A method for analyzing flight incidents
By dividing the eye-tracking gaze area, collecting and processing pilot eye-tracking data, and using the FCM algorithm and fuzzy width learning system, the flight roles of the pilot and co-pilot are simulated, solving the problem of the lack of common operation simulation in existing technologies and improving flight safety and the accuracy of environmental simulation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CIVIL AVIATION FLIGHT UNIV OF CHINA
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies in simulation studies only focus on the flight operations of a single pilot, lacking simulations of the joint operations of the pilot and co-pilot in actual passenger aircraft, and failing to effectively reflect the differences in visual attention and operation between the two in special flight situations, thus affecting flight safety.
By dividing the eye-tracking gaze region, collecting pilot eye-tracking data, performing preprocessing and feature extraction, using the FCM algorithm to calculate membership degrees and generate adjusted weight values, and combining this with a fuzzy width learning system for classification and evaluation, the flight roles of pilots and co-pilots are simulated, improving the environmental simulation effect of flight missions.
It effectively reduces the impact of individual differences on eye-tracking data, improves the accuracy of flight environment simulation, enhances pilot operational compatibility, provides flight warnings, and improves passenger aircraft flight safety.
Smart Images
Figure CN122156799A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of aviation flight safety technology, and in particular relates to a method for analyzing flight emergencies. Background Technology
[0002] Flight safety is closely related to the pilot's flight status and flight operations. During flight, special circumstances such as single-engine failure or attitude control system malfunction can affect flight attitude and airframe stability. These special circumstances increase the difficulty of pilot operations, thus seriously threatening flight safety. Existing simulation studies focus on monitoring the operations of a single pilot during special circumstances. However, in actual passenger aircraft flights, the flight mission is usually carried out jointly by the pilot (PF) and the co-pilot (PM). Simulating only the PF lacks a complete simulation and representation of the actual flight environment. Furthermore, since the visual attention of the PF and PM differs significantly at different stages of flight, simple data statistics cannot clearly demonstrate the impact of their visual attention and operations on the handling of special circumstances. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention proposes a flight emergency analysis method to resolve the issues present in the prior art.
[0004] To achieve the above objectives, the present invention provides a flight emergency analysis method, comprising: Divide the eye-tracking fixation areas and collect eye-tracking data from the pilots in different eye-tracking fixation areas; The eye movement data is preprocessed to obtain eye movement features; Calculate the membership degree of eye movement features from the cluster center in different eye movement fixation regions, and generate an adjusted weight value based on the membership degree; The eye movement features of the pilot under test are adjusted based on the adjusted weight values, and a classification model is used to classify and evaluate the pilot under test; wherein the classification model is a fuzzy width learning system.
[0005] Optionally, the defined eye-tracking gaze areas include: the in-cabin instrument area, the out-of-cabin scene area, and the operating area; The pilots include: the pilot and the co-pilot.
[0006] Optionally, the process of preprocessing the eye-tracking data to obtain eye-tracking features includes: The collected eye movement data is preliminarily processed to extract fixation area data and generate eye movement trajectory maps and eye movement trajectory recording data; The eye-tracking data is denoised to obtain denoised data. Record fixation time, first fixation time, and number of visits within different eye-tracking fixation areas; Based on the fixation time, first fixation time, and number of visits, eye movement features are extracted from the denoised data.
[0007] Optionally, the eye movement features include: the variance of the number of times the pilot visits each eye movement fixation area, the Markov transition probability, the emergency reaction time, and the average fixation duration.
[0008] Optionally, the process of calculating the membership degree of eye movement features from the cluster center in different eye movement fixation regions, and generating adjusted weight values based on the membership degree, includes: The training datasets were based on historical data of the pilot and co-pilot during flight emergencies, respectively. The training dataset is input into the FCM model to obtain cluster centers, and the corresponding membership degrees are calculated. An adjusted weight value is generated based on the membership degree; The adjusted weight values are fitted using the maximum likelihood estimation method to obtain the adjusted weight value matrix.
[0009] Optionally, the expression for generating the adjusted weight value based on the membership degree is: ; In the formula, , , These are the first error constant term, the second error constant term, and the third error constant term, respectively. , , The first adjustment weight value, the second adjustment weight value, and the third adjustment weight value are respectively. U1, U2, and U3 are the membership values corresponding to the instrument area inside the cabin, the scene area outside the cabin, and the operation area, respectively. X1(i), X2(i), and X3(i) are the feature data of the instrument area inside the cabin, the scene area outside the cabin, and the operation area, respectively.
[0010] Optionally, the adjusted weight values are fitted based on the maximum likelihood estimation method, and the expression for calculating the adjusted weight value matrix is as follows: ; In the formula, k is the number of independent variable factors, and L is the maximum likelihood estimate. For variance, Factors for evaluating and adjusting weight values.
[0011] Optionally, the classification model is a fuzzy width learning system, and the process of classifying and evaluating the pilot under test using the classification model includes: The eye-tracking features are weighted based on the adjusted weight matrix, and then processed using a trained fuzzy width learning system to obtain the role classification results for the driver and co-driver. Based on the role classification results, the values in the classification matrix are extracted as evaluation values; Based on the aforementioned evaluation values, a weighted summation calculation is used to obtain the comprehensive effectiveness evaluation value for handling flight emergencies.
[0012] The present invention also provides a computer terminal device, comprising: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method.
[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.
[0014] Compared with the prior art, the present invention has the following advantages and technical effects: This invention proposes a flight emergency analysis method. By simulating the flight roles of the pilot and co-pilot in an actual passenger aircraft, it simulates aviation missions when flight emergencies occur, improving the simulation effect of the flight environment in actual passenger aircraft flight missions. In addition, after obtaining the eye movement feature data of the pilot and co-pilot through feature processing, the eye movement data of the pilot and co-pilot are clustered using the FCM algorithm to obtain cluster centers. The clustering effect is used to assist in the calculation and assignment of the eye movement features of the tested pilots, thereby effectively reducing the influence of eye movement data caused by individual differences. This assignment method can effectively assign values to the eye movement attention area and eye movement features of each pilot. Finally, the data after feature assignment is classified and evaluated through a fuzzy width learning system, which can effectively match the flight attention allocation and operation effect of the pilot and co-pilot, and provide operation reminders when passenger aircraft pilots have two pilot roles. Attached Figure Description
[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a structural diagram of a flight emergency analysis method according to an embodiment of the present invention. Detailed Implementation
[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0018] This embodiment provides a method for analyzing flight emergencies, including the following steps: This embodiment proposes a flight emergency analysis method. During actual passenger aircraft flight, the pilot (PF) typically focuses their visual attention primarily on the external field of vision, flight control, and the left-hand display area, while the co-pilot (PM) focuses more on the right-hand display area, checklists, and other monitoring tasks. In certain flight phases, such as takeoff and landing, a statistically significant correlation is observed between the gaze indices (e.g., gaze duration) of the PF and PM. To further utilize this correlation to improve the effectiveness of flight emergency handling during passenger aircraft flight, cluster analysis and feature assignment are used to perform synchronous feature analysis and evaluation prediction for both the pilot and co-pilot roles, thereby improving flight safety.
[0019] like Figure 1 The diagram shown is a structural diagram of a flight emergency analysis method according to this embodiment. The main steps of the method are as follows: S1. Divide the eye-tracking fixation area and use an eye tracker to collect eye movement data of the pilot in different eye-tracking fixation areas.
[0020] S2. Preprocess the eye movement data and extract eye movement features. This process includes: performing preliminary processing on the collected eye movement data, extracting fixation area data, generating eye movement trajectory maps and eye movement trajectory recording data; performing noise reduction processing on the eye movement trajectory recording data to obtain noise-reduced data; recording fixation time, first fixation time, and number of visits in different eye movement fixation areas; and extracting eye movement features from the noise-reduced data based on fixation time, first fixation time, and number of visits.
[0021] S3. Input the eye movement features into the fuzzy clustering module, calculate the membership degree of the feature distance of different eye movement gaze regions to the cluster center, and generate an adjusted weight value based on the membership degree.
[0022] S4. Adjust the weight values to assign values to the eye movement features of the pilot under test, and then use the fuzzy width learning system as a classification model to classify and evaluate the pilot under test.
[0023] The experiment used a desktop primary flight program trainer with Microsoft Flight Simulator 2020 software, a TobiiGlasses 3 Eyetracker, and flight simulation scoring software to collect pilots' eye tracker data during flight. The pilots' roles included the pilot (PF) and the co-pilot (PM). In other words, both the experimental training set data and the test set data were composed of the pilot (PF) and the co-pilot (PM).
[0024] Next, the flight path and mission were designed, and the operations that the pilots needed to perform during flight were defined. Both the primary and co-pilots were equipped with eye trackers, and their eye movement data was stored separately in the system's backend. The pilots performed the flight mission on a simulation platform according to the mission requirements, and eye movement data was acquired using an eye-tracking device during the experimental flight mission. Furthermore, during the approach phase of the flight mission, the pilots were not informed in advance of any upcoming flight emergencies.
[0025] Preliminary processing of eye-tracking data involves extracting fixation area data, visualizing the eye-tracking data, and generating trajectory maps of the eye-tracking data along with recorded data of the pilot's eye-tracking trajectories. Furthermore, noise reduction processing is required to eliminate the influence of environmental factors, nystagmus, microsaccades, and other influences. In the preprocessing steps, to facilitate feature extraction, fixation time, initial fixation time, and number of visits within different defined eye-tracking fixation areas are recorded for subsequent processing. The extracted eye-tracking features used for analysis and classification include: the variance of the number of times the pilot visits each eye-tracking fixation area (VAR), the Markov transition probability (MK), the emergency reaction time (RT), and the average fixation duration (EPT).
[0026] Specifically, the method for extracting the above features is as follows: The formula for calculating VAR is: VC represents the number of times the pilot visited each attention instrument distribution area. denoted as MK, where MK is the average number of visits by all participating pilots, and n is the total number of areas of interest. The formula for calculating MK is: ,in Let be the Markov transition probability of the nth region of interest when a special situation occurs. The duration of fixation in region of interest j after the pilot's attention moves from region i to region j. The total duration of gaze at all remaining regions of interest after the pilot's gaze point is moved out of region i; , where TM is the time point when the test pilot discovered the special situation, and T0 is the actual time point when the special situation occurred; EPT is the average fixation duration at each fixation point. These values are then normalized using a commonly used method.
[0027] Then, in the fuzzy clustering module, the FCM algorithm is used to analyze the eye movement data of each eye movement fixation area of the driver (PF) and the co-driver (PM) to obtain the corresponding membership values. The eye movement features extracted from the eye movement fixation area are used as independent variable factors, and the membership values are used as dependent variable factors. The weight values of each independent variable factor are automatically calculated through the statistical model.
[0028] The specific steps for automatically calculating the weight values of each independent variable factor using a statistical model are as follows: 1) Use historical data of the pilot and co-pilot at the time of flight emergencies as training datasets; 2) Input the training datasets into the FCM model to obtain cluster centers and calculate the corresponding membership degrees; 3) Generate adjusted weight values based on the membership degrees; 4) Fit the adjusted weight values using the maximum likelihood estimation method to obtain the adjusted weight value matrix.
[0029] S3.1, using historical data of normal and abnormal handling results of pilot PF and co-pilot PM when flight emergencies occur as training dataset, the feature data of the cockpit instrument area are defined as X1(i), the external scene area X2(i) and the operation area X3(i), i=VAR, MK, RT, EPT; that is, the feature dimension of each pilot input is 12-dimensional.
[0030] S3.2 After inputting the training dataset into the FCM model to obtain cluster centers, X1(i), X2(i), and X3(i) are input into the FCM model to calculate the corresponding membership values U1, U2, and U3. This method first inputs the partition features of each eye-tracking fixation region, determines the distance between different regions and the cluster centers, and performs preliminary weighting adjustments on the eye-tracking fixation regions based on the distance values. The resulting number of cluster centers is 4, representing the cluster centers of the driver PF when the processing result is normal, the cluster centers of the driver PF when the processing result is abnormal, the cluster centers of the co-driver PM when the processing result is normal, and the cluster centers of the co-driver PM when the processing result is abnormal.
[0031] In one embodiment, after inputting the pilot (PM) and co-pilot (PF), the membership values of PM are U1=0.7324, U2=0.4921, and U3=0.1328; the membership values of PF are U1=0.4722, U2=0.8200, and U3=0.2873. That is, this membership calculation method can reflect the differences in the areas of focus for the pilot (PM) and co-pilot (PF) when actual flight emergencies occur. Therefore, to address these differences, the feature values for each pilot in each eye-tracking fixation area need to be adjusted individually.
[0032] S3.3, Generate adjustment weight values based on membership degrees, and set the calculation formula: ; in, , , For the error constant term, , , Adjust the weight values for the corresponding feature values.
[0033] This step involves classifying the membership of each pilot in different eye-tracking fixation regions, then further adjusting the weight values of the eye-tracking features in each eye-tracking fixation region, and performing noise reduction on the eye-tracking features while taking into account the influence of the eye-tracking regions.
[0034] S3.4, the above weight values are fitted using the maximum likelihood estimation method, and the maximum likelihood estimate L and variance are obtained. Then calculate the formula Where k is the number of independent variable factors, As an evaluation factor, when The calculation is completed when the value of is minimized. The weights when the value is minimum are obtained. , , As an adjustment weight value.
[0035] Finally, the adjusted weight matrix can be obtained, which will... Each dimension is assigned a corresponding feature and its weight is adjusted accordingly.
[0036] Then, historical data on the normal and abnormal handling results of pilot PF and co-pilot PM during flight emergencies were used as the training dataset to train the fuzzy width learning system. Classification accuracy was set as the evaluation criterion for model training completion. Model training was stopped when the model classification accuracy was greater than 95%. Finally, the number of fuzzy rules NR in the fuzzy width learning network was set to 11, the number of fuzzy systems NF to 9, and the number of augmentation nodes NE to 13.
[0037] After the model completes the above training, the feature datasets of all test pilots are adjusted with weights and then input into the trained classification model as the test set to classify and evaluate the test pilots.
[0038] The process of classifying and evaluating the pilots under test using a classification model includes: adjusting the weights of eye-tracking features based on the weight matrix, processing the data using a trained fuzzy width learning system to obtain the role classification results for the pilot and co-pilot; extracting the values from the classification matrix as evaluation values based on the role classification results; and calculating the comprehensive evaluation value of the flight emergency handling effect using a weighted summation based on the evaluation values.
[0039] The fuzzy width learning system classifies the two roles, the pilot (PF) and the co-pilot (PM), to initially determine whether the operation is abnormal or normal. Then, it uses the classification matrix value in the classification result as the evaluation value, evaluates the effect of the emergency handling based on the evaluation results of the two roles, and issues a flight warning signal to the pilot based on the evaluation of the effect of the emergency handling.
[0040] Specifically, since FBLS is a network variation based on the BLS model, its main difference lies in the construction of the intermediate fuzzy layer. However, the network width setting and the inverse operation of the coefficient matrix are consistent. Therefore, in terms of network output representation, the FBLS model can also be expressed as follows: In the feature layer and output layer, the feature node matrix y is set to have dimensions s×(N2×N1), where s is the scaling factor of the augmented nodes. The fuzzy system node matrix y is standardized and augmented to obtain H2. Then, the coefficient matrix wh of the augmented nodes can be represented as a (N2×N1)×N3 dimensional orthogonally normalized random matrix. Then, the tansig activation function is used to activate the augmented nodes, and the network output is generated. Where xx is the final classification result matrix, with values of 0 or 1, where 0 represents normal operation and 1 represents abnormal operation, and W is the calculation process matrix of the fuzzy system. As the input for the final generation network, It includes the numerical values of the classification matrix.
[0041] In this embodiment, the classification results for the pilot (PF) and co-pilot (PM) are xx=[0,0], meaning both the pilot and co-pilot are normal, T3=[0.82, 0.77], and the final processing effect evaluation for this flight is S=0+0+0.82. 0.5 + 0.77 0.5 = 0.795. Based on the comprehensive experimental data, when the final processing evaluation value is greater than 1.42, the system should issue a warning such as a buzzer to remind the pilot and co-pilot in the cockpit of a flight attention error, thereby improving the accuracy of emergency handling and enhancing the flight safety performance of the passenger aircraft.
[0042] The present invention also provides a computer terminal device, comprising: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method.
[0043] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.
[0044] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for analyzing flight special situations, characterized in that, Includes the following steps: Divide the eye-tracking fixation areas and collect eye-tracking data from the pilots in different eye-tracking fixation areas; The eye movement data is preprocessed to obtain eye movement features; Calculate the membership degree of eye movement features from the cluster center in different eye movement fixation regions, and generate an adjusted weight value based on the membership degree; The eye movement characteristics of the pilot under test are adjusted based on the adjusted weight values, and a classification model is used to classify and evaluate the pilot under test. The classification model described therein is a fuzzy width learning system.
2. The flight emergency analysis method according to claim 1, characterized in that, The eye-tracking fixation areas are divided into: the instrument area inside the cabin, the scene area outside the cabin, and the operating area; The pilots include: the pilot and the co-pilot.
3. The flight emergency analysis method according to claim 1, characterized in that, The process of preprocessing the eye-tracking data to obtain eye-tracking features includes: The collected eye movement data is preliminarily processed to extract fixation area data and generate eye movement trajectory maps and eye movement trajectory recording data; The eye-tracking data is denoised to obtain denoised data. Record fixation time, first fixation time, and number of visits within different eye-tracking fixation areas; Based on the fixation time, first fixation time, and number of visits, eye movement features are extracted from the denoised data.
4. The flight emergency analysis method according to claim 3, characterized in that, The eye movement features include: the variance of the number of times the pilot visits each eye movement fixation area, the Markov transition probability, the emergency reaction time, and the average fixation duration.
5. The flight emergency analysis method according to claim 1, characterized in that, The process of calculating the membership degree of eye movement features from cluster centers in different eye-movement fixation regions, and generating adjusted weight values based on the membership degree, includes: The training datasets were based on historical data of the pilot and co-pilot during flight emergencies, respectively. The training dataset is input into the FCM model to obtain cluster centers, and the corresponding membership degrees are calculated. An adjusted weight value is generated based on the membership degree; The adjusted weight values are fitted using the maximum likelihood estimation method to obtain the adjusted weight value matrix.
6. The flight emergency analysis method according to claim 5, characterized in that, The expression for generating the adjusted weight value based on the membership degree is: ； In the formula, , , These are the first error constant term, the second error constant term, and the third error constant term, respectively. , , The first adjustment weight value, the second adjustment weight value, and the third adjustment weight value are respectively. U1, U2, and U3 are the membership values corresponding to the instrument area inside the cabin, the scene area outside the cabin, and the operation area, respectively. X1(i), X2(i), and X3(i) are the feature data of the instrument area inside the cabin, the scene area outside the cabin, and the operation area, respectively.
7. The flight emergency analysis method according to claim 6, characterized in that, The adjusted weight values are fitted using the maximum likelihood estimation method, and the expression for the adjusted weight value matrix is as follows: ； In the formula, k is the number of independent variable factors, and L is the maximum likelihood estimate. For variance, Factors for evaluating and adjusting weight values.
8. The flight emergency analysis method according to claim 7, characterized in that, The process of classifying and evaluating pilots using a classification model includes: The eye-tracking features are weighted based on the adjusted weight matrix, and then processed using a trained fuzzy width learning system to obtain the role classification results for the driver and co-driver. Based on the role classification results, the values in the classification matrix are extracted as evaluation values; Based on the aforementioned evaluation values, a weighted summation calculation is used to obtain the comprehensive effectiveness evaluation value for handling flight emergencies.
9. A computer terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-8.