CN111047916B - A heavy landing risk identification method based on the area characteristic of QAR curve - Google Patents

Abstract

The invention discloses a hard landing risk identification method based on a QAR curve area feature. The identification method is specifically: S1: extracting QAR parameters required for risk identification of all flight segments of a single aircraft; S2: performing data analysis on the extracted QAR parameters Cleaning; S3: Based on S2, extract the QAR data of the landing stage of all flight segments; S4: According to the vertical velocity average value of the fixed descent distance of all flight segments during the landing stage, construct the mean curve, and then extract the curve area feature of each flight segment; S5: Construct a loss function, and combine the curve area features of each flight segment to determine the curve area feature threshold; S6: Compare the curve area feature of each flight segment with the curve area feature threshold, if it is greater than that, it is considered that there is a risk of heavy landing , otherwise considered safe. The invention defines the calculation method of the loss value by establishing the vertical velocity mean value curve and the loss function, thereby obtaining the heavy landing risk identification threshold.

Description

A heavy landing risk identification method based on the area characteristic of QAR curve

technical field

The invention relates to the field of heavy-landing risk research of civil passenger aircraft, in particular to a heavy-landing risk identification method based on the area characteristic of a QAR curve.

Background technique

According to Boeing's major flight safety accident data from 1959 to 2016, the approach and landing stages are the flight stages most prone to major safety accidents, and the incidence of accidents and unsafe events is significantly higher than other flight stages. The landing phase averages only 1% of flight time, but its accident rate is as high as 24%.

Among the safety incidents in the landing stage, heavy landing is one of the frequent unsafe incidents. From 2006 to 2011, there were 125 unsafe heavy landing incidents in civil aviation in my country, accounting for about 20% of the total unsafe incidents in the landing stage. As a type of risk event, a heavy landing will not only bring a bad flight experience to passengers, but also damage the airline's image. Frequent heavy landings will accelerate the fatigue damage or even breakage of the wing, landing gear, and engine structure, and increase landing safety accidents. The probability of occurrence of the accident will bring huge economic losses to airlines, and in severe cases, it will lead to catastrophic accident consequences, threatening the life safety of passengers.

SUMMARY OF THE INVENTION

In view of this, the purpose of the present invention is to provide a method for identifying the risk of heavy landing based on the area characteristic of the QAR curve. The method has high stability and is suitable for promotion in this field.

The purpose of this invention is to realize through the following technical solutions:

A heavy landing risk identification method based on QAR curve area characteristics, the identification method is specifically: S1: extracting QAR parameters required for risk identification of all flight segments of a single aircraft;

S2: perform data cleaning on the extracted QAR parameters;

S3: Based on S2, extract the QAR data of the landing phase of all flight segments;

S4: According to the vertical velocity mean value of the fixed descent distance during the landing phase of all segments, construct the mean value curve, and then extract the curve area feature of each segment;

S5: Construct a loss function, and combine the curve area characteristics of each flight segment to determine the curve area characteristic threshold;

S6: Compare the curve area feature of each flight segment with the curve area feature threshold. If it is greater than the curve area feature threshold, it is considered that there is a risk of heavy landing, otherwise, it is considered safe.

Further, the S1 is specifically:

S11: Decode and parse the QAR parameters of all flight segments in the civil aircraft to obtain CSV files of all flight segments;

S12: Extract the parameter data required for risk identification of the CSV file of all flight segments, the parameter data includes the aircraft's longitudinal load, radio altitude, engine speed, longitudinal acceleration, airspeed, ground speed, vertical speed, flap state, seam Wing status, landing gear status, spoiler status, true altitude, pitch angle.

Further, the S3 is specifically:

S31: Divide the flight stage according to the value of the parameter data, and extract the landing stage data;

S32: In the landing stage data, identify the landing time point of the aircraft through the landing gear state parameters;

S33: In the landing phase data of all flight segments, extract the time point t _start when the radio altitude is 50ft, and determine the touchdown time point through the parameters of the landing gear state, longitudinal acceleration, and spoiler state, and record it as t _end ;

S34: Extract the vertical velocity at all time points in the [t _start , t _end ] time period in all the CSV files.

Further, the S4 is specifically:

S41: According to the S34, use the B-spline interpolation method to expand the vertical velocity points in each [t _start , t _end ] time period of all flight segments to at least 50, so that the vertical velocity points correspond to the flight height, All flight segments generate their own single-segment vertical speed curve;

S42: Calculate the average value of the vertical speed per 1ft in each [t _start , t _end ] time period of all flight segments to form a vertical speed average curve, where the ordinate represents the flight altitude, and the abscissa represents the flight height mean vertical velocity;

S43: Compare the vertical speed curves and the average vertical speed curves of all single flight segments. The area above the average vertical speed curve is a positive area, and the area below the average vertical speed curve is a negative area. Calculate the area of each flight. said curve area characteristic of the segment;

Wherein the curve area feature = positive value area - negative value area.

Further, the S5 is specifically:

S51: Construct a loss function, wherein the loss function is defined as:

Loss value = a × false negative rate + b × false positive rate;

Among them: the rate of false negative examples = the number of positive examples that are judged to be negative examples/the total number of positive examples that are judged to be negative examples, the rate of false positive examples = the number of positive examples that are judged to be actually negative examples/the total number of positive examples that are judged to be positive;

a and b are the coefficients determined according to the severity of the corresponding misjudgment type, respectively;

S52: Set different thresholds, and judge the positive and negative examples according to whether the curve area feature exceeds the threshold. If the threshold is exceeded, it is considered as a positive example, and if it does not exceed the threshold, it is considered as a negative example. This step is a method for actually judging positive and negative benefits.

S53: Calculate loss values under different thresholds according to the loss function, and take the threshold with the smallest loss value as the threshold for re-landing risk identification of this feature.

The beneficial effects of the present invention are:

Based on the QAR data of the aircraft, the present invention establishes a vertical speed mean curve, so that the vertical speed curve of a single flight segment is compared with it and obtains the curve area characteristics, and then by constructing a loss function, the calculation method of the loss value is defined, and the weight value is obtained. The landing risk identification threshold, the method has high stability and is suitable for promotion in this field.

Other advantages, objects, and features of the present invention will be set forth in the description that follows, and will be apparent to those skilled in the art based on a study of the following, to the extent that is taught in the practice of the present invention. The objectives and other advantages of the present invention may be realized and attained by the following description.

Description of drawings

In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with the accompanying drawings, wherein:

Accompanying drawing 1 is the flow chart of the present invention;

Accompanying drawing 2 is the sample curve area graph.

Detailed ways

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the protection scope of the present invention.

Example 1

This embodiment proposes a hard landing risk identification method based on the area feature of the QAR curve. The identification method is specifically as follows:

S1: Extract the QAR parameters required for risk identification of all flight segments;

S11: Decode and parse the QAR parameters of all flight segments to obtain CSV files of all flight segments;

S12: Extract the parameter data required for risk identification of the CSV file of all flight segments, the parameter data includes the aircraft's longitudinal load, radio altitude, engine speed, longitudinal acceleration, airspeed, ground speed, vertical speed, flap status, slat status , Landing gear status, spoiler status, true altitude, pitch angle.

S2: perform data cleaning on the extracted QAR parameters;

Due to factors such as decoding misalignment or acquisition error, the original QAR data may have obvious abnormalities such as misalignment of some data fields or missing information. Combined with all the parameter data of the aircraft state for a period of time near the time point where the abnormal data is located, the abnormal data is identified, deleted, and inferred and completed.

Abnormal data identification range: The CSV file is incomplete, and there is no whole process from take-off to landing; the CSV file is the flight training data with the same departure and destination; the parameters of the CSV file output by decoding are misplaced, that is, in the parameter 1 column A row of , displays the data of parameter 2; the parameter value exceeds the theoretical value range; the parameter value has an illogical jump, etc.

Delete operation: For the abnormal situation of the CSV file format mentioned above, it is regarded as invalid data and discarded; for the data of the correct format of the CSV file itself, only the parameter values are occasionally abnormal, only the abnormal data in the CSV file is deleted. Completion is then inferred in combination with other parameters.

The method of inference and completion: For continuous numerical parameters such as speed, latitude and longitude, and altitude, the average value of the front and rear is generally taken; for discrete state parameters such as flap status and slat status, the front value or the rear value is generally used for filling.

S3: As shown in Figure 1, based on S2, extract the QAR data of the landing phase of all flight segments;

S31: Divide the flight stages of all flight segments according to the value of the parameter data, and extract the landing stage data;

S32: In the landing stage data, identify the landing time point of the aircraft through the landing gear state parameters, that is, only use the parameter data of those lines corresponding to the landing stage in the CSV file;

S33: In the landing phase data of each flight segment, extract the time point t _start when the radio altitude is 50ft, and judge the touchdown time point of the corresponding flight segment through the parameters of the landing gear state, longitudinal acceleration, and spoiler state, which is recorded as t _end ;

The method of determining the time point of the aircraft touchdown is as follows:

Extract the highest frequency data of five types of parameters including radio altitude, landing gear air-ground switch status, spoiler position, longitudinal acceleration and radio altitude of the landing stage data respectively;

The data of the five types of parameters except the highest frequency data are processed to the same frequency as the corresponding highest frequency data. The frequency of these data ranges from once per second to eight times per second, so it is necessary to increase the frequency of low-frequency data. To be consistent with the highest frequency data, the accuracy of the grounding time point is guaranteed to be higher.

Different upscaling processing methods are used for different data, such as: the state of the landing gear air-ground switch is filled with the previous value; the position of the spoiler is linearly interpolated (the average value of front and rear); the longitudinal acceleration is calculated by using the vertical speed to calculate the proportion of each frame of data, and then Proportional allocation; radio heights are calculated using vertical velocity combined with quadratic spline interpolation.

Based on the decision-making conditions, determine the grounding time of the aircraft. Specifically:

After the start of the landing phase, find the first time point t _start with the radio height less than 3, as the time starting point for the loop judgment; traverse each time point from t _start until it encounters any one of the decision conditions. , mark the point as the ground point t _TD and output. Among them, the decision-making conditions include the first condition, the second condition and the third condition, and satisfying any one of the conditions is the decision-making condition, wherein the first condition is: the spoiler position at any time point traversed from t _start to the back is relatively The position of the spoiler at the last time point has a change greater than the sudden change value I, and the sudden change value I is 4-6; the second condition is: the longitudinal acceleration of any time point traversed from t _start onward is higher than the longitudinal acceleration of the previous time point. The acceleration is larger than the sudden change value II, and the sudden change value II is 0.025-0.035; the third condition is: the state transition of the landing gear air-ground switch occurs at any time point traversed from t _start to the back.

S34: Extract the vertical speed at all time points within [t _start , t _end ] in the landing time period corresponding to each flight segment.

S4: According to the vertical velocity mean value of the fixed descent distance during the landing phase of all segments, construct the mean value curve, and then extract the curve area feature of each segment;

S41: According to the S34, use the B-spline interpolation method to expand the vertical speed points in the [t _start , t _end ] time period of each flight segment to 50, so that the vertical speed points correspond to the flight altitude, and all flight Each segment generates its own single-segment vertical speed curve;

The B-spline interpolation method can generate a smooth function relationship curve according to the known vertical speed and the corresponding time point, and through the function relationship curve, the vertical speed points in the time period can be expanded to 50, so that the vertical speed Corresponding to the flight altitude of the aircraft,

S42: Calculate the average value of the vertical speed per 1ft in each [t _start , t _end ] time period of all flight segments, and form a vertical speed average curve. The vertical speed average curve is that the ordinate represents the flight altitude, and the abscissa represents the vertical speed mean;

S43: Compare the vertical speed curves and the average vertical speed curves of all single flight segments. The area above the average vertical speed curve is a positive area, and the area below the average vertical speed curve is a negative area. Calculate the area of each flight. Curve area feature of the segment;

Wherein the curve area feature = positive value area - negative value area.

S5: Construct a loss function, and combine the curve area characteristics of each flight segment to determine the curve area characteristic threshold.

S51: Construct a loss function, where the loss function is defined as:

Loss value = a × false negative rate + b × false positive rate;

Among them: the rate of false negative examples = the number of positive examples that are initially judged to be negative examples/the total number of positive examples that are judged to be negative examples, and the rate of false positive examples = the number of positive examples that are initially judged to be negative examples/the total number of positive examples that are judged to be positive examples;

The methods for judging positive and negative examples are as follows:

Extract landing loads from parametric data, where:

If the flight segment with the landing load > 1.5, it is considered that there is a landing risk, which is a positive example, otherwise it is a negative example.

a and b are the coefficients determined according to the severity of the corresponding misjudgment type, respectively;

S52: Set different thresholds, and judge the positive and negative examples according to whether the curve area feature exceeds the threshold value, and if it exceeds the threshold value, it is considered as a positive example, and if it does not exceed the threshold, it is considered as a negative example;

S53: Calculate loss values under different thresholds according to the loss function, and take the threshold with the smallest loss value as the threshold for re-landing risk identification of this feature.

This model can then be applied to other flight segment data (non-sample data) as a feature to identify the risk of a hard landing. The specific operation is to calculate the characteristic value of the curve area of the new sample, and compare its relationship with the threshold value. If it is greater than the threshold value, it is considered that there is a risk of heavy landing, otherwise it is considered safe.

Example 2

In the actual application process of this model, more than 20,000 pieces of data are used for fitting. Taking one of the sample data as an example, the vertical speed data of the landing stage are extracted as shown in Table 1:

Table 1 Sample data

According to the method for judging the grounding time point described in Embodiment 1, t _start is specifically 21.75 seconds, t _end is 30.5 seconds, and the gray color is a time period consisting of the nearest two integer time points including this time point.

B-spline interpolation is performed on the data in the interval from t _start to t _end , and the vertical velocity data is obtained as a feature vector. The data are as follows:

[-651.33,-654.81,-656.86,-657.13,-655.71,-652.71,-648.23,-642.37,-635.25,-626.86,-617.15,-606.07,-593.54,-579.52,-564.0,-547.37 530.22,-513.16,-496.77,-481.67,-468.2,-455.46,-442.05,-426.59,-407.71,-384.0,-354.75,-321.89,-287.99,-255.64,-227.41,-5,-205.9,- -186.3, -184.98, -186.81, -189.92, -192.45, -193.15, -192.06, -189.38, -185.29, -179.99, -173.67, -166.88, -160.69, -156.22, -154.56, -156.83 ]

Repeat the above operations, after calculating all the sample feature vectors, calculate the mean vector, and the data of the sample and mean are shown in Table 2:

Table 2 Sample and mean data

SAMPLE MEAN 1 -651.325 -745.227 2 -654.812 -744.545 3 -656.864 -744.042 4 -657.134 -743.565 5 -655.713 -743.229 6 -652.708 -743.225 7 -648.226 -742.053 8 -642.373 -740.15 9 -635.249 -737.825 10 -626.862 -735.542 11 -617.153 -733.083 12 -606.066 -730.075 13 -593.541 -727.254 14 -579.522 -725.014 15 -563.999 -723.034 16 -547.367 -721.284 17 -530.222 -719.316 18 -513.159 -716.87 19 -496.775 -714.035 20 -481.666 -709.201 twenty one -468.202 -699.505 twenty two -455.459 -679.604 twenty three -442.051 -642.977 twenty four -426.595 -584.347 25 -407.706 -505.942 26 -384 -417.118 27 -354.754 -332.954 28 -321.889 -265.253 29 -287.988 -219.815 30 -255.635 -195.643 31 -227.412 -153.816 32 -205.901 -120.611 33 -192.651 -127.079 34 -186.301 -78.3901 35 -184.977 -53.0421 36 -186.808 -40.3578 37 -189.92 -41.8736 38 -192.445 -44.5733 39 -193.147 -43.1673 40 -192.061 -43.3547 41 -189.379 -42.1593 42 -185.292 -41.3004 43 -179.989 -37.6459 44 -173.669 -34.7692 45 -166.881 -31.4725 46 -160.694 -27.5653 47 -156.216 -23.0006 48 -154.558 -17.6923 49 -156.829 -11.5501 50 -164.087 -4.32967

Establish the curve graph as shown in Figure 2, carry out the area integration of SAMPLE-MEAN, and calculate the characteristic value of the curve area of this sample to be 853.2365075. Repeat the above operation to calculate the curve area eigenvalues of all samples. Select a certain threshold, if the area characteristic of the sample curve is greater than the threshold, the sample is judged as a positive example, otherwise it is judged as a negative example, judge all samples according to this method, compare the actual number of positive and negative examples of the sample, and calculate the false negative example rate and false positive example rate, and calculate the loss value corresponding to this threshold based on this, where loss value=a×false negative rate+b×false positive rate. In this example, take a=0.75-1.25, b=1.75-2.25, change the threshold and repeat this operation to calculate the loss values corresponding to multiple thresholds. The threshold corresponding to the minimum loss value is taken as the characteristic threshold for identifying the risk of heavy landing. In this example, the threshold corresponding to the minimum loss value is 10287.72. The corresponding loss values for the 5 thresholds are as follows:

table 3

The characteristic value of the area of the above sample curve is 853.2365075, which is much smaller than the threshold value, and the landing load is 1.148, which is much smaller than the 1.5 risk value of heavy landing.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be Modifications or equivalent replacements, without departing from the spirit and scope of the technical solution, should all be included in the scope of the claims of the present invention.

Claims (2)

1. A heavy landing risk identification method based on QAR curve area characteristics is characterized in that: the identification method specifically comprises the following steps: s1: extracting QAR parameters required by risk identification of all flight segments of a single airplane;

s2: carrying out data cleaning on the extracted QAR parameters;

s3: based on S2, extracting QAR data of landing stages of all the legs;

s4: constructing a mean value curve according to the vertical speed mean values of the fixed descending distances of all flight sections in the landing stage, and then extracting the curve area characteristics of each flight section;

s5: constructing a loss function, and determining a curve area characteristic threshold value by combining the curve area characteristic of each flight segment;

s6: comparing the curve area characteristic of each flight segment with a curve area characteristic threshold, if the curve area characteristic is larger than the curve area characteristic threshold, determining that the risk of heavy landing exists, otherwise, determining that the safety exists;

the S3 specifically includes:

s31: dividing flight phases according to the values of the parameter data, and extracting landing phase data;

s32: identifying the landing time point of the airplane through the landing gear state parameters in the landing stage data;

s33: in landing phase data of all legs, a time point t at which the radio altitude is 50ft is extracted_startJudging the grounding time point by the landing gear state, the longitudinal acceleration and the spoiler state parameter, and recording the grounding time point as t_end；

S34: extracting [ t ] from all CSV files_start，t_end]Vertical velocities of all time points within a time period;

the S4 specifically includes:

s41: according to the S34, each [ t ] of all the legs is interpolated by B splines_start，t_end]The number of the vertical speed points in the time period is expanded to at least 50, so that the vertical speed points correspond to the flying height, and all the flight sections generate respective single flight section vertical speed curves;

s42: calculate each of all legs [ t ]_start，t_end]Forming a vertical speed mean value curve by the mean value of every 1ft of vertical speed in a time period, wherein the vertical speed mean value curve is a vertical coordinate representing the flight height, and the horizontal coordinate representing the vertical speed mean value;

s43: comparing all the single-flight-section vertical speed curves with the vertical speed mean value curve, enabling the area of the area above the vertical speed mean value curve to be a positive-value area, enabling the area of the area below the vertical speed mean value curve to be a negative-value area, and calculating the curve area characteristic of each flight section;

wherein the curve area characteristic is positive value area-negative value area;

the S5 specifically includes:

s51: constructing a loss function, wherein the loss function is defined as:

the loss value is a multiplied by the false negative rate + b multiplied by the false positive rate;

wherein: the false positive rate is the number of the positive examples/the total number of the positive examples;

a and b are coefficients determined according to the severity of the corresponding misjudgment type respectively;

s52: setting different thresholds, judging positive and negative examples according to whether the curve area characteristics exceed the thresholds, and considering the positive examples and not considering the negative examples if the curve area characteristics exceed the thresholds;

s53: and calculating loss values under different thresholds according to the loss function, and taking the threshold with the minimum loss value as the threshold for identifying the characteristic re-landing risk.

2. The QAR curve area feature-based heavy landing risk identification method according to claim 1, wherein: the S1 specifically includes:

s11: decoding and analyzing QAR parameters of all flight segments in the civil aircraft to obtain CSV files of all flight segments;

s12: extracting parameter data required by risk identification of CSV files of all the flight segments, wherein the parameter data comprise longitudinal load, radio altitude, engine rotating speed, longitudinal acceleration, airspeed, ground speed, vertical speed, flap state, slat state, undercarriage state, spoiler state, true altitude and pitch angle of the airplane.

Images