A method for recognizing a dangerous action of a pilot
Patent Information
- Application Number
- CN202610937793.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-26
- Publication Date
- 2026-10-27
AI Technical Summary
[0003]然而，现有的飞行安全防护体系主要依赖于静态的包线保护或单一阈值的超限告警，其响应机制相对滞后，难以在危险发生的萌芽阶段进行有效识别与阻断
[0026]本申请能够保障通用飞机的飞行安全、降低飞行员的驾驶负担、提升使用效能。
Smart Images
Figure CN122986576A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of flight safety design technology, and specifically relates to a method for identifying dangerous actions of pilots. Background Technology
[0002] With the rapid development of the general aviation industry, modern aircraft are evolving rapidly towards high performance and multi-purpose capabilities. The widespread application of variable configurations, wide speed ranges, and high-maneuverability layouts, while significantly expanding the flight envelope, also poses unprecedented challenges to flight control system design and flight safety assurance. In high-maneuverability, high-reliability flight missions, pilots often face complex situational judgments and precise control requirements, making them highly susceptible to errors due to reaction delays, environmental perception biases, or improper handling. These errors often lead to uncontrollable states such as stall, deviation, post-stall spin, or even tailspin, seriously threatening flight safety. The industry typically defines such potentially serious maneuvers as dangerous actions.
[0003] However, existing flight safety protection systems mainly rely on static envelope protection or single-threshold over-limit alarms. Their response mechanisms are relatively lagging, making it difficult to effectively identify and prevent hazards in their early stages. Especially when dealing with dangerous maneuvers involving dynamic characteristics, such as human-machine interface (PIO), traditional methods lack the ability to extract and analyze the frequency domain characteristics of the maneuvers in real time, failing to accurately distinguish between normal maneuvers and dangerous actions that could lead to loss of control. Furthermore, for low-energy states during approach or boundary approaches in complex maneuvers, existing systems mostly rely on post-event alarms, lacking predictive and proactive judgment capabilities. These technological limitations make it difficult for flight control systems to provide sufficiently accurate and timely safety protection in the face of complex and ever-changing real-world flight scenarios. Summary of the Invention
[0004] To address the aforementioned problems, this application provides a method for identifying dangerous actions by pilots, mainly including:
[0005] Step S1: Based on the flight safety restrictions during flight, determine the dangerous actions and operations required for a flight to enter a dangerous state, and establish a dangerous action library that includes dangerous states, dangerous actions, and dangerous operations;
[0006] Step S2: Based on flight test data, extract the action feature parameters of dangerous or normal actions in real time. The action feature parameters include feature parameters in the pilot's control input signals and aircraft response signals. The feature parameters include time domain features and frequency domain features.
[0007] Step S3: Establish a quantitative correspondence between action characteristic parameters and flight safety boundaries through mathematical simulation, and determine the criteria for judging dangerous or normal actions;
[0008] Step S4: Construct a relationship model between action feature parameters and the identification logic of dangerous or normal actions based on the judgment criteria, and perform real-time online identification of dangerous actions based on the relationship model.
[0009] Preferably, in step S1, the flight safety restrictions include angle of attack restrictions, load restrictions, flight energy restrictions, altitude restrictions, and speed restrictions. When the flight state exceeds any of the safety restrictions, the flight is determined to be in a dangerous state.
[0010] Preferably, in step S2, the characteristic parameters of the pilot control input signal include the pilot control amplitude, control rate, and control frequency; the characteristic parameters of the aircraft response signal include angle of attack, overload, speed, altitude, and track angle.
[0011] Preferably, in step S2, the motion feature parameters are processed using the sliding window Fourier transform method to extract frequency domain feature parameters in real time.
[0012] Preferably, the sliding window Fourier transform method specifically includes:
[0013] Step S21: Select a time window of sample data from the signal and calculate the difference between the sample and its regression line;
[0014] Step S22: Perform a Fourier transform on the difference to obtain the spectral characteristics of the signal within the window;
[0015] Step S23: Control the window to slide forward, remove the first data point in the window, add a new data point, and update the spectral characteristics.
[0016] Preferably, the relational model in step S4 is a fuzzy inference model.
[0017] Preferably, step S4 further includes:
[0018] Based on the combination of action feature parameters and their membership functions, the membership degree of each parameter is obtained;
[0019] The parameters of each fuzzy rule are calculated based on the membership degree. The fuzzy rules include positive rules that can increase the estimated value of dangerous actions and negative rules that can reduce the estimated value of dangerous actions.
[0020] Calculate the expected dangerous action based on the formula. :
[0021] ;
[0022] in, To coordinate parameters, For the first One regular parameter The weight, For positive regularity, For the first Negative rule parameters The weight, The number of negative rules;
[0023] If the expected dangerous action exceeds the set value, a dangerous action is determined to have occurred; otherwise, no dangerous action has occurred.
[0024] Preferably, step S4 further includes:
[0025] Upon determining that a dangerous action has been taken, an alarm is triggered or intervention measures to disconnect the driver from the vehicle are taken.
[0026] This application can ensure the flight safety of general aviation aircraft, reduce the pilot's workload, and improve the efficiency of use. Attached Figure Description
[0027] Figure 1 This is a flowchart of a preferred embodiment of the pilot dangerous maneuver identification method of this application.
[0028] Figure 2 This is a flowchart of the calculation process for the fuzzy logic method. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of this application, not all of them. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0030] This application provides a method for identifying dangerous actions by pilots, such as... Figure 1 As shown, it mainly includes:
[0031] Step S1: Based on the flight safety restrictions during flight, determine the dangerous actions and operations required for a flight to enter a dangerous state, and establish a dangerous action library that includes dangerous states, dangerous actions, and dangerous operations;
[0032] Step S2: Based on flight test data, extract the action feature parameters of dangerous or normal actions in real time. The action feature parameters include feature parameters in the pilot's control input signals and aircraft response signals. The feature parameters include time domain features and frequency domain features.
[0033] Step S3: Establish a quantitative correspondence between action characteristic parameters and flight safety boundaries through mathematical simulation, and determine the criteria for judging dangerous or normal actions;
[0034] Step S4: Construct a relationship model between action feature parameters and the identification logic of dangerous or normal actions based on the judgment criteria, and perform real-time online identification of dangerous actions based on the relationship model.
[0035] This application addresses the flight control requirements of general aviation aircraft, aiming to balance high reliability and high maneuverability. It conducts research on pilot dangerous action identification technology and proposes a research scheme with engineering feasibility. Specifically, step S1 primarily defines dangerous actions and establishes a database of typical dangerous tasks. Step S2 aims to extract feature parameters from flight test data, including normal and dangerous actions. Step S3 aims to establish a mapping relationship between feature parameters and flight boundaries through mathematical simulation. Using the features extracted in step S2 and combined with the dangerous action database from step S1, dangerous action judgment criteria are obtained. Finally, in step S4, the criteria from step S3 are applied to determine whether an action is dangerous.
[0036] In some optional implementations, in step S1, the flight safety restrictions include angle of attack restrictions, load restrictions, flight energy restrictions, altitude restrictions, and speed restrictions. When the flight condition exceeds any of the safety restrictions, the flight is determined to be in a dangerous state.
[0037] During flight, to ensure flight safety and controllability, flight state variables must remain within multiple safety boundaries. Exceeding these boundaries will lead to dangerous situations. For example, excessive angle of attack can easily cause the aircraft to stall, or even veer off course and spin; or, performing a dive recovery maneuver at low altitude may result in a crash due to insufficient altitude margin; and high-frequency, large-amplitude pilot maneuvers may induce pitch-in-the-air (PIO) phenomena. Therefore, in step S1 of this application, a series of safety boundaries during flight must first be organized and summarized. Based on these boundaries, flight states that are likely to exceed these boundaries are defined as dangerous states.
[0038] Once a hazardous situation is defined, specific hazardous maneuvers that meet the definition can be selected and organized based on the pilot's flight course. These include hazardous maneuvers that trigger flight state limitations during maneuvers, such as half-loops and convergent turns; hazardous maneuvers that trigger flight state (low energy) limitations during approaches, such as precise offset landings; and hazardous maneuvers that trigger aircraft dynamic characteristic limitations during maneuvers, such as pitch tracking.
[0039] Based on specific dangerous flight maneuvers and the flight safety boundaries they may exceed, the corresponding specific operations that may threaten flight safety are determined. For example, pulling over the stick at low speeds may cause the aircraft's angle of attack to increase rapidly, leading to stall and affecting flight safety. Or, during a low-altitude dive recovery maneuver, insufficient pull-up may easily cause a crash. Thus, a dangerous maneuver library containing dangerous states and corresponding dangerous operations is established.
[0040] Step S2 is used to extract action feature parameters. In some optional embodiments, in step S2, the feature parameters of the pilot control input signal include the pilot control amplitude, control rate and control frequency; the feature parameters of the aircraft response signal include angle of attack, overload, speed, altitude and track angle.
[0041] In some optional implementations, in step S2, the motion feature parameters are processed using the sliding window Fourier transform method to extract frequency domain feature parameters in real time.
[0042] In some optional implementations, the sliding window Fourier transform method specifically includes:
[0043] Step S21: Select a time window of sample data from the signal and calculate the difference between the sample and its regression line;
[0044] Step S22: Perform a Fourier transform on the difference to obtain the spectral characteristics of the signal within the window;
[0045] Step S23: Control the window to slide forward, remove the first data point in the window, add a new data point, and update the spectral characteristics.
[0046] Since observing the characteristics of dangerous maneuvers directly from the time history curves of pilot inputs and aircraft response signals is not readily apparent, numerical transformation is necessary to enhance the feature information. Furthermore, the feature codes of dangerous maneuvers typically contain information in both the time and frequency domains, and accurate and timely extraction of these parameters is crucial when issuing dangerous maneuver warnings. Therefore, the method for extracting dangerous maneuver feature parameters must effectively resolve the contradictions between time and frequency domain localization, accurately and quickly extracting the feature parameters of dangerous maneuvers.
[0047] This application employs a sliding window Fourier transform method to extract feature parameters. The sliding window Fourier transform selects a small time window within the signal and performs a Fourier transform to obtain the spectral characteristics of the signal within that window. To facilitate handling signal variations, a sample data point within the window is initially selected, and the difference between that sample and its regression line is calculated. This difference is then subjected to a Fourier transform. Subsequently, as the window slides forward, the first data point in the window is removed, the remaining data points are retained, and a new data point is added at the end of the window. The new data point is then subjected to a Fourier transform along with the remaining samples, continuously updating the frequency domain characteristics. This method enables real-time acquisition of the signal's spectral characteristics and is suitable for analyzing dynamically changing signals.
[0048] In step S3, mathematical simulation is performed to obtain the criteria for judging dangerous actions. In this step, stability and maneuverability are typically determined through mathematical simulation of a six-degree-of-freedom nonlinear model of the aircraft. First, a six-degree-of-freedom nonlinear mathematical model of the general aviation aircraft is established. Then, by analyzing the aircraft's body parameters, speed limits, altitude limits, and structural load limits are determined. Next, under different maneuvering flight states, different control commands are input, and mathematical simulations are performed on the aircraft. Combining these with the aircraft's speed limits, altitude limits, and structural load limits, pilot control boundary values related to the aircraft's thrust, aerodynamic coefficient, mass, inertial characteristics, and other body parameters are determined. Finally, the data is processed to establish a pilot control envelope corresponding to the aircraft's maneuvering flight envelope, providing a basis for identifying dangerous maneuvers.
[0049] Finally, dangerous actions are identified in step S4.
[0050] In some optional implementations, the relational model in step S4 is a fuzzy inference model. Dangerous action recognition is based on pattern recognition technology and employs fuzzy logic methods for online dangerous action recognition. The calculation flow of the fuzzy logic method is as follows: Figure 2 As shown. First, based on the combination of dangerous action parameters obtained through real-time processing of the characteristic parameter signals of dangerous actions, the membership function of each parameter is determined, thereby obtaining the membership degree of each parameter; then, reasoning is performed according to fuzzy logic rules to obtain the estimated value of the dangerous action, and to determine whether a dangerous action has occurred.
[0051] In some alternative implementations, step S4 further includes:
[0052] Based on the combination of action feature parameters and their membership functions, the membership degree of each parameter is obtained;
[0053] The parameters of each fuzzy rule are calculated based on the membership degree. The fuzzy rules include positive rules that can increase the estimated value of dangerous actions and negative rules that can reduce the estimated value of dangerous actions.
[0054] Calculate the expected dangerous action based on the formula. :
[0055] ;
[0056] in, To coordinate parameters, For the first One regular parameter The weight, For positive regularity, For the first Negative rule parameters The weight, The number of negative rules;
[0057] If the expected dangerous action exceeds the set value, a dangerous action is determined to have occurred; otherwise, no dangerous action has occurred.
[0058] It should be noted that membership functions can map feature parameters to a fuzzy logic space, obtaining the membership degree for determining dangerous actions based on those feature parameters. Combining the membership degrees of all feature parameters, a judgment on whether an action is dangerous is then made according to logical rules. These logical rules consist of positive and negative rules. Positive rules are used to identify dangerous action characteristics, while negative rules are used to distinguish and identify normal flight characteristics; that is, positive rules increase the estimated value of dangerous actions, while negative rules decrease it. Finally, the expected value of the dangerous action is used to determine whether an action is dangerous. For example, if the value is set to 0.5, then if the calculated expected value of the dangerous action exceeds 0.5, the action is considered dangerous; otherwise, it is considered not dangerous.
[0059] This fuzzy logic method can identify dangerous maneuvers under different triggering conditions: For dangerous maneuvers that trigger flight state limitations during maneuvers (e.g., half-loop, convergent turn, etc.), it extracts real-time aircraft time-domain data such as flight angle of attack, overload, speed, altitude, and track angle; calculates real-time overload prediction values, angle of attack prediction values, the minimum distance between the predicted recovery altitude and the altitude boundary, and the predicted recovery maximum speed, and substitutes these values into the dangerous maneuver judgment criteria that trigger angle of attack, overload, altitude, and speed limitations, thereby achieving the identification and early warning of dangerous maneuvers that trigger angle of attack, overload, altitude, and speed limitations during maneuvers; For dangerous maneuvers that trigger flight state (low energy) limitations during approach (e.g., precise offset landing, etc.), it can identify dangerous maneuvers that trigger flight state (low energy) limitations during approach. This method extracts real-time aircraft time-domain data such as speed, altitude, overload, and track angle to calculate and predict the minimum distance between the recoverable altitude and the lower boundary of the glide slope, as well as the minimum recoverable speed. These parameters are then incorporated into the criteria for identifying dangerous maneuvers that trigger low-energy limits, enabling the identification and early warning of such maneuvers during approach. For dangerous maneuvers that trigger aircraft dynamic characteristic limits (e.g., pitch tracking), a sliding window Fourier transform is used for time-frequency transformation to calculate the human-machine coupling frequency and the phase lag of attitude angles on control. Combined with the maximum pilot control amplitude and maximum control surface rate within the corresponding time window obtained from the time-domain response, typical PIO characteristic parameters are extracted for identification. In summary, this dangerous maneuver identification method can identify three major categories of dangerous maneuvers with different triggering conditions.
[0060] In some alternative implementations, step S4 is further followed by:
[0061] Upon determining that a dangerous action has been taken, an alarm is triggered or intervention measures to disconnect the driver from the vehicle are taken.
[0062] This application employs a fuzzy logic method for online identification of dangerous actions. This method transforms and processes signals into characteristic parameters for identifying dangerous actions, establishes a fuzzy logic inference system, and obtains an estimate of the dangerous action. This application fills a gap in the research field of dangerous action identification methods for pilots, supplements and improves existing flight safety protection systems, provides a new method for real-time flight safety monitoring of various types of aircraft, ensures flight safety of general aviation aircraft, reduces pilot workload, and improves operational efficiency.
[0063] The above description is merely a specific embodiment 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 technical scope 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 identifying dangerous actions by pilots, characterized in that, include: Step S1: Based on the flight safety restrictions during flight, determine the dangerous actions and operations required for a flight to enter a dangerous state, and establish a dangerous action library that includes dangerous states, dangerous actions, and dangerous operations; Step S2: Based on flight test data, extract the action feature parameters of dangerous or normal actions in real time. The action feature parameters include feature parameters in the pilot's control input signals and aircraft response signals. The feature parameters include time domain features and frequency domain features. Step S3: Establish a quantitative correspondence between action characteristic parameters and flight safety boundaries through mathematical simulation, and determine the criteria for judging dangerous or normal actions; Step S4: Construct a relationship model between action feature parameters and the identification logic of dangerous or normal actions based on the judgment criteria, and perform real-time online identification of dangerous actions based on the relationship model.
2. The method for identifying dangerous pilot maneuvers as described in claim 1, characterized in that, In step S1, the flight safety restrictions include angle of attack restrictions, load restrictions, flight energy restrictions, altitude restrictions, and speed restrictions. When the flight status exceeds any of the safety restrictions, the flight is determined to be in a dangerous state.
3. The method for identifying dangerous pilot maneuvers as described in claim 1, characterized in that, In step S2, the characteristic parameters of the pilot control input signal include the pilot control amplitude, control rate, and control frequency; the characteristic parameters of the aircraft response signal include angle of attack, overload, speed, altitude, and track angle.
4. The method for identifying dangerous pilot maneuvers as described in claim 1, characterized in that, In step S2, the motion feature parameters are processed using the sliding window Fourier transform method to extract frequency domain feature parameters in real time.
5. The method for identifying dangerous pilot maneuvers as described in claim 4, characterized in that, The sliding window Fourier transform method specifically includes: Step S21: Select a time window of sample data from the signal and calculate the difference between the sample and its regression line; Step S22: Perform a Fourier transform on the difference to obtain the spectral characteristics of the signal within the window; Step S23: Control the window to slide forward, remove the first data point in the window, add a new data point, and update the spectral characteristics.
6. The method for identifying dangerous pilot maneuvers as described in claim 1, characterized in that, The relational model in step S4 is a fuzzy inference model.
7. The method for identifying dangerous pilot maneuvers as described in claim 6, characterized in that, Step S4 further includes: Based on the combination of action feature parameters and their membership functions, the membership degree of each parameter is obtained; The parameters of each fuzzy rule are calculated based on the membership degree. The fuzzy rules include positive rules that can increase the estimated value of dangerous actions and negative rules that can reduce the estimated value of dangerous actions. Calculate the expected dangerous action based on the formula. : ； in, To coordinate parameters, For the first One regular parameter The weight, For positive regularity quantity, For the first Negative rule parameters The weight, The number of negative rules; If the expected dangerous action exceeds the set value, a dangerous action is determined to have occurred; otherwise, no dangerous action has occurred.
8. The method for identifying dangerous pilot maneuvers as described in claim 1, characterized in that, Step S4 is followed by the following: Upon determining that a dangerous action has been taken, an alarm is triggered or intervention measures to disconnect the driver from the vehicle are taken.