A method for recognizing and quantitatively evaluating take-off and landing effective operation of a copilot of a civil aircraft

Abstract

The application discloses a kind of civil aircraft co-pilot take-off and landing effective operation recognition and quantification evaluation method, its method includes: constructing flight feature database, historical flight sample database and operation quality comprehensive evaluation model, operation quality comprehensive evaluation model is based on flight feature database to obtain historical flight sample database according to the characteristic data of flight sequentially gathered, to all flights using K-Means clustering module obtains K clustering results, to K clustering results is sorted according to the order of co-pilot operation quality from low to high order classification processing;The main pilot operation time length, co-pilot operation time length, main co-pilot collaborative operation time length and main pilot intervention data corresponding to the take-off stage and landing stage of the to-be-evaluated research flight are input into operation quality comprehensive evaluation model, and operation quality comprehensive evaluation model outputs the grade of co-pilot operation quality.The application can provide data support for the determination of effective take-off and landing of civil aviation, and promote the development of pilot training system to refinement and data.

Description

Technical Field

[0001] This invention relates to the field of co-pilot operation assessment, and in particular to a method for identifying and quantifying effective takeoff and landing operations of co-pilots in civil aircraft. Background Technology

[0002] The current civil aviation pilot training and qualification assessment system generally suffers from a prominent problem of "emphasizing flight hours while neglecting control quality." The evaluation of co-pilot capabilities has long relied on instructors' subjective judgment and flight experience statistics, lacking objective, quantitative, and scientific assessment methods. Particularly during takeoff and landing, the co-pilot's actions are deeply coupled with the captain's input, autopilot commands, and external environmental disturbances, making it difficult to accurately identify their true control intentions and scientifically quantify their operational contributions. In mechanically driven aircraft (such as the B737), the left and right control sticks are physically linked, and the co-pilot's and captain's operational signals are superimposed, making direct separation difficult. Existing assessment methods mainly rely on flight instructors' subjective experience or determine operational effectiveness based on simple flight parameter thresholds. These methods have significant limitations: first, in large amounts of flight data, they cannot effectively distinguish whether the pilot is "actively executing precise controls" or "passively following changes in aircraft status"; second, they fail to fully consider the essential characteristics of mechanically driven aircraft control systems, resulting in a simplistic and unsuitable assessment standard; and third, they lack a tiered evaluation mechanism for operational quality, making it difficult to differentiate the performance differences of co-pilots at different levels under various operational scenarios.

[0003] Current technology lacks a comprehensive operational identification and quantitative assessment method that integrates multi-dimensional flight parameters, is interpretable, and can grade operational quality. This results in inaccurate statistics on co-pilot "effective maneuvering" time, unfair assessments of operational contributions, and an inability to achieve refined grading of co-pilot operational capabilities. This not only hinders the implementation of the Civil Aviation Administration's "Guidelines for Effective Takeoff and Landing Monitoring," but also fails to provide reliable quantitative data for co-pilot upgrade training, impeding the transformation of pilot training from "experience-driven" to "data-driven." Therefore, there is an urgent need for a technical assessment method that can automatically and accurately identify effective co-pilot operational behaviors during takeoff and landing from real flight QAR data, and objectively quantify and grade their operational effectiveness. This would improve the scientific rigor of the assessment and the effectiveness of training, enabling precise and personalized training of co-pilot capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying and quantifying the effective operations of a co-pilot during takeoff and landing in civil aircraft. It achieves precise decoupling of the co-pilot's actual maneuvering behavior from highly coupled flight data, accurately distinguishing various behavioral patterns such as independent operations, collaborative operations, and ineffective operations. Combined with cluster analysis, it classifies the quality of co-pilot operations into levels, achieving refined evaluation from two dimensions: operation duration and operation quality. By statistically analyzing operation duration, percentage, and captain intervention frequency, it quantifies the co-pilot's maneuvering contribution and stability, forming objective and traceable evaluation results.

[0005] The objective of this invention is achieved through the following technical solution: A method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot, the method comprising: S1. Obtain QAR time-series data of the research flight, identify the takeoff and landing phases based on the QAR time-series data, and collect the stick force time-series data in segments according to the takeoff and landing phases. The stick force time-series data includes pitch stick force time-series data and roll stick force time-series data. S2. Construct a driver and co-pilot determination rule module based on the combined action of stick forces on both sides and the master control status. The driver and co-pilot determination rule module extracts the stick forces on both sides from the stick force time sequence data frame by frame to determine whether the action of the time frame is the driver's operation, the co-pilot's operation, or the driver and co-pilot's collaborative operation. The driver's operation time, co-pilot's operation time, driver and co-pilot's collaborative operation time, and driver intervention data are counted according to the takeoff and landing phases, respectively. S3. Construct a flight feature database, a historical flight sample database, and a comprehensive operational quality evaluation model. The features of the flight feature database include the proportion of primary pilot operations, primary pilot operations, and collaborative operations during takeoff; the proportion of captain operations, primary pilot operations, and collaborative operations during landing; primary pilot intervention data during takeoff; primary pilot intervention data during landing; and total intervention data for both takeoff and landing phases. Principal component analysis focusing on primary pilot takeoff and landing operations is used to reduce the dimensionality of the feature space of the flight feature database. The comprehensive operational quality evaluation model obtains feature data from the historical flight sample database based on the flight feature database, which is then sequentially aggregated by flight. K-Means clustering is then used to obtain K clustering results for all flights. These K clustering results are then sorted in order of primary pilot operation quality from low to high. S4. Following methods S1 and S2, obtain the pilot's operation time, co-pilot's operation time, pilot-co-pilot collaborative operation time, and pilot intervention data for the takeoff and landing phases of the flight to be evaluated. Input the data into the comprehensive operation quality assessment model. The comprehensive operation quality assessment model outputs the co-pilot's operation quality level.

[0006] To better implement the present invention, in method S2, the judgment rules of the driver and passenger judgment rule module are as follows: if the absolute values ​​of the lever forces on both sides are less than P1, the operation is judged to be invalid. If the absolute value of the lever force on one side is ≥ P1 and the absolute value of the lever force on the other side is < P1, then the side with the master control state of SET in the QAR time series data is identified as the operating side, and the driver attributes corresponding to the operating side are extracted. The driver attributes include the driver and the co-driver. If the absolute values ​​of the lever forces on both sides are ≥ P1, the further judgment method is as follows: if the difference in the absolute values ​​of the lever forces on both sides is ≤ P2, it is determined that the driver and co-driver are cooperating; if the difference in the absolute values ​​of the lever forces on both sides is > P2, the side with the master control status of SET in the QAR time series data is identified as the operating side, and the driver attributes corresponding to the operating side are extracted.

[0007] Preferably, if the driver and passenger determination rule module determines that the time frame is an operation performed by the driver, then that time frame is counted as a driver operation; if the driver and passenger determination rule module determines that the time frame is an operation performed by the passenger, then that time frame is counted as a passenger operation; if the driver and passenger determination rule module determines that the time frame is a collaborative operation between the driver and passenger, then that time frame is counted as a collaborative operation between the driver and passenger.

[0008] Preferably, the driver and co-pilot determination rule module performs frame-by-frame determination of the stick force timing data and counts the driver's operation time, the co-pilot's operation time, and the driver and co-pilot's collaborative operation time; the driver intervention data is the intervention data of the driver intervening in the co-pilot's operation when the pitch stick or / and roll stick are operated, and the intervention data includes the number of interventions.

[0009] Preferably, the principal component analysis is performed on the feature space based on the historical flight sample database and projected onto the principal component space. The principal component space includes the ratio of the proportion of co-pilot operations to the proportion of pilot operations during takeoff and landing, as well as the number or frequency of pilot intervention during takeoff and landing.

[0010] Preferably, the K-Means clustering module performs Z-score normalization on the feature dimensions of flights and uses the elbow rule and contour coefficient to determine the optimal number of clusters K.

[0011] Preferably, the comprehensive operational quality evaluation model further includes a co-pilot scoring module, which obtains the co-pilot's operational participation level according to the following expression. : ,in , These represent the percentage of independent operations by the co-pilot during takeoff and landing, respectively. , These are the weights corresponding to the takeoff and landing phases, respectively. The initial score for the co-pilot is then calculated using the following linear weighted formula. : ,in Weighting of co-pilot's operational involvement. To intervene in the frequency penalty weight, For intervention parameters, For bias terms; Obtain the initial scores of all flights in the historical flight sample database. The scores are standardized from 0 to 100, and then aggregated into a comprehensive score range based on K clustering results. Each clustering result corresponds to a comprehensive score range for the co-pilot's operational quality level.

[0012] Preferably, in method S4, the operational quality comprehensive evaluation model calculates the initial score of the flight under evaluation. The scores are then standardized to obtain a comprehensive score.

[0013] Preferably, in method S1, the takeoff phase identification method is as follows: Takeoff-related data is initially screened from the QAR time-series data; the data frame in which both landing gears simultaneously meet the takeoff state is identified and recorded as the start frame of the takeoff phase; the autopilot is identified and recorded as the end frame of the takeoff phase according to the order of the QAR time-series data; the duration between the start and end frames of the takeoff phase is the takeoff duration. The landing phase identification method is as follows: Landing-related data is initially screened from the QAR time-series data; the data frame in which the autopilot is disconnected is identified and recorded as the start frame of the landing phase; the data frame in which both landing gears simultaneously touch down is identified and recorded as the end frame of the landing phase according to the order of the QAR time-series data; the duration between the start and end frames of the landing phase is the landing duration.

[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) This invention achieves precise decoupling of the co-pilot's actual control behavior from highly coupled flight data, accurately distinguishes various behavior modes such as independent operation, collaborative operation and invalid operation, and classifies the co-pilot's operation quality by combining cluster analysis method, and realizes refined evaluation from two dimensions: operation duration and operation quality; by statistically analyzing operation duration, proportion and captain intervention frequency, the co-pilot's control contribution and stability are quantified, forming an objective and traceable evaluation result; this invention can provide data support for the determination of effective take-off and landing in civil aviation, provide scientific basis for co-pilot upgrade training and capability assessment, and promote the development of pilot training system towards refinement and data-driven direction.

[0015] (2) Based on QAR time series data, this invention obtains lever force time series data. By using the master and co-driver judgment rule module based on the joint master and co-driver control status, the master and co-driver control of the lever force time series data is extracted frame by frame to determine whether the master driver operation, co-driver operation or master and co-driver collaborative operation is performed in that frame. This invention achieves precise decoupling of co-driver operation. This invention can effectively distinguish between co-driver independent operation, collaborative operation and invalid operation, significantly improve the recognition accuracy, and introduce an operation classification mechanism to achieve a refined evaluation leap from whether there is an operation to the quality of the operation.

[0016] (3) This invention not only achieves accurate identification of operational behavior, but also constructs a complete quantitative grading evaluation index system, refines the calculation of key indicators such as the proportion of independent operation by the co-pilot, the proportion of collaborative operation, and the number of times the captain intervenes, and introduces PCA and K-Means clustering algorithms to form an operational quality grading feature vector. The clustering results are sorted and graded according to the co-pilot's operational quality from low to high, realizing the grading evaluation of the co-pilot's operational quality. Combined with the operation proportion map, timeline analysis, and multi-flight trend comparison, it provides an objective and traceable quantitative basis for the co-pilot's operational contribution, making the evaluation results directly comparable and grade distinguishable.

[0017] (4) This invention effectively solves the problems of time-consuming and labor-intensive traditional manual assessment methods and subjective bias, and realizes automated and batch operation behavior analysis and grade assessment. The assessment results are objective and traceable. It can be used for the phased ability evaluation and grade assessment of co-pilots, and can also support the formulation of personalized training programs, promoting the evolution of the pilot training system towards a data-driven, precise and efficient modern model. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for identifying and evaluating the effective takeoff and landing operations of the co-pilot of a civil aircraft, as described in this invention. Figure 2 This is a schematic diagram illustrating the principle of the takeoff phase identification method in the embodiment; Figure 3 This is a schematic diagram illustrating the principle of the landing phase identification method in the embodiment; Figure 4 This is a schematic diagram illustrating the logical determination rules for effective operations, co-pilot operations, and simultaneous operations during the takeoff and landing phases of the B737 aircraft in this embodiment. Figure 5 The example provided is a bar chart illustrating the percentage of operations during the flight's takeoff and landing phases. Figure 6 The example provided is a pie chart showing the percentage of operations during the flight's takeoff and landing phases. Figure 7 This is a clustering hierarchy diagram illustrating the effective operations of the co-pilot during takeoff and landing in this embodiment. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to embodiments: Example like Figure 1 As shown, a method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot is provided, the method comprising: S1. Obtain QAR time-series data of the research flight, identify the takeoff and landing phases based on the QAR time-series data, and collect the stick force time-series data segmented according to the takeoff and landing phases. The stick force time-series data includes pitch stick force time-series data and roll stick force time-series data.

[0020] In some embodiments, the takeoff phase identification method is as follows: Takeoff-related data is initially filtered from the QAR time-series data; the data frame in which both left and right landing gears simultaneously meet the takeoff requirements is recorded as the start frame of the takeoff phase; the autopilot engagement is identified according to the order of the QAR time-series data, and the recorded data frame is recorded as the end frame of the takeoff phase; the duration between the start and end frames of the takeoff phase is the takeoff duration. The specific method is as follows: Figure 2 As shown, QAR time-series data is retrieved point-by-point in chronological order. When FLIGHT_PHASE is found to be TAKE_OFF (takeoff phase), the takeoff phase identification window is entered. Within this window, the left and right landing gear states are sequentially determined. When LGL_L (left landing gear) and LGL_R (right landing gear) simultaneously meet the AIR (airborne / takeoff) condition, this moment is recorded as T1, the start time of the takeoff phase. Subsequently, the autopilot status is retrieved from T1 onwards. When AP (autopilot) changes to ENGAGE (engaged, engaged, connected), this moment is recorded as T2, the end time of the takeoff phase. Finally, T2-T1 is used as the output of the takeoff phase determination duration for this flight segment. The landing phase identification method is as follows: landing-related data is initially filtered from the QAR time-series data. The data frame where the autopilot disconnects is identified as the start frame of the landing phase. The data frame where the left and right landing gears simultaneously touch down is identified as the end frame of the landing phase, following the order of the QAR time-series data. The duration between the start and end frames of the landing phase is the landing duration. The specific methods are as follows: Figure 3As shown, QAR time-series data is retrieved point by point in chronological order. When FLIGHT_PHASE is found to be APPROACH, the landing phase identification window is entered. Within this window, the autopilot status is further determined. When AP is NONE (no, not engaged, disconnected), this moment is recorded as T3, serving as the start time of the landing phase. Subsequently, starting from T3, the landing gear touchdown status is retrieved. When LGL_N (nose landing gear) meets the touchdown status (GROUND, touching down, already on the ground), this moment is recorded as T4, serving as the end time of the landing phase. Finally, T4-T3 is used as the determination duration of the landing phase for this flight segment. This invention proposes a calculable definition of "effective operation" suitable for data analysis: it refers to the continuous and dominant control operation of the aircraft attitude and trajectory independently performed by the co-pilot while acting as the primary pilot, without monitoring the pilot's input of controls or only making minor corrections. This operation should form a clear "control input - aircraft dynamic response" closed loop in terms of timing, thus distinguishing it from unconscious stick position holding or passive following behavior.

[0021] S2. Construct a driver / passenger determination rule module based on the combined force of the two levers and the master control state. This module extracts the force of both levers from the time-series data frame by frame and determines whether the action at that time frame is a driver operation, a passenger operation, or a collaborative driver / passenger operation. In some embodiments, such as... Figure 4 As shown, the determination rules for the driver and passenger seat determination module are as follows: If the absolute values ​​of the stick forces on both sides are less than P1 (P1 is 4 pounds in this example), the operation is deemed invalid. The preferred value for the stick force threshold P1 is 4 pounds, determined based on the control sensitivity description in the B737 Flight Crew Operations Manual and the pilot's actual perception threshold. A 4-pound threshold is higher than the minimum control force that the pilot can clearly perceive (approximately 2-3 pounds). Furthermore, through statistical analysis of historical QAR data from a large number of flights during the stable cruise phase (considered as having no intention of active control), a stick force threshold of 4 pounds ensures effective filtering of environmental disturbances and sensor noise.

[0022] If the absolute value of the lever force on one side is ≥ P1 (P1 is 4 pounds in this example) and the absolute value of the lever force on the other side is < P1, then the side with the master control status of SET in the QAR time series data is identified as the operating side, and the driver attributes corresponding to the operating side are extracted. The driver attributes include the driver and the co-driver. If the absolute values ​​of the lever forces on both sides are ≥ P1 (P1 is 4 lbs in this example), the further determination method is as follows: If the difference in the absolute values ​​of the lever forces on both sides is ≤ P2 (P1 is 1.5 lbs in this example), then it is determined that the driver and co-pilot are cooperating. The preferred value of the absolute value difference of the lever forces in this invention is 1.5 lbs, which is derived from the force difference analysis of the measured data of the typical "cooperative lever" scenario of the B737 model, reflecting the physical boundary for determining the control dominance in the mechanical linkage system. If the difference in the absolute values ​​of the lever forces on both sides is > P2, then the side with the master control status of SET in the QAR time series data is identified as the operating side, and the driver attributes corresponding to the operating side are extracted. This invention simultaneously extracts the pitch lever force (CTL_CL_FC_A_L active, CTL_CL_FC_B_L auxiliary) and roll lever force of the left and right control sticks, as well as the status of the key parameter LOCAL LIMITED MASTER_L / R, to indicate which control stick currently has master control over the input of the other side. This invention is the first to combine lever force difference analysis with master control status, enabling precise differentiation of independent, collaborative, and invalid operations by the co-pilot under physical linkage conditions, overcoming the limitation of traditional methods in decoupling mixed signals. The system's thresholds P1 and P2 are configurable parameters, allowing users to fine-tune the authorization within the management interface according to airline standard operating procedures, ensuring the method's adaptability.

[0023] The duration of the primary pilot's operation, the co-pilot's operation, the duration of the primary and co-pilot's coordinated operation, and the primary pilot's intervention data are recorded separately for the takeoff and landing phases. Preferably, if the primary pilot / co-pilot determination rule module determines that a time frame represents a primary pilot operation, then that time frame is counted as a primary pilot operation; if the primary pilot / co-pilot determination rule module determines that a time frame represents a co-pilot operation, then that time frame is counted as a co-pilot operation; if the primary pilot / co-pilot determination rule module determines that a time frame represents a primary and co-pilot coordinated operation, then that time frame is counted as a primary and co-pilot coordinated operation. The primary pilot / co-pilot determination rule module performs frame-by-frame determination of the stick force timing data and records the duration of the primary pilot's operation, the co-pilot's operation, and the duration of the primary and co-pilot's coordinated operation. The primary pilot's intervention data refers to the intervention data of the primary pilot intervening in the co-pilot's operation during pitch stick and / or roll stick operations, including the number of interventions or frequency. Figure 5 , Figure 6 As shown, this invention statistically studies the pilot's operation time, co-pilot's operation time, pilot-co-pilot collaborative operation time, and pilot intervention data for flights during the takeoff and landing phases, respectively. It can generate structured reports, including timelines for takeoff and landing phases, operation time and percentage graphs for each role, preliminary judgment suggestions for effective takeoff and landing based on preset standards, and trend analysis and comparison of multiple flights.

[0024] S3. Construct a flight feature database, a historical flight sample database, and a comprehensive operational quality evaluation model. The flight feature database includes features such as the proportion of primary pilot operations during takeoff, the proportion of first officer operations during takeoff, the proportion of coordinated operations during takeoff, the proportion of captain operations during landing, the proportion of first officer operations during landing, the proportion of coordinated operations during landing, primary pilot intervention data during takeoff, primary pilot intervention data during landing, and total intervention data for both takeoff and landing phases. Principal Component Analysis (PCA) is used to reduce the dimensionality of the feature space based on the features of the flight feature database, focusing on first officer takeoff and landing operations. PCA performs dimensionality reduction within the feature space based on the historical flight sample database and projects it onto the principal component space. The principal component space includes the ratio of first officer operations to primary pilot operations during takeoff and landing (a higher value indicates a higher proportion of first officer operations), and the number or frequency of primary pilot interventions during takeoff and landing. The primary pilot intervention frequency is the ratio of primary pilot interventions to pitch stick operations (a higher value indicates more frequent captain interventions). Principal component analysis was used to capture key feature information, providing a clear and condensed feature foundation for subsequent K-Means clustering.

[0025] The comprehensive operational quality assessment model is based on a flight feature database, which obtains feature data from a historical flight sample database, sequentially grouped by flight. Then, a K-Means clustering module is used to obtain K clustering results for all flights. These K clustering results are then sorted in ascending order of co-pilot operational quality. The K-Means clustering module performs Z-score normalization on the flight feature dimensions and uses the elbow rule and silhouette coefficient to determine the optimal number of clusters K. Figure 7 For example, the optimal number of clusters K after clustering by the K-Means clustering module is three clustering results, and the three clustering results correspond to the level of co-pilot operation quality assigned (such as level A, level B, and level C). Figure 7 In the case study, the correspondence between the three clustering results and the co-pilot's operational quality level is shown in the table below:

[0026] As shown in Table 1, cluster 2 corresponds to the highest operational quality, grade A, characterized by a high proportion of independent operation by the co-pilot and moderate captain intervention, demonstrating independent control capabilities; cluster 0 corresponds to the medium operational quality, grade B, characterized by active co-pilot participation but unstable control, requiring frequent captain intervention for correction; and cluster 1 corresponds to the operational quality needing improvement, grade C, characterized by extremely low co-pilot participation and severely insufficient independent control capabilities. This embodiment maps the clustering results to the two-dimensional space after PCA dimensionality reduction, generating a cluster distribution map (e.g., ...). Figure 7As shown in the figure, different colors represent different operational levels, intuitively revealing the group differences and distribution patterns of co-pilot operational behavior. Simultaneously, an operational level radar chart is generated for each operational level, displaying its mean performance across various assessment characteristics, forming a clear operational profile for easy comparison and analysis of operational characteristics at different levels. Furthermore, the system can build a personal profile for each co-pilot, statistically analyzing their operational level distribution across multiple flights, identifying individual operational stability and skill evolution trends, and supporting dynamic capability tracking and growth path analysis. At the application level, this invention introduces operational level as a weighted indicator into the effective takeoff and landing determination system. For example, setting "operation level A with an independent operation rate greater than 50%" is considered a high-quality effective takeoff and landing, significantly improving the scientific rigor and discriminative power of the determination. The system automatically outputs the cluster label and corresponding operational level for each flight, serving as a quantitative basis for subsequent co-pilot operational quality assessment, effective takeoff and landing monitoring, and the development of personalized training programs. By combining PCA dimensionality reduction and cluster analysis, this invention achieves an objective classification of co-pilot operational behavior during takeoff and landing based on actual flight data. This solves the problems of strong subjectivity and inability to quantify in traditional assessments, and provides scientific support for pilot capability profiling and refined training.

[0027] This embodiment, based on QAR data from 2000 flights, uses PCA dimensionality reduction and K-Means clustering analysis to classify the first officer's operational behavior during takeoff and landing into three levels: Cluster 2 (A-level, excellent, 22.4%) is characterized by a high proportion of independent operations by the first officer (62.97% during takeoff and 53.79% during landing), with moderate captain (i.e., primary pilot) intervention (3.47 interventions in total), and stable and mature handling; Cluster 0 (B-level, good, 45.3%) is characterized by active participation by the first officer (47.21% during takeoff and 40.99% during landing), but the handling is still unstable, requiring frequent corrections from the captain (13.44 interventions in total), indicating a stage of practical training; Cluster 1 (C-level, needs improvement, 32.3%) is characterized by minimal first officer participation (both percentages below 6%), with the captain taking the lead throughout (1.74 interventions in total), indicating a severe lack of independent handling ability. These classification results highly align with the experience-based judgments of flight instructors, validating the model's effectiveness.

[0028] S4. Following methods S1 and S2, obtain the pilot's operation time, co-pilot's operation time, pilot-co-pilot collaborative operation time, and pilot intervention data for the takeoff and landing phases of the flight to be evaluated. Input the data into the comprehensive operation quality assessment model. The comprehensive operation quality assessment model outputs the co-pilot's operation quality level.

[0029] In some embodiments, the operational quality comprehensive evaluation model further includes a co-pilot scoring module, which obtains the co-pilot's operational participation level according to the following expression. : ,in , These represent the percentage of independent operations by the co-pilot during takeoff and landing, respectively. , These are the weights corresponding to the takeoff and landing phases, respectively.

[0030] The initial score for the co-pilot is then calculated using the following linear weighted formula. : ,in Weighting of co-pilot's operational involvement. To intervene in the frequency penalty weight, For intervention parameters, This is a bias term.

[0031] Obtain the initial scores of all flights in the historical flight sample database. The scores are standardized from 0 to 100, and the initial scores are... The expression mapping to the standardized interval of 0-100 is as follows: ,in The standardized composite score (after normalization, the closer the score is to 100, the higher the quality of the operation). , These are the minimum and maximum initial scores from the historical flight sample database, respectively, representing the initial score before standardization. .

[0032] The comprehensive score range is aggregated based on K clustering results, with each clustering result corresponding to a comprehensive score range for the co-pilot's operational quality level. This embodiment outputs an independent comprehensive score for each flight sample, with the score result and cluster label output together, thus achieving a dual assessment of the co-pilot's operational ability: including both discrete level classification and continuous score quantification. Based on the score distribution of each flight obtained from cluster analysis, scoring thresholds for operational levels are further set: Level A (Excellent) corresponds to a comprehensive score ≥ 75 points, i.e., the comprehensive score range for Level A (Excellent) is: comprehensive score ≥ 75 points and ≤ 100 points; Level B (Good) has a comprehensive score range: comprehensive score ≥ 45 points and < 75 points; Level C (Needs Improvement) has a comprehensive score range: comprehensive score < 45 points. These thresholds can be dynamically adjusted based on actual fleet data to adapt to different training stages and aircraft characteristics. The operational quality comprehensive assessment model calculates the initial score for the flight under evaluation. The system standardizes and processes the data to obtain a comprehensive score, then determines the score range to which the co-pilot belongs and the co-pilot's operational quality level. This invention can define a high-quality effective takeoff and landing as "operation level A with co-pilot independent operation accounting for more than 50%". In this embodiment, the percentage of flights meeting the high-quality effective takeoff standard is 18.7%, and the percentage meeting the high-quality effective landing standard is 15.2%. This determination method not only considers "whether operation occurred" but also integrates "how well the operation was performed," improving the scientific rigor and discriminative power of effective takeoff and landing determination. This invention achieves fully automated processing, reducing the analysis time for 2000 flights from approximately 200 hours of manual evaluation to within 2 hours, increasing processing efficiency by nearly 100 times. Simultaneously, the evaluation results are objective and consistent, eliminating subjective bias. A random sample of 50 flights was manually reviewed by three senior flight instructors. The accuracy rate of operator identification reached 92.7%, and the consistency between the operation classification results and the instructors' experience judgment reached 89.3%, fully verifying the reliability and practicality of the technical solution.

[0033] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot, characterized in that: The methods include: S1. Obtain QAR time-series data of the research flight, identify the takeoff and landing phases based on the QAR time-series data, and collect the stick force time-series data in segments according to the takeoff and landing phases. The stick force time-series data includes pitch stick force time-series data and roll stick force time-series data. S2. Construct a driver / co-pilot determination rule module based on the combined action of both control stick forces and the master control status. This module extracts the action of both control stick forces from the time-series data frame by frame to determine whether the action at that time frame is a driver operation, a co-pilot operation, or a coordinated driver / co-pilot operation. It also statistically analyzes the duration of driver operation, co-pilot operation, coordinated driver / co-pilot operation, and driver intervention data for both takeoff and landing phases. The determination rules of the driver / co-pilot determination rule module are as follows: If the absolute values ​​of the forces on both sides are less than P1, the operation is deemed invalid. If the absolute value of the lever force on one side is ≥ P1 and the absolute value of the lever force on the other side is < P1, then the side with the master control state of SET in the QAR time series data is identified as the operating side, and the driver attributes corresponding to the operating side are extracted. The driver attributes include the driver and the co-driver. If the absolute values ​​of the lever forces on both sides are ≥ P1, the further judgment method is as follows: if the difference in the absolute values ​​of the lever forces on both sides is ≤ P2, it is determined that the driver and co-driver are cooperating; if the difference in the absolute values ​​of the lever forces on both sides is > P2, the side with the master control status of SET in the QAR time series data is identified as the operating side, and the driver attributes corresponding to the operating side are extracted. S3. Construct a flight feature database, a historical flight sample database, and a comprehensive operational quality evaluation model. The features of the flight feature database include the proportion of primary pilot operations, primary pilot operations, and collaborative operations during takeoff; the proportion of captain operations, primary pilot operations, and collaborative operations during landing; primary pilot intervention data during takeoff; primary pilot intervention data during landing; and total intervention data for both takeoff and landing phases. Principal component analysis focusing on primary pilot takeoff and landing operations is used to reduce the dimensionality of the feature space of the flight feature database. The comprehensive operational quality evaluation model obtains feature data from the historical flight sample database based on the flight feature database, which is then sequentially aggregated by flight. K-Means clustering is then used to obtain K clustering results for all flights. These K clustering results are then sorted in order of primary pilot operation quality from low to high. S4. Following methods S1 and S2, obtain the pilot's operation time, co-pilot's operation time, pilot-co-pilot collaborative operation time, and pilot intervention data for the takeoff and landing phases of the flight to be evaluated. Input the data into the comprehensive operation quality assessment model. The comprehensive operation quality assessment model outputs the co-pilot's operation quality level.

2. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 1, characterized in that: If the driver and passenger seat determination rule module determines that the time frame is an operation performed by the driver, then that time frame is counted as an operation performed by the driver; if the driver and passenger seat determination rule module determines that the time frame is an operation performed by the passenger, then that time frame is counted as an operation performed by the passenger. If the driver and co-driver determination rule module determines that the time frame is a collaborative operation between the driver and co-driver, then that time frame is counted as a collaborative operation between the driver and co-driver.

3. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 2, characterized in that: The driver and co-pilot determination rule module performs frame-by-frame determination of the lever force timing data and counts the driver's operation time, co-pilot's operation time, and driver and co-pilot collaborative operation time. The driver intervention data refers to the intervention data of the driver intervening in the passenger's operation when the pitch stick and / or roll stick are operated, and the intervention data includes the number of interventions.

4. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 1, characterized in that: The principal component analysis is based on the historical flight sample database. The feature space is reduced and then projected to the principal component space. The principal component space includes the ratio of the proportion of co-pilot operations to the proportion of pilot operations during takeoff and landing, as well as the number or frequency of pilot intervention during takeoff and landing.

5. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 1, characterized in that: The K-Means clustering module performs Z-score normalization on the feature dimensions of flights and uses the elbow rule and contour coefficient to determine the optimal number of clusters K.

6. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 1, characterized in that: The comprehensive operational quality evaluation model also includes a co-pilot scoring module, which calculates the co-pilot's operational participation level according to the following expression. : ,in , These represent the percentage of independent operations by the co-pilot during takeoff and landing, respectively. , These are the weights corresponding to the takeoff and landing phases, respectively. The initial score for the co-pilot is then calculated using the following linear weighted formula. : ,in Weighting of co-pilot's operational involvement. To intervene in the frequency penalty weight, For intervention parameters, For bias terms; Obtain the initial scores of all flights in the historical flight sample database. The scores are standardized from 0 to 100, and then aggregated into a comprehensive score range based on K clustering results. Each clustering result corresponds to a comprehensive score range for the co-pilot's operational quality level.

7. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 6, characterized in that: In method S4, the operational quality comprehensive assessment model calculates the initial score of the flight under study to be evaluated. The scores are then standardized to obtain a comprehensive score.

8. The method for identifying and quantifying effective takeoff and landing operations of a civil aircraft co-pilot according to claim 1, characterized in that: In method S1, the takeoff phase identification method is as follows: Takeoff-related data is initially filtered from the QAR time-series data; the data frame in which both landing gears simultaneously meet the takeoff state is identified and recorded as the start frame of the takeoff phase; the autopilot is identified and recorded as the end frame of the takeoff phase according to the order of the QAR time-series data; the duration between the start and end frames of the takeoff phase is the takeoff duration. The landing phase identification method is as follows: Landing-related data is initially filtered from the QAR time-series data; the data frame in which the autopilot is disconnected is identified and recorded as the start frame of the landing phase; the data frame in which both landing gears simultaneously touch down is identified and recorded as the end frame of the landing phase according to the order of the QAR time-series data; the duration between the start and end frames of the landing phase is the landing duration.

Images