overall error rate is still higher than the acceptable level (e.g., E> 15%);

[0158] Execution action: retrain the entire model using all available data, update all parameters. Specifically, first, merge the original training set with the newly labeled data, and after cleaning and normalization, divide the training, validation, and test sets. Then, load the complete structure of the model without locking any parameters, and re-execute the complete training process with the integrated full amount of data as input. Through forward propagation, calculate the global loss, and update all trainable parameters of the model through back propagation iteration. Combine the validation set early stopping mechanism to avoid overfitting. After training is completed, replace all old parameters of the original model with the new full amount of parameters and versionize the storage. After the performance of the test set is verified to meet the performance, load it to the inference service, and monitor the core indicators to ensure that the update is effective. This is a very common and mature method in industrial-level adaptive model optimization, and will not be described in detail here.

[0159] ​In this way, through the mixed strategy of "micro-update + periodic full update", the model is kept stable and continuously optimized, forming a truly self-learning prediction system. The closed-loop feedback mechanism enables the model to continuously iterate and optimize, solving the problem of traditional model drift and gradual decline in prediction accuracy.

[0160] The following is an example of a servo actuator in a motion platform system. This component is a repairable turnover part, and its running state directly affects the flight attitude simulation effect. The system collects cumulative running time, vibration level, current, and working temperature running state data through sensors, and synchronously collects environmental temperature and humidity information. The data is input into the life prediction module as multi-source running data,

[0161] 1. The life prediction module uses physical degradation models and data-driven models to predict the remaining life of the servo actuator based on the above data, and obtains the final life prediction result and the corresponding remaining life probability distribution interval through a dynamic weight fusion method.

[0162] 2. When the servo actuator is replaced and repaired, the system calculates the post-repair health coefficient based on the post-repair function test results, replacement component ratio, and repair work order qualification, and recalibrates the life prediction initial value after the component is put into use again to reflect the influence of repair quality on the degree of life recovery.

[0163] 3. The inventory decision module calculates the cumulative failure probability of the servo actuator within the future order lead time based on the remaining life prediction result. When the cumulative failure probability exceeds the risk tolerance threshold corresponding to the current flight training task level, the system automatically triggers the inventory decision model to output the reordering time and order quantity of the component, in order to reduce the risk of missing parts and control inventory costs.

[0164] Taking the projection light source module in the visual display system as an example, this component is a non-repairable consumable part. The system collects data such as cumulative lighting time and working temperature to predict its remaining life, and dynamically adjusts the safety stock level accordingly. When the prediction failure risk increases, the replenishment operation is triggered in advance to avoid the interruption of flight simulation training due to light source failure.

[0165] Therefore, through the whole-process technical system of multi-source perception of the Internet of Things, fusion life prediction, probability-driven inventory decision, and closed-loop adaptive correction, the intelligent upgrading of flight simulator equipment management and inventory allocation is realized. In practical application, after the flight simulation training base adopts the method, the accuracy of equipment residual life prediction is significantly improved, compared with the traditional manual cycle estimation mode, the life prediction error is greatly reduced, and the waste of resources caused by "premature replacement" and the equipment failure risk caused by "delayed replacement" are avoided. In addition, through the dynamic weighted fusion of the "physical degradation model + data-driven model", the running characteristics and environmental differences of different types of equipment can be accurately adapted, and the life prediction of different types of equipment such as turnover parts and consumable parts is stable and reliable without manual intervention to adjust model parameters. Finally, based on the inventory triggering logic of the residual life failure probability and the cost optimization mechanism of the particle swarm algorithm, the optimal reordering time and batch can be automatically generated, and the model closed-loop iteration of maintenance data feedback can greatly reduce the occurrence frequency of equipment AOG (Aircraft On Ground), effectively improve the continuity and safety of flight training tasks, and reduce the inventory fund occupation and overall maintenance cost.

[0166] 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 above-described method can refer to the corresponding process in the foregoing system embodiment, which will not be repeated here.

[0167] The second embodiment of the application is a flight simulation equipment management and inventory optimization system based on the flight simulation equipment management and inventory optimization method. The system comprises:

[0168] A data acquisition module configured to acquire multi-source operation data of the flight simulation equipment, wherein the multi-source operation data comprises operation state data and environment data;

[0169] A life prediction module configured to calculate a prediction result of the flight simulation equipment by using a life prediction model that fuses physical degradation rules and data-driven learning based on the multi-source operation data, wherein the prediction result comprises a life prediction value and a probability distribution interval of residual life;

[0170] An inventory decision module configured to construct and solve an inventory decision model with the objective of minimizing total cost based on the prediction result to determine a dynamic inventory strategy, comprising: determining the cumulative failure probability of the equipment component within a preset future support time window based on the probability distribution interval of residual life;

[0171] compare the accumulated failure probability with a dynamic risk tolerance threshold, and trigger an inventory decision module when the accumulated failure probability exceeds the dynamic risk tolerance threshold; wherein the dynamic risk tolerance threshold is automatically adjusted according to the type of the equipment component and the criticality level of the associated flight training mission plan;

[0172] the inventory strategy comprises reorder point, order quantity and risk level;

[0173] an error calculation module configured to obtain a maintenance result of the flight simulation equipment in actual operation, and calculate a prediction error between the prediction result and the maintenance result;

[0174] a model correction module configured to adaptively correct the life prediction model and the inventory decision model according to the prediction error.

[0175] It should be noted that the flight simulation equipment management and inventory optimization system provided in the above embodiment 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 embodiment of the present application are further decomposed or combined, for example, the modules of the above embodiment 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.

[0176] Those skilled in the art should be able to realize that the modules and method steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. The programs corresponding to the software modules and method steps can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the components and steps of each example have been generally described in the above description. Whether the functions are performed by electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0177] The terms "first", "second", and the like are used to distinguish similar objects, and are not used to describe or indicate a specific order or sequence.

[0178] The term "comprising" or any other similar word is intended to encompass the inclusion of one or more steps, features, or elements but not to the exclusion of any other steps, features, or elements. The term "comprising" therefore indicates that the inclusion of one or more steps, features, or elements is not a requirement and that other steps, features, or elements can also be included.

[0179] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to 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 deviating from the principles of the present application, and the technical solutions after these changes or replacements will all fall within the protection scope of the present application.

Claims

1. A method for managing and optimizing the inventory of flight simulator equipment, characterized in that, The method includes the following steps: Collect multi-source operational data from flight simulator equipment, including operational status data and environmental data; Based on the multi-source operational data, a life prediction model that integrates physical degradation laws and data-driven learning is used to calculate the prediction results for flight simulator equipment; the prediction results include the predicted life value and the probability distribution range of the remaining life. Based on the prediction results, an inventory decision model with the goal of minimizing total cost is constructed and solved to determine a dynamic inventory strategy, which includes reordering timing, order quantity, and risk level. Obtain the maintenance results of the flight simulator equipment in actual operation, and calculate the prediction error between the predicted results and the maintenance results; Based on the prediction error, the lifetime prediction model and the inventory decision model are adaptively corrected. The lifetime prediction model includes a physical model and a data-driven model. The expression is: ； in, These are dynamic weighting coefficients; The predicted lifetime calculated for the physical model; The predicted lifetime value is calculated for the data-driven model; the dynamic weighting coefficient is obtained as follows: Obtain the The first average error in the most recent W predictions; Obtain the The second average error in the most recent W predictions; The dynamic weighting coefficient is determined based on the inverse relationship between the first average error and the second average error.

2. The method for managing and optimizing the inventory of flight simulator equipment as described in claim 1, characterized in that, For repairable components in the flight simulator equipment that have been repaired and put back into use, the method further includes: calculating a post-repair health coefficient based on the multi-dimensional evaluation results of this repair, and using the post-repair health coefficient to calibrate the prediction results of the repairable components; The multi-dimensional evaluation includes at least: post-repair functional test score, the proportion of replacement parts value, and the pass rate of repair work orders.

3. The method for managing and optimizing the inventory of flight simulator equipment as described in claim 2, characterized in that, The method for calibrating the prediction results of repairable parts using the post-repair health coefficient is as follows: The revised health coefficient is used as a multiplication factor and applied to the life prediction value output by the life prediction model to obtain the calibrated life prediction value.

4. The method for managing and optimizing the inventory of flight simulator equipment as described in claim 1, characterized in that, Based on the prediction results, an inventory decision model with the objective of minimizing total cost is constructed and solved to determine a dynamic inventory strategy. The method is as follows: The probability distribution range of the remaining lifetime is mapped to the cumulative failure probability of the component during the future order lead time. When the cumulative failure probability exceeds a dynamic risk tolerance threshold, the solution of the inventory decision model is triggered. The total cost of the inventory decision model includes: inventory holding costs, procurement processing costs, and shortage risk costs.

5. The method for managing and optimizing the inventory of flight simulator equipment as described in claim 4, characterized in that, The cost of missing parts is related to the impact level of the flight training mission, and when the duration of missing parts exceeds a preset threshold, the cost of missing parts increases in a non-linear manner.

6. The method for managing and optimizing the inventory of flight simulator equipment as described in claim 4, characterized in that, The inventory decision model is solved using the particle swarm optimization algorithm.

7. The method for managing and optimizing the inventory of flight simulator equipment as described in claim 4, characterized in that, Based on the prediction error, the lifetime prediction model and the inventory decision model are adaptively corrected, and the method is as follows: When the prediction error exhibits a statistically directional bias, the dynamic weighting coefficients of the physical model and the data-driven model in the lifetime prediction model are fine-tuned. When the proportion of the prediction error in a specific feature subdomain exceeds a preset concentration threshold, the lifetime prediction model is locally recalibrated or locally retrained. When the number of newly labeled samples added since the last training exceeds the preset proportion of the original training set, or when the overall error of the model continues to deteriorate, the life prediction model is fully retrained.

8. A management and inventory optimization system for flight simulator equipment, based on the management and inventory optimization method for flight simulator equipment according to any one of claims 1-7, characterized in that, The system includes: The data acquisition module is configured to acquire multi-source operational data of the flight simulator equipment, including operational status data and environmental data. The lifespan prediction module is configured to: calculate the prediction results of the flight simulator equipment based on the multi-source operational data and using a lifespan prediction model that integrates physical degradation laws and data-driven learning; the prediction results include the lifespan prediction value and the probability distribution range of the remaining lifespan. The inventory decision module is configured to: construct and solve an inventory decision model with the goal of minimizing total cost based on the prediction results, so as to determine a dynamic inventory strategy, including: determining the cumulative failure probability of equipment components within a preset future warranty time window based on the probability distribution range of the remaining lifespan; The cumulative failure probability is compared with a dynamic risk tolerance threshold, and the inventory decision module is triggered when the cumulative failure probability exceeds the dynamic risk tolerance threshold; wherein, the dynamic risk tolerance threshold is automatically adjusted according to the type of equipment component and the criticality level of the associated flight training mission plan; The inventory strategy includes reordering timing, order quantity, and risk level. The error calculation module is configured to: obtain the maintenance results of the flight simulator equipment in actual operation, and calculate the prediction error between the predicted results and the maintenance results; The model correction module is configured to adaptively correct the lifetime prediction model and the inventory decision model based on the prediction error.

Images