Prediction method for managing aviation turnover accessory deployment

Abstract

The invention discloses a prediction method for managing aviation turnover accessory deployment, and the method comprises the steps: obtaining the sample input data of a large-scale passenger plane and a preset damage replacement threshold value; acquiring sample output data of the large passenger plane; establishing a neural network model as a damage prediction model and a resource allocation notification model; assigning the sample input data and the sample output data to a damage prediction model for training; obtaining target input data after the turnover accessories of the large passenger plane are replaced, and assigning the target input data to the damage prediction model; the damage prediction model outputs target output data in the next voyage of the passenger plane; assigning the target output data and a preset damage replacement threshold to a resource allocation notification model for calculation and comparison; when a calculation and comparison result reaches a maintenance and replacement standard, a notification switch is triggered; and the resource allocation notification model sends a notification instruction to notify a turnover accessory storage warehouse of a next voyage destination to convey the corresponding turnover accessories to an airport maintenance area so as to complete accessory allocation.

Description

Technical Field

[0001] The invention relates to the technical field of aviation spare parts management and dispatching, and in particular to a forecasting method for managing the dispatching of aviation turnover spare parts. Background Art

[0002] In the aviation field, the rational allocation of turnover parts is crucial to ensure the safe flight and efficient operation of large passenger aircraft. The various key turnover parts of large passenger aircraft, such as vortex fan blades, landing support rods and brake discs, will continue to wear and damage during the taxiing, flying and landing of the aircraft. At present, the allocation and management of aviation turnover parts mostly adopts the method of regular inspection and replacement. This method lacks accurate prediction of the actual damage of the parts, which easily leads to waste of resources or failure to replace the parts in time when the parts are seriously damaged, thus affecting the flight safety and operational efficiency of the aircraft. At the same time, due to the complexity and uncertainty of aviation operations, the flight and taxiing conditions of different aircraft vary greatly. Traditional management methods are difficult to adapt to such dynamic changes and cannot accurately allocate turnover parts in advance according to the actual operating conditions of the aircraft, resulting in shortages or backlogs of parts in some cases, increasing operating costs and management difficulties.

[0003] Therefore, there is an urgent need to design a method that can accurately predict the damage of turnover parts based on the actual operating data of large passenger aircraft, and allocate resources efficiently and reasonably accordingly. Summary of the invention

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a forecasting method for managing the allocation of aviation turnover parts.

[0005] The present application example provides a forecasting method for managing the allocation of aviation turnover parts, the method comprising the following steps: S1: Obtain sample input data and preset damage replacement thresholds of large passenger aircraft, wherein the sample input data includes the number of taxiing, taxiing time, taxiing speed, number of flights, flight time, flight speed, temperature data of vortex fan blades and acceleration during landing of the large passenger aircraft in each historical year, and the preset damage replacement thresholds include the preset minimum replacement damage gap between adjacent blade tips of the vortex fan blades of the large passenger aircraft, the preset minimum replacement damage thickness of the brake disc, the preset minimum replacement compression damage deformation of the landing support rod, and the preset minimum replacement bending damage deformation of the landing support rod; S2: Obtain sample output data of large passenger aircraft, including the gap between adjacent blade tips of vortex fan blades, brake disc thickness, axial compression deformation and lateral bending deformation of landing support rods of large passenger aircraft in various historical years; S3: Establish a neural network model as a damage prediction model and establish a resource allocation notification model; S4: Assigning the sample input data as input features and the sample output data as output features to the damage prediction model for damage prediction training; S5: Obtain target input data of a large passenger aircraft after the last replacement of a turnover part, and assign the target input data to a trained damage prediction model for damage prediction, wherein the turnover parts include vortex blades, landing support rods, and brake discs, and the target input data include the number of taxiing, taxiing time, taxiing speed, number of flights, flight time, flight speed, temperature data of vortex blades, and acceleration during landing after the last replacement of the turnover parts; S6: the damage prediction model outputs target output data of the large passenger aircraft in the next flight, wherein the target output data includes the clearance between adjacent blade tips of the vortex fan blades of the large passenger aircraft in the next flight, the thickness of the brake disc, the axial compression deformation and the lateral bending deformation of the landing support rod; S7: assigning the target output data and the corresponding preset damage replacement thresholds to the resource allocation notification model for calculation and comparison; S8: When the calculation and comparison results meet the maintenance and replacement standards, the notification switch of the resource allocation notification model is triggered; S9: The resource allocation notification model sends a notification instruction to notify the nearest turnover parts storage warehouse at the next flight destination to transport the corresponding turnover parts to the airport maintenance area to complete the resource allocation of turnover parts.

[0006] The purpose of the present invention is to overcome the problem that the traditional aviation turnover parts allocation management method lacks accurate prediction of the actual damage of the parts, resulting in waste of resources or safety hazards, and is difficult to adapt to the dynamic changes in aviation operations. Based on this, the inventor of the present application provides a prediction method for managing the allocation of aviation turnover parts. This method, through in-depth analysis of historical operating data of large passenger aircraft, construction of an accurate damage prediction model and an intelligent resource allocation notification model, ensures that aviation turnover parts can be allocated in real time and efficiently according to the actual damage of the aircraft, greatly improving the safety and efficiency of aviation operations and ensuring the stable operation of aviation business.

[0007] In view of the above-mentioned shortcomings, this method introduces a neural network model into the allocation process of circulating spare parts, and realizes the optimal scheduling of circulating spare parts resources by training the neural network model for damage prediction. First, the detailed historical operation data of large passenger aircraft is collected as sample input data, and the reasonable preset damage replacement threshold of each type of circulating spare parts is determined. At the same time, the actual damage status data of each type of accessories of large passenger aircraft under detailed historical operation data is collected as sample output data. The sample input data and sample output data provide an accurate data basis for subsequent model training. These historical data cover the key parameters of various stages such as taxiing, flight and landing of large passenger aircraft, as well as the actual damage status of various types of accessories of large passenger aircraft, which can fully reflect the operating status of the aircraft.

[0008] Next, a neural network model is established as a damage prediction model to accurately predict the damage of different types of turnover accessories. During the training process, specific training strategies and optimization algorithms are used to improve the accuracy and generalization ability of the model.

[0009] Then, using the trained damage prediction model, the target input data of the large passenger aircraft after the last replacement of the turnover parts is input into the damage prediction model for damage prediction, and the target output data of the large passenger aircraft in the next flight is predicted, that is, the damage status of various types of turnover parts. This prediction method based on actual operation data can more accurately and quickly reflect the actual damage status of the parts.

[0010] Then, the predicted target output data is calculated and compared with the preset damage replacement threshold. When the calculation and comparison result reaches the maintenance and replacement standard, the notification switch of the resource allocation notification model is triggered. The resource allocation notification model sends the resource allocation information to the aviation material management center at the next flight destination by calling the information service interface. Based on the information, the aviation material management center takes out the corresponding turnover parts from the turnover parts storage warehouse closest to the destination and transports them to the airport maintenance area to ensure that the required turnover parts can be replaced in a timely and fast manner when the aircraft arrives at the destination.

[0011] Finally, through this precise prediction and efficient allocation mechanism, the shortcomings of traditional aviation circulating spare parts allocation management methods can be overcome in all aspects, the safety and economy of aviation operations can be improved, and the high standards of modern aviation business for circulating spare parts management can be met.

[0012] Furthermore, establishing a neural network model as a damage prediction model includes: A first neural network model is established as a vortex fan blade damage prediction model, a second neural network model is established as a landing component damage prediction model, and a third neural network model is established as a brake disc damage prediction model.

[0013] Further, step S3 includes: The number of flights of large passenger aircraft in each historical year, as well as the flight duration, flight speed and temperature data of vortex fan blades corresponding to each flight number are used as the first sample input data; The clearance between adjacent blade tips of vortex fan blades corresponding to each flight number of large passenger aircraft in each historical year is used as the first sample output data; The first sample input data is used as input feature data, and the first sample output data is used as output feature data to be assigned to the eddy current blade damage prediction model for damage prediction training.

[0014] Furthermore, step S3 also includes: The number of taxiing of large passenger aircraft in each historical year, as well as the taxiing time, taxiing speed and acceleration of large passenger aircraft during landing corresponding to each taxiing number are used as the second sample input data; The axial compression deformation and lateral bending deformation of the landing support rod corresponding to each taxiing number of the large passenger aircraft in each historical year are used as the second sample output data; The second sample input data is used as input feature data, and the second sample output data is used as output feature data to be assigned to the landing component damage prediction model for damage prediction training.

[0015] Furthermore, step S3 also includes: The taxiing times of large passenger aircraft in each historical year, as well as the taxiing duration, taxiing speed and acceleration during landing corresponding to each taxiing time are used as the third sample input data; The brake disc thickness corresponding to each taxiing number of large passenger aircraft in each historical year is used as the third sample output data; The third sample input data is used as input feature data, and the third sample output data is used as output feature data to be assigned to the brake disc damage prediction model for damage prediction training.

[0016] Furthermore, the training steps of the vortex fan blade damage prediction model include: Integrate the first sample input data and the first sample output data according to the historical year relationship to form a first training sample data set, and assign the first training sample data set to the vortex fan blade damage prediction model; Perform normalization preprocessing on the first training sample data set so that its value range is between [0, 1]; The preprocessed first training sample data set is randomly divided into a training set and a test set in a ratio of 80% to 20%; The eddy current fan blade damage prediction model is trained using the training set. The initial learning rate of the model is set to 0.001. As the number of training rounds increases, the learning rate decays to 0.9 times the original value after every 10 epochs. At the same time, the EarlyStopping mechanism is introduced to monitor the loss changes on the test set to prevent overfitting. The performance of the trained model is evaluated using the test set. If the accuracy between the calculated predicted value and the corresponding true value in the test set is less than 88%, retraining is continued until the performance evaluation standard is met.

[0017] Furthermore, the training steps of the landing component damage prediction model include: Integrate the second sample input data and the second sample output data according to the historical year relationship to form a second training sample data set, and assign the second training sample data set to the landing component damage prediction model; Perform normalization preprocessing on the second training sample data set so that its value range is between [0, 1]; The preprocessed second training sample data set is randomly divided into a training set and a test set in a ratio of 75% to 25%; The training set is used to train the landing component damage prediction model. The Adagrad optimization algorithm is used to set the initial learning rate to 0.01. After every 15 epochs, the batch size is increased to 1.5 of the original. At the same time, the L1 regularization term is introduced, and the regularization coefficient is set to 0.0001. The loss change on the test set is monitored to prevent overfitting. The performance of the trained model is evaluated using the test set, with mean square error and mean absolute error as evaluation indicators. If the mean square error is greater than 0.05 or the mean absolute error is greater than 0.03, retraining is continued until the performance evaluation criteria are met.

[0018] Furthermore, the training steps of the brake disc damage prediction model include: Integrate the third sample input data and the third sample output data according to the historical year relationship to form a third training sample data set, and assign the third training sample data set to the brake disc damage prediction model; Perform normalization preprocessing on the third training sample data set so that its value range is between [0, 1]; The third training sample data set after preprocessing is randomly divided into a training set and a test set at a ratio of 70% and 30%; The brake disc damage prediction model is trained using the training set, and the RMSprop optimization algorithm is used to set the initial learning rate to 0.005. After every 20 epochs, the learning rate decays to 0.8 times the original value. At the same time, the L2 regularization term is introduced, and the regularization coefficient is set to 0.0005. The loss change on the test set is monitored to prevent overfitting; The performance of the trained model is evaluated using the test set and the coefficient of determination is used and root mean square error as evaluation indicators, if the determination coefficient If the value is less than 0.9 or the RMS error is greater than 0.04, the training will continue until the performance evaluation criteria are met.

[0019] Further, step S5 includes: The number of flights of a large passenger aircraft after the last replacement of a turbine blade, as well as the flight time, flight speed and temperature data of the vortex blade corresponding to each flight number are assigned as the first target input data to the vortex blade damage prediction model for damage prediction; The number of taxiing of the large passenger aircraft after the last replacement of the landing support rod, as well as the taxiing time, taxiing speed and acceleration during landing corresponding to each taxiing number are assigned as the second target input data to the landing component damage prediction model for damage prediction; The number of taxiing times of a large passenger aircraft after the most recent replacement of the brake disc, as well as the taxiing time, taxiing speed and acceleration during landing corresponding to each taxiing number are assigned as the third target input data to the brake disc damage prediction model for damage prediction.

[0020] Further, step S8 includes: When the gap between adjacent blade tips of the vortex fan blades deviates from the preset minimum replacement damage gap by more than 5%, and / or the thickness of the brake disc is lower than the preset minimum replacement damage thickness, and / or the compression damage deformation and bending damage deformation of the landing support rod exceed the preset threshold, the notification switch that triggers the resource allocation notification model is in the on state.

[0021] In summary, the present invention has the following beneficial effects compared with the prior art: The present invention provides a prediction method for managing the allocation of aviation turnover parts. Through the above technical scheme, on the one hand, the historical operation data of large passenger aircraft in multiple dimensions is collected as sample input, combined with the corresponding actual damage status data of the parts, to lay a solid foundation for building an accurate damage prediction model, and to establish special prediction models for different types of turnover parts, so as to further improve the accuracy of damage prediction, and to grasp the damage of each part in advance, effectively avoiding the waste of resources and potential flight safety hazards caused by the lack of accurate prediction. On the other hand, by presetting a reasonable damage replacement threshold, the predicted damage data of the parts is calculated and compared with it, and with the help of an intelligent resource allocation notification model, the allocation process is triggered in time when the parts are about to reach the replacement standard, ensuring that the required turnover parts can be quickly and accurately transported to the corresponding airport maintenance area, ensuring that the aircraft can replace the parts in time and maintain normal operation. This refined allocation strategy based on data-driven and model prediction overcomes the problems that the traditional aviation turnover parts allocation management method is difficult to adapt to the dynamic changes of aviation operations, and is prone to shortages or backlogs of parts, and effectively meets the high standards of modern aviation business for the efficiency, accuracy and safety of turnover parts allocation, and effectively improves the overall efficiency of aviation operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 is a flow chart of the preparation method of the present invention; Figure 2 The present invention is a flow chart of the training of the eddy current fan blade damage prediction model; Figure 3 The present invention is a flow chart of training a landing component damage prediction model; Figure 4 The present invention is a flow chart of the training of the brake disc damage prediction model. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0024] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0025] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0026] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0027] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0028] In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of the present invention.

[0029] In the following description, suffixes such as "module", "component", "assembly" or "unit" are used only to facilitate the description of the present invention and have no specific meanings. Therefore, they can be used in combination.

[0030] The present invention will be further described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0031] Please refer to Figure 1 , is a flow chart of a prediction method for managing the allocation of aviation turnover parts provided by an embodiment of the present invention. Further, the specific steps of the prediction method for managing the allocation of aviation turnover parts include: S1: Obtaining sample input data and preset damage replacement thresholds of a large passenger aircraft, wherein the sample input data includes the taxiing times, taxiing duration, taxiing speed, flight times, flight duration, flight speed, temperature data of vortex blades, and acceleration during landing of the large passenger aircraft in each historical year, and the preset damage replacement thresholds include the preset minimum replacement damage gap between adjacent blade tips of the vortex blades of the large passenger aircraft, the preset minimum replacement damage thickness of the brake disc, the preset minimum replacement compression damage deformation of the landing support rod, and the preset minimum replacement bending damage deformation of the landing support rod; S2: Obtaining sample output data of the large passenger aircraft, wherein the sample output data includes the clearance between adjacent blade tips of vortex fan blades, brake disc thickness, axial compression deformation and lateral bending deformation of landing support rods of the large passenger aircraft in each of the historical years; S3: Establish a neural network model as a damage prediction model and establish a resource allocation notification model; S4: using the sample input data as input features and the sample output data as output features to assign values ​​to the damage prediction model for damage prediction training; S5: Obtain target input data of the large passenger aircraft after the last replacement of the turnover parts, and assign the target input data to the trained damage prediction model for damage prediction, wherein the turnover parts include vortex blades, landing support rods and brake discs, and the target input data include the number of taxiing, taxiing time, taxiing speed, number of flights, flight time, flight speed, temperature data of vortex blades and acceleration during landing after the last replacement of the turnover parts; S6: the damage prediction model outputs target output data of the large passenger aircraft in the next voyage, wherein the target output data includes the clearance between adjacent blade tips of the vortex fan blades of the large passenger aircraft in the next voyage, the thickness of the brake disc, the axial compression deformation and the lateral bending deformation of the landing support rod; S7: assigning the target output data and the corresponding preset damage replacement thresholds to the resource allocation notification model for calculation and comparison; S8: When the calculation and comparison result reaches the maintenance and replacement standard, triggering the notification switch of the resource allocation notification model; S9: The resource allocation notification model sends a notification instruction to notify the nearest turnover parts storage warehouse at the next flight destination to transport the corresponding turnover parts to the airport maintenance area to complete the resource allocation of the turnover parts.

[0032] The purpose of the present invention is to overcome the problem that the traditional aviation turnover parts allocation management method lacks accurate prediction of the actual damage of the parts, resulting in waste of resources or safety hazards, and is difficult to adapt to the dynamic changes in aviation operations. Based on this, the inventor of the present application provides a prediction method for managing the allocation of aviation turnover parts. This method, through in-depth analysis of historical operating data of large passenger aircraft, construction of an accurate damage prediction model and an intelligent resource allocation notification model, ensures that aviation turnover parts can be allocated in real time and efficiently according to the actual damage of the aircraft, greatly improving the safety and efficiency of aviation operations and ensuring the stable operation of aviation business.

[0033] In response to the above-mentioned shortcomings, the inventors of the present application found that as the number of flights increases, there is a significant change relationship between the flight time, flight speed, and temperature data of the vortex blades and the gaps between adjacent blade tips of the vortex blades. Specifically, an increase in the number of flights usually causes the vortex blades to be subjected to greater wear during long periods of high-speed flight, which in turn causes an increase in the gaps between adjacent blade tips. Through the analysis of historical data, it was found that when the flight time exceeds a certain threshold, the temperature data of the vortex blades will rise significantly. This high temperature state exacerbates material fatigue and increases the change in tip clearance.

[0034] Similarly, as the number of taxiing increases, there is an obvious relationship between the acceleration, taxiing speed, taxiing time and brake discs of large passenger aircraft during landing. Specifically, an increase in the number of taxiing means an increase in the frequency of braking, which leads to increased wear of the brake discs and a gradual decrease in thickness. In addition, an increase in acceleration usually causes a greater impact force, which has a more significant impact on the brake discs and landing support rods. Therefore, when the number of taxiing reaches a certain threshold, the thickness of the brake disc will be significantly lower than the preset minimum replacement damage thickness.

[0035] Therefore, the number of flights, flight duration, flight speed, and temperature data of vortex blades are used as input features, and the gap between adjacent blade tips of vortex blades is assigned as output features to the damage prediction model, and damage prediction training of vortex blades is performed through the damage prediction model. At the same time, the number of taxiing, taxiing duration, taxiing speed, and acceleration during landing are used as input features, and the compression deformation and lateral bending deformation of the landing support rod are assigned as output features to the damage prediction model, and damage prediction training of landing components is performed. Furthermore, the number of taxiing, taxiing duration, taxiing speed, and acceleration during landing are used as input features, and the thickness of the brake disc is assigned as an output feature to the damage prediction model, and damage prediction training of the brake disc is performed. In this embodiment, when screening the data input for training the damage prediction model, with respect to the flight speed, the speed data during the flight process is filtered to extract the average flight speed value in each flight cycle to reduce the impact of noise on the data, so as to accurately reflect the flight status. For the temperature data of the vortex blades, , using the piecewise normalization strategy, the temperature data is standardized by intervals to highlight the nonlinear damage effects under different temperature conditions, so as to obtain the key temperature values ​​in each flight stage (that is, the highest temperature, lowest temperature and average temperature experienced by the vortex fan blades during the flight, which directly affect the fatigue life and damage degree of the material); for the taxiing speed, the weighted average method is used to calculate the speed data during the taxiing process, and a weighted taxiing speed is extracted as a representative value to reflect the actual usage during the taxiing period; for the acceleration during landing, the peak acceleration during the landing process is calculated by the three-axis synthetic vector, and the maximum acceleration value is extracted as a representative feature; finally, these extracted feature values ​​are assigned to the damage prediction model for damage prediction training.

[0036] After training, the neural network model is used to predict the damage of the turnover parts of large passenger aircraft. When the predicted turnover parts reach the maintenance or replacement standards, the corresponding turnover parts storage warehouse nearest to the next flight destination of the corresponding passenger aircraft is notified to transport the corresponding turnover parts, and the corresponding turnover parts are transported to the airport maintenance area, thereby realizing the optimal scheduling of turnover parts resources.

[0037] First, we collect detailed historical operation data of large passenger aircraft as sample input data, determine reasonable preset damage replacement thresholds for various types of turnover parts, and collect detailed actual damage status data of various types of parts of large passenger aircraft under historical operation data as sample output data. Sample input data and sample output data provide accurate data basis for subsequent model training. These historical data cover key parameters of large passenger aircraft in various stages such as taxiing, flight and landing, as well as the actual damage status of various types of parts of large passenger aircraft, which can fully reflect the operation status of the aircraft.

[0038] Next, a neural network model is established as a damage prediction model to accurately predict the damage of different types of turnover accessories. During the training process, specific training strategies and optimization algorithms are used to improve the accuracy and generalization ability of the model.

[0039] Then, using the trained damage prediction model, the target input data of the large passenger aircraft after the last replacement of the turnover parts is input into the damage prediction model for damage prediction, and the target output data of the large passenger aircraft in the next flight is predicted, that is, the damage status of various types of turnover parts. This prediction method based on actual operation data can more accurately and quickly reflect the actual damage status of the parts.

[0040] Then, the predicted target output data is calculated and compared with the preset damage replacement threshold. In this embodiment, the replacement threshold specifically includes the preset minimum replacement damage gap between adjacent blade tips of the vortex fan blades, the preset minimum replacement damage thickness of the brake disc, the preset minimum replacement compression damage deformation of the landing support rod, and the preset minimum replacement bending damage deformation of the landing support rod. The setting of these thresholds is based on the relevant standards in the aviation maintenance manual to ensure that the parts that need to be replaced can be identified in a timely and accurate manner during aviation operations to ensure flight safety. When the calculation and comparison results meet the maintenance and replacement standards, the notification switch of the resource allocation notification model is triggered, and the information service interface is called to send the parts demand information of the large passenger aircraft to the aviation material management center of the next flight destination. The aviation material management center extracts the corresponding parts from the nearest turnover parts storage warehouse and transports them to the airport maintenance area based on the received parts demand information and the confirmed turnover parts demand to ensure the smooth maintenance of the passenger aircraft.

[0041] Finally, through this precise prediction and efficient allocation mechanism, the shortcomings of traditional aviation circulating spare parts allocation management methods can be overcome in all aspects, the safety and economy of aviation operations can be improved, and the high standards of modern aviation business for circulating spare parts management can be met.

[0042] In one possible implementation, establishing a neural network model as a damage prediction model includes: A first neural network model is established as a vortex fan blade damage prediction model, a second neural network model is established as a landing component damage prediction model, and a third neural network model is established as a brake disc damage prediction model.

[0043] When implementing this embodiment, in view of the differences in the damage mechanisms and working conditions of different aviation turnover accessories, different neural network models are constructed to make accurate predictions. For vortex blades, they are in a long-term high-speed flight state and will be affected by a variety of complex alternating stresses, such as centrifugal force, vibration stress, and aerodynamic force. Under the long-term influence of these stresses, the blades are prone to fatigue deformation and other damage. Moreover, no matter whether the damage occurs at the root, middle, or the connection between the sleeve and the bearing of the blade, under the action of high-speed airflow pressure, the damaged part will be deformed, and this deformation will cause the gap between the adjacent blade tips to change. Therefore, the present invention selects the parameter of the gap between the adjacent blade tips to characterize the damage condition of the vortex blades, and uses this as the basis to establish the first neural network model as the vortex blade damage prediction model. As for the landing components, they can be further subdivided into landing support rods and brake discs, which are mainly damaged during the landing and taxiing stage of the aircraft. Among them, the landing gear support column will bear the impact load from the ground when the aircraft lands. After long-term repeated action, it is easy to have axial compression deformation and lateral bending deformation. When the aircraft is at a high taxiing speed, the brake disc will generate a large friction between the ground. With the increase in the number of uses and the passage of time, its thickness will gradually become thinner. Based on the above characteristics, a second neural network model is established as a landing component damage prediction model to predict the compression deformation and bending deformation of the landing gear support column. A third neural network model is established as a brake disc damage prediction model to predict damage conditions such as changes in brake disc thickness.

[0044] Since the damage formation mechanism of vortex fan blades is mainly related to various complex stresses during flight, and the damage of landing gear support columns and brake discs is mainly caused by impact and friction during aircraft landing and taxiing, the causes and manifestations of damage to the three are different. Therefore, three different neural network models are introduced for separate training, which can more accurately capture the key features of damage to each component, thereby improving the accuracy of damage prediction, and then realizing the scientific allocation of aviation turnover parts resources, effectively avoiding resource waste or safety hazards caused by the shortcomings of traditional allocation management methods, ensuring the efficient and safe development of aviation operations, and better meeting the high standards of modern aviation business for turnover parts management.

[0045] In a possible implementation, step S3 includes: The number of flights of the large passenger aircraft in each historical year, and the flight duration, flight speed and temperature data of vortex blades corresponding to each of the flight numbers are used as first sample input data; The clearance between adjacent blade tips of the vortex fan blades corresponding to the number of flights of the large passenger aircraft in each historical year is used as the first sample output data; The first sample input data is used as input feature data, and the first sample output data is used as output feature data to be assigned to the vortex blade damage prediction model for damage prediction training.

[0046] In a possible implementation, step S3 further includes: The number of taxiing of the large passenger aircraft in each historical year, as well as the taxiing duration, taxiing speed and acceleration of the large passenger aircraft when landing corresponding to each taxiing number are used as second sample input data; The axial compression deformation and lateral bending deformation of the landing support rod corresponding to each of the taxiing times of the large passenger aircraft in each historical year are used as second sample output data; The second sample input data is used as input feature data, and the second sample output data is used as output feature data to be assigned to the landing component damage prediction model for damage prediction training.

[0047] In a possible implementation, step S3 further includes: The taxiing times of the large passenger aircraft in each historical year, and the taxiing duration, taxiing speed and acceleration during landing corresponding to each taxiing times are used as third sample input data; The brake disc thickness corresponding to each taxiing number of the large passenger aircraft in each historical year is used as the third sample output data; The third sample input data is used as input feature data, and the third sample output data is used as output feature data to be assigned to the brake disc damage prediction model for damage prediction training.

[0048] In one possible implementation, please refer to Figure 2 The training steps of the eddy current blade damage prediction model include: Integrate the first sample input data and the first sample output data according to the historical year relationship to form a first training sample data set, and assign the first training sample data set to the vortex fan blade damage prediction model; Performing normalization preprocessing on the first training sample data set so that its value range is between [0, 1]; The first training sample data set after preprocessing is randomly divided into a training set and a test set in a ratio of 80% to 20%; The eddy current fan blade damage prediction model is trained using the training set, and the initial learning rate of the model is set to 0.001. As the number of training rounds increases, the learning rate decays to 0.9 times of the original after every 10 epochs. At the same time, an EarlyStopping mechanism is introduced to monitor the loss changes on the test set to prevent overfitting; The performance of the trained model is evaluated using the test set. If the accuracy between the calculated predicted value and the corresponding true value in the test set is lower than 88%, retraining is continued until the performance evaluation standard is met.

[0049] In the implementation of this embodiment, in order to ensure the efficiency and accuracy of the vortex fan blade damage prediction model, a multi-layer perceptron (MLP) neural network is used as the basic model architecture. Specifically, the model includes an input layer, two hidden layers, and an output layer. The input layer receives the first sample input data, including the number of flights, flight time, flight speed, and vortex fan blade temperature data, and its dimension is 4. The first hidden layer contains 64 neurons and uses the ReLU activation function to capture nonlinear relationships. The second hidden layer contains 32 neurons and also uses the ReLU activation function. The output layer uses a linear activation function to output the predicted gap between adjacent blade tips of the vortex fan blade.

[0050] When this embodiment is implemented, in order to ensure that the vortex fan blade damage prediction model fully learns the flight timing characteristics, the periodic granularity of "year-month-day" is used for division in the data integration stage. Specifically, the historical data of each passenger aircraft for three consecutive years is extracted according to the flight segments with a three-day interval every month (such as the 1st, 4th, and 7th of each month), which can not only cover the impact of temperature on metal fatigue in different seasons, but also avoid the redundancy of adjacent date data. In the normalization process, Min-Max linear scaling is used for the flight speed (range 0-950km / h), and the vortex fan blade temperature data (range -50℃ to 650℃) adopts a segmented normalization strategy, compressing the interval below 300℃ to [0,0.5], and mapping above 300℃ to [0.5,1] to highlight the nonlinear damage effect under high temperature conditions. A dynamic learning rate mechanism is used in the training process. The initial learning rate of 0.001 converges stably in the first 10 epochs, and then decays by 0.9 times every 10 epochs, so that the model can avoid oscillation during fine adjustment in the later stage. The EarlyStopping mechanism is set when the test set loss decreases by less than 5 epochs in a row The training is terminated when the error is less than 500°C, effectively preventing overfitting while retaining the best weights. A triple check is used in the performance evaluation phase: through the above algorithms and mechanisms, in addition to limiting the global accuracy threshold to 88%, the prediction error under high temperature conditions (>500°C) is also specifically limited to no more than 3% to ensure the reliability of predictions under extreme conditions.

[0051] In one possible implementation, please refer to Figure 3 The training steps of the landing component damage prediction model include: Integrating the second sample input data and the second sample output data according to the historical year relationship to form a second training sample data set, and assigning the second training sample data set to the landing component damage prediction model; Performing normalization preprocessing on the second training sample data set so that its value range is between [0, 1]; The preprocessed second training sample data set is randomly divided into a training set and a test set according to a ratio of 75% to 25%; The landing component damage prediction model is trained using the training set, and the initial learning rate is set to 0.01 using the Adagrad optimization algorithm. After every 15 epochs, the batch size is increased to 1.5 of the original. At the same time, an L1 regularization term is introduced, and the regularization coefficient is set to 0.0001. The loss change on the test set is monitored to prevent overfitting; The performance of the trained model is evaluated using the test set, with mean square error and mean absolute error used as evaluation indicators. If the mean square error is greater than 0.05 or the mean absolute error is greater than 0.03, retraining is continued until the performance evaluation criteria are met.

[0052] In the implementation of this embodiment, a convolutional neural network (CNN) architecture is used for the landing component damage prediction model. The model includes a convolution layer, a maximum pooling layer, and two fully connected layers. The convolution layer uses 32 filters of size 3, a step size of 1, and a ReLU activation function. The maximum pooling layer uses a pooling window of size 2, a step size of 2, the first fully connected layer contains 64 neurons, and the second fully connected layer contains 32 neurons, both using the ReLU activation function. The output layer uses a linear activation function to output the axial compression deformation and lateral bending deformation of the landing support rod.

[0053] When implementing this embodiment, in view of the transient characteristics of the landing components subjected to impact loads, an event-triggered data acquisition mode is adopted when constructing the second training sample data set. Each complete landing and sliding action is taken as an independent sample unit, and the set composed of each independent unit sample is assigned to the landing component damage prediction model for training, that is, the axial compression deformation and lateral bending deformation of the landing support rod during each complete landing and sliding, and its corresponding number of sliding times, sliding duration, sliding speed and acceleration during landing are assigned as sample sets to the landing component damage prediction model for training. The sliding speed data is processed by Kalman filtering to eliminate instrument jitter noise, and the landing acceleration uses the kurtosis coefficient of the three-axis composite vector value to characterize the impact intensity. The application of the Adagrad optimizer can adaptively adjust the learning rate of each parameter. The batch size starts from the initial 32 and expands by 1.5 times every 15 epochs to a maximum of 128, which not only ensures the stability of the initial training, but also improves the convergence speed through large batches in the later stage. The introduction of L1 regularization (λ=0.0001) enables the model to automatically filter out key factors such as taxiing duration and number of taxiing times before inputting features into the convolutional layer, reducing the sparsity of hidden layer node connections from 78% to 43%, significantly improving the interpretability of the model. During the performance evaluation phase, in addition to the hard indicators of MSE<0.05 and MAE<0.03, a special abnormal operating condition test set was set up, including 20% ​​of special scenario data such as slippery runways and heavy-load landings, so that the overall prediction error fluctuation does not exceed 150% of the baseline indicator.

[0054] In one possible implementation, please refer to Figure 4 The training steps of the brake disc damage prediction model include: Integrate the third sample input data and the third sample output data according to the historical year relationship to form a third training sample data set, and assign the third training sample data set to the brake disc damage prediction model; Performing normalization preprocessing on the third training sample data set so that its value range is between [0, 1]; The preprocessed third training sample data set is randomly divided into a training set and a test set in a ratio of 70% to 30%; The brake disc damage prediction model is trained using the training set, and the initial learning rate is set to 0.005 using the RMSprop optimization algorithm. After every 20 epochs, the learning rate decays to 0.8 times of the original value. At the same time, an L2 regularization term is introduced, and the regularization coefficient is set to 0.0005. The loss change on the test set is monitored to prevent overfitting; The performance of the trained model was evaluated using the test set, using the coefficient of determination and root mean square error as evaluation indicators, if the determination coefficient If the value is less than 0.9 or the RMS error is greater than 0.04, the training will continue until the performance evaluation criteria are met.

[0055] In the implementation of this embodiment, a gradient boosted decision tree (GBDT) architecture is used for the brake disc damage prediction model. The model contains 100 decision trees, each with a maximum depth of 5, a learning rate of 0.1, a minimum sample split number of 2, and a minimum sample leaf node number of 1. The input features of the model include the number of gliding times, gliding duration, gliding speed, and acceleration during landing, and the output feature is the thickness of the brake disc. Through the integrated learning ability of GBDT, the nonlinear relationship of brake disc wear can be effectively captured.

[0056] When implementing this embodiment, in order to ensure the efficiency and accuracy of the brake disc damage prediction model, a time series window sampling strategy is adopted when constructing the third training sample data set. Specifically, the historical data of each passenger aircraft for three consecutive years is divided according to the complete cycle of each landing taxiing to form independent sample units. The taxiing speed data is processed by moving average filtering, and the landing acceleration uses the root mean square (RMS) of the three-axis synthetic vector value to characterize the impact intensity. In the normalization process, Min-Max linear scaling is used for the taxiing speed (range 0-300km / h), while the landing acceleration (range 0-3g) adopts a piecewise normalization strategy to compress the interval below 1g to [0,0.5], and map above 1g to [0.5,1] to highlight the nonlinear wear effect under high impact conditions. The RMSprop optimization algorithm is used in the training process, and the initial learning rate is set to 0.005, which decays by 0.8 times every 20 epochs to avoid oscillation during fine adjustment of the model in the later stage. The introduction of L2 regularization (λ=0.0005) enables the model to automatically filter out key factors such as the number of sliding times and braking duration from the input features, reducing the sparsity of hidden layer node connections from 65% to 38%, significantly improving the model interpretability. In addition to the hard indicators of ±0.9 and root mean square error RMSE<0.04, a special abnormal operating condition test set is set up, including 20% ​​of special scenario data such as slippery runways and heavy-load landings, so that the fluctuation of the entire prediction error does not exceed 150% of the baseline indicators.

[0057] In a possible implementation, step S5 includes: The number of flights of the large passenger aircraft after the last replacement of the turbine blades, as well as the flight duration, flight speed and temperature data of the vortex blades corresponding to each of the flight numbers are assigned as first target input data to the vortex blade damage prediction model for damage prediction; Assigning the number of taxiing of the large passenger aircraft after the last replacement of the landing support rod, and the taxiing duration, taxiing speed and acceleration during landing corresponding to each of the taxiing numbers as second target input data to the landing component damage prediction model for damage prediction; The number of taxiing of the large passenger aircraft after the most recent replacement of the brake disc, as well as the taxiing duration, taxiing speed and acceleration during landing corresponding to each of the taxiing numbers are assigned as third target input data to the brake disc damage prediction model for damage prediction.

[0058] In a possible implementation, step S8 includes: When the gap between adjacent blade tips of the vortex fan blades deviates from the preset minimum replacement damage gap by more than 5%, and / or the thickness of the brake disc is lower than the preset minimum replacement damage thickness, and / or the compression damage deformation and bending damage deformation of the landing support rod exceed the preset threshold, the state of the notification switch that triggers the resource allocation notification model is on.

[0059] When this embodiment is implemented, for the predicted value of the adjacent blade tip clearance of the vortex fan, the resource allocation notification model calculates the deviation rate from the preset threshold in real time. When ΔG≥5%, a level 1 alarm is activated, and the corresponding parts need to be repaired after the large passenger aircraft flies to the destination of the next flight mission. When ΔG≥8%, a replacement instruction is directly triggered, and the corresponding parts need to be replaced after the large passenger aircraft flies to the destination of the next flight mission. For the prediction of brake disc thickness, a thickness attenuation trend function is established , where k is the current thickness, λ is the material wear coefficient, and when F≤threshold×(1+3%), the replacement pre-instruction is triggered, and the corresponding parts need to be replaced after the large passenger aircraft flies to the destination of the next flight mission; for the deformation of the landing support rod, the vector synthesis method is used to calculate the comprehensive deformation and compare it with the preset deformation threshold. When the D value of three consecutive prediction cycles exceeds the threshold, the mandatory replacement procedure is initiated, and the corresponding parts need to be replaced after the large passenger aircraft flies to the destination of the next flight mission.

[0060] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A forecasting method for managing the allocation of aviation turnover parts, characterized in that: The following steps are involved: S1: Obtaining sample input data and preset damage replacement thresholds of a large passenger aircraft, wherein the sample input data includes the taxiing times, taxiing duration, taxiing speed, flight times, flight duration, flight speed, temperature data of vortex blades, and acceleration during landing of the large passenger aircraft in each historical year, and the preset damage replacement thresholds include the preset minimum replacement damage gap between adjacent blade tips of the vortex blades of the large passenger aircraft, the preset minimum replacement damage thickness of the brake disc, the preset minimum replacement compression damage deformation of the landing support rod, and the preset minimum replacement bending damage deformation of the landing support rod; S2: Obtaining sample output data of the large passenger aircraft, wherein the sample output data includes the clearance between adjacent blade tips of vortex fan blades, brake disc thickness, axial compression deformation and lateral bending deformation of landing support rods of the large passenger aircraft in each of the historical years; S3: Establish a neural network model as a damage prediction model and establish a resource allocation notification model; S4: using the sample input data as input features and the sample output data as output features to assign values ​​to the damage prediction model for damage prediction training; S5: Obtain target input data of the large passenger aircraft after the last replacement of the turnover parts, and assign the target input data to the trained damage prediction model for damage prediction, wherein the turnover parts include vortex blades, landing support rods and brake discs, and the target input data include the number of taxiing, taxiing time, taxiing speed, number of flights, flight time, flight speed, temperature data of vortex blades and acceleration during landing after the last replacement of the turnover parts; S6: the damage prediction model outputs target output data of the large passenger aircraft in the next voyage, wherein the target output data includes the clearance between adjacent blade tips of the vortex fan blades of the large passenger aircraft in the next voyage, the thickness of the brake disc, the axial compression deformation and the lateral bending deformation of the landing support rod; S7: assigning the target output data and the corresponding preset damage replacement thresholds to the resource allocation notification model for calculation and comparison; S8: When the calculation and comparison result reaches the maintenance and replacement standard, triggering the notification switch of the resource allocation notification model; S9: The resource allocation notification model sends a notification instruction to notify the nearest turnover parts storage warehouse at the next flight destination to transport the corresponding turnover parts to the airport maintenance area to complete the resource allocation of the turnover parts.

2. A forecasting method for managing the allocation of aviation turnover parts according to claim 1, characterized in that: Establishing a neural network model as a damage prediction model includes: A first neural network model is established as a vortex fan blade damage prediction model, a second neural network model is established as a landing component damage prediction model, and a third neural network model is established as a brake disc damage prediction model.

3. A forecasting method for managing the allocation of aviation turnover parts according to claim 2, characterized in that: The step S3 comprises: The number of flights of the large passenger aircraft in each historical year, and the flight duration, flight speed and temperature data of the vortex fan blades corresponding to each of the flight numbers are used as first sample input data; The clearance between adjacent blade tips of the vortex fan blades corresponding to the number of flights of the large passenger aircraft in each historical year is used as the first sample output data; The first sample input data is used as input feature data, and the first sample output data is used as output feature data to be assigned to the vortex blade damage prediction model for damage prediction training.

4. A forecasting method for managing the allocation of aviation turnover parts according to claim 2, characterized in that: The step S3 further comprises: The number of taxiing of the large passenger aircraft in each historical year, as well as the taxiing duration, taxiing speed and acceleration of the large passenger aircraft when landing corresponding to each taxiing number are used as second sample input data; The axial compression deformation and lateral bending deformation of the landing support rod corresponding to each of the taxiing times of the large passenger aircraft in each historical year are used as second sample output data; The second sample input data is used as input feature data, and the second sample output data is used as output feature data to be assigned to the landing component damage prediction model for damage prediction training.

5. A forecasting method for managing the allocation of aviation turnover parts according to claim 2, characterized in that: The step S3 further comprises: The taxiing times of the large passenger aircraft in each historical year, and the taxiing duration, taxiing speed and acceleration during landing corresponding to each taxiing times are used as third sample input data; The brake disc thickness corresponding to each taxiing number of the large passenger aircraft in each historical year is used as the third sample output data; The third sample input data is used as input feature data, and the third sample output data is used as output feature data to be assigned to the brake disc damage prediction model for damage prediction training.

6. A forecasting method for managing the allocation of aviation turnover parts according to claim 3, characterized in that: The training steps of the eddy current fan blade damage prediction model include: Integrate the first sample input data and the first sample output data according to the historical year relationship to form a first training sample data set, and assign the first training sample data set to the vortex fan blade damage prediction model; Performing normalization preprocessing on the first training sample data set so that its value range is between [0, 1]; The first training sample data set after preprocessing is randomly divided into a training set and a test set in a ratio of 80% to 20%; The eddy current fan blade damage prediction model is trained using the training set, and the initial learning rate of the model is set to 0.

001. As the number of training rounds increases, the learning rate decays to 0.9 times of the original after every 10 epochs. At the same time, an EarlyStopping mechanism is introduced to monitor the loss changes on the test set to prevent overfitting; The performance of the trained model is evaluated using the test set. If the accuracy between the calculated predicted value and the corresponding true value in the test set is lower than 88%, retraining is continued until the performance evaluation standard is met.

7. A forecasting method for managing the allocation of aviation turnover parts according to claim 4, characterized in that: The training steps of the landing component damage prediction model include: Integrating the second sample input data and the second sample output data according to the historical year relationship to form a second training sample data set, and assigning the second training sample data set to the landing component damage prediction model; Performing normalization preprocessing on the second training sample data set so that its value range is between [0, 1]; The preprocessed second training sample data set is randomly divided into a training set and a test set according to a ratio of 75% to 25%; The landing component damage prediction model is trained using the training set, and the initial learning rate is set to 0.01 using the Adagrad optimization algorithm. After every 15 epochs, the batch size is increased to 1.5 of the original. At the same time, an L1 regularization term is introduced, and the regularization coefficient is set to 0.0001. The loss change on the test set is monitored to prevent overfitting; The performance of the trained model is evaluated using the test set, with mean square error and mean absolute error used as evaluation indicators. If the mean square error is greater than 0.05 or the mean absolute error is greater than 0.03, retraining is continued until the performance evaluation criteria are met.

8. A forecasting method for managing the allocation of aviation turnover parts according to claim 5, characterized in that: The training steps of the brake disc damage prediction model include: Integrate the third sample input data and the third sample output data according to the historical year relationship to form a third training sample data set, and assign the third training sample data set to the brake disc damage prediction model; Performing normalization preprocessing on the third training sample data set so that its value range is between [0, 1]; The preprocessed third training sample data set is randomly divided into a training set and a test set in a ratio of 70% and 30%; The brake disc damage prediction model is trained using the training set, and the initial learning rate is set to 0.005 using the RMSprop optimization algorithm. After every 20 epochs, the learning rate decays to 0.8 times of the original value. At the same time, an L2 regularization term is introduced, and the regularization coefficient is set to 0.0005. The loss change on the test set is monitored to prevent overfitting; The performance of the trained model was evaluated using the test set, using the coefficient of determination and root mean square error as evaluation indicators, if the determination coefficient If the value is less than 0.9 or the RMS error is greater than 0.04, the training will continue until the performance evaluation criteria are met.

9. A forecasting method for managing the allocation of aviation turnover parts according to claim 2, characterized in that: The step S5 comprises: The number of flights of the large passenger aircraft after the last replacement of the turbine blades, as well as the flight duration, flight speed and temperature data of the vortex blades corresponding to each of the flight numbers are assigned as first target input data to the vortex blade damage prediction model for damage prediction; Assigning the number of taxiing of the large passenger aircraft after the last replacement of the landing support rod, and the taxiing duration, taxiing speed and acceleration during landing corresponding to each of the taxiing numbers as second target input data to the landing component damage prediction model for damage prediction; The number of taxiing of the large passenger aircraft after the most recent replacement of the brake disc, as well as the taxiing duration, taxiing speed and acceleration during landing corresponding to each of the taxiing numbers are assigned as third target input data to the brake disc damage prediction model for damage prediction.

10. A forecasting method for managing the allocation of aviation turnover parts according to claim 9, characterized in that: The step S8 comprises: When the gap between adjacent blade tips of the vortex fan blades deviates from the preset minimum replacement damage gap by more than 5%, and / or the thickness of the brake disc is lower than the preset minimum replacement damage thickness, and / or the compression damage deformation and bending damage deformation of the landing support rod exceed the preset threshold, the state of the notification switch that triggers the resource allocation notification model is on.

Images