Key flight parameter screening method based on flight parameter-load correlation

Abstract

The present application belongs to the field of helicopter structure fatigue design, and relates to a key flight parameter screening method based on flight parameter-load correlation, which is realized based on Pearson correlation coefficient. Through helicopter scientific research load flight test, a large amount of flight data can be obtained, which contains numerous flight parameters and flight load data of each component. These flight data can be used for flight reproduction, fault analysis, health monitoring, fatigue life determination and the like of the helicopter. The correlation coefficient can obtain the correlation law of flight parameter-load, which can help to establish the quantitative / qualitative relationship between flight parameter and load. The method can be used for subsequent load abnormality cause diagnosis, load abnormality early warning and the like in the field of helicopter flight test.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of helicopter structure fatigue design, and relates to a key flight parameter screening method based on flight parameter-load correlation. BACKGROUND

[0002] A large amount of flight data, including numerous flight parameters and flight load data of each component, can be obtained through helicopter scientific research load flight test. The flight data can be used for flight reproduction, fault analysis, health monitoring, fatigue life determination and the like of the helicopter. How to efficiently analyze and process the high-dimensional data to obtain the correlation between numerous flight parameters and comprehensively consider the influence of the flight parameters on the flight load is a key to the above problems.

[0003] One method to solve the difficulty in high-dimensional data analysis and processing is variable screening. Variable screening can reduce the data from super-high dimension or high dimension to appropriate dimension, so that the traditional statistical model has better applicability and can help to establish a more reliable prediction and inference model. SUMMARY

[0004] OBJECTIVE

[0005] The application proposes a key flight parameter screening method based on flight parameter-load correlation, which can effectively reduce the size of flight parameters and improve efficiency, and has great engineering application prospect.

[0006] TECHNICAL SCHEME

[0007] The key flight parameter screening method based on flight parameter-load correlation is realized based on Pearson correlation coefficient.

[0008] The operation steps are as follows:

[0009] (1) Because there are many types of sensors on the helicopter, the working environment is poor, and the interference is more, so it is necessary to pretreat the flight data collected by the sensors. The pretreatment mainly removes some data jump points. Through the jump point removal, the data curve is smoother and more close to the true value.

[0010] (2) Determine the standard blade azimuth angle signal: process the blade azimuth angle signal into a standard blade azimuth angle signal, that is, convert the azimuth angle data into a peak signal that only jumps one point each time; the blade azimuth angle data recorded in the flight test is P j (j=1,…,n, n is the total number of test points), and the standard blade azimuth angle signal data is

[0011] (3) Determine the fundamental frequency period and the corresponding data points: the time T corresponding to each standard azimuth angle signal TOPi The time difference corresponding to the adjacent standard blade azimuth angle signals is a base frequency period T i+1 -T i , wherein T i and T i+1 are time points corresponding to two continuous standard azimuth angle signals; the number of data points in a base frequency period m = f × (T i+1 -T i , wherein f (unit: Hz) is a sampling rate;

[0012] (4) Processing flight parameter data and component load data in the i-th base frequency period: calculating the time start T i and the end T i+1 of the corresponding i-th base frequency period, and taking the maximum value S max,i , the minimum value S min,i and the average value S mean,i of the original data in the time period. For the flight parameter data, the average value in the time period from the start T i to the end T i+1 is taken: For the component load data, the static load and the dynamic load in the time period from the start T i to the end T i+1 are taken.

[0013] (5) The parameters of different categories and sizes are different, so it is necessary to normalize each category of parameters before using these parameters. The normalization formula can be various, and the following formula is generally used.

[0014]

[0015] After such processing, the value range of the parameters is converted to the range of (-1, 1).

[0016] (6) According to the Pearson correlation coefficient, the correlation coefficient between each input flight parameter and the flight parameter is calculated. For a two-dimensional random variable vector (X, Y), the Pearson correlation coefficient is defined as

[0017]

[0018] (X i ,Y i ), i = 1, … n is the sample data of (X, Y), and the calculation formula of the Pearson correlation coefficient is:

[0019]

[0020] Generally, the closer the absolute value of p is to 1, the stronger the monotonic relationship between the two entities.

[0021] Generally, the absolute value of the correlation coefficient ρ in the range of (0, 0.3) represents a weak correlation between variables X and Y, in the range of (0.3, 0.5) represents a real correlation, in the range of (0.5, 0.8) represents a significant correlation, and in the range of (0.8, 1) represents a high correlation. In this patent, if the absolute value of the Pearson correlation coefficient of variables X and Y is greater than or equal to 0.7, it is considered that the variables X and Y are highly correlated and have mutual replaceability. If the absolute value of the Pearson correlation coefficient is between (0, 0.7), it is considered that there is a certain correlation between the variables X and Y. Accordingly, the flight parameter matrix can be preliminarily screened and reduced in dimension;

[0022] (7) Calculate the correlation coefficient between the component load and the flight parameters after preliminary screening according to the Pearson correlation coefficient calculation formula in (6). The proportion of each flight parameter correlation coefficient is calculated using the following formula:

[0023]

[0024] According to the correlation coefficient proportion, the flight parameters with a cumulative proportion of 90% are screened out, and the remaining flight parameters are discarded, thereby further reducing the dimension of the flight parameters.

[0025] Further, before step 1 is implemented, data conversion and channel decomposition should be performed.

[0026] Further, the data conversion is specifically:

[0027] The voltage signal recorded by the measurement is converted into corresponding channel data using a data processing equation consistent with the calibration equation. The typical data calibration equation and the corresponding data processing equation are shown in the following formula:

[0028] S=k×V+b

[0029]

[0030] In the formula:

[0031] S——load value;

[0032] f b ——calibration voltage amplification factor;

[0033] f c ——measurement voltage amplification factor;

[0034] k——calibration equation slope;

[0035] b——calibration equation intercept;

[0036] V——recorded voltage;

[0037] V0——initial (parked state) voltage average value;

[0038] S0——initial (parked state) load, generally determined by calculation analysis method.

[0039] Further, in the implementation of step 1, the jump point removal method can be removed by manual removal of abnormal points through artificial judgment, or by programming means such as matlab and python, and batch removal based on statistical methods. The implementation method can be achieved by calculating the mean, variance and standard deviation of the data. Abnormal values usually cause the deviation of data points, which can be identified by statistical indicators to identify abnormal values in the data and determine whether it is a jump point. Further, ready-made algorithms can be used to detect jump points, such as Slope Change Detection (Slope Change Detection), Outlier Detection (Outlier Detection) and other automatic identification of jump points in the data, and removal. Further, if there are noise or pulse peaks in the data, filtering methods can be used to smooth the data and remove abnormal values. Common filtering methods include moving average, median filtering, Gaussian filtering, etc.

[0040] Further, the standard blade azimuth angle signal determined in step 2 can be generated based on the integral of the rotor speed to generate the azimuth angle file.

[0041] Further, the specific implementation of the azimuth angle file based on the integral of the rotor speed: whether it is a research machine or an in-service machine, a flight parameter acquisition and processing system is installed, and the rotor speed data of the helicopter can be extracted through the system. Rotor speed data is usually recorded by a series of data points and displayed in percentage. The data form is shown in the table below, and the total number of data points n is obtained.

[0042] Rotor speed percentage data

[0043]

[0044] According to the rotor speed parameters of the helicopter, the rated speed n (r / min) of the main rotor (or tail rotor) is obtained, and the formula is Convert the rotor speed percentage of the helicopter into instantaneous speed (r / s) as shown in the table below

[0045] Rotor instantaneous speed

[0046]

[0047] The total number of data points of the azimuth angle signal is calculated according to the total number of rotor speed data points n and the data sampling rate as needed. The method for calculating the total number of data points N of the azimuth angle signal is: Where f z is the rotor speed data sampling rate, fr The sampling rate for main propeller load or tail propeller load data. Based on the total number of data points N, establish the initial azimuth sequence, as shown in the table below. The initial azimuth value can be set to 0, i.e., p = 0.

[0048] Azimuth initial sequence

[0049]

[0050] via S = ∑v i / f z The instantaneous rotational speed of the rotor is accumulated. Where v i Let f be the instantaneous rotor speed at time i. z The rotor speed data sampling rate is [value]. When the cumulative result S≥1, record the data sequence number n′ at this time, and [replace the corresponding value in the initial azimuth sequence]. The azimuth signal at the location is reassigned p = p′ = 1 to form a pulse signal, and the cumulative result is set to 0. This process is repeated to finally generate the azimuth signal based on the rotor speed.

[0051] Furthermore, the normalization process in step 5 can be performed using the mean-variance method;

[0052] Furthermore, the mean-variance method generally uses the following formula:

[0053] x k =(x k -x mean ) / x var

[0054] In the formula x k For flight data, x mean Let x be the mean of the flight data series. var The variance of the flight data.

[0055] Furthermore, in step 6, the correlation coefficient between each input flight parameter and the flight parameters can be calculated using the Spearman correlation coefficient.

[0056] Furthermore, the specific implementation of the Spearman correlation coefficient: Similarly, for a random sample (X) i ,Y i ), i = 1,...n, and its sample estimation form is:

[0057]

[0058] in X represents i Similarly, the rank of {X1,...,Xn} is... Y represents i In {Y1,...,Yn The rank in}.

[0059] Furthermore, after step 7 is completed, the flight parameter screening method is verified using a flight parameter-dynamic load prediction algorithm based on a BP neural network.

[0060] The training samples were selected to predict the same payload using both unselected and selected flight parameters. The effectiveness of the flight parameter selection method was verified by comparing the accuracy and speed of the neural network payload prediction method.

[0061] The mean squared error of the validation set samples is used to measure the prediction accuracy of the neural network. The formula for calculating the mean squared error is:

[0062]

[0063] The beneficial effects of this application are as follows:

[0064] The advantages of this invention are: Through helicopter research payload flight tests, massive amounts of flight data can be obtained, including numerous flight parameters and flight load data for various components. This flight data can be used for helicopter flight reproduction, fault analysis, health monitoring, fatigue life determination, etc. The correlation coefficient reveals the correlation law between flight parameters and loads, which helps establish a quantitative / qualitative relationship between flight parameters and loads. This method can be used for subsequent diagnosis of load anomalies and early warning of load anomalies in helicopter flight testing. This method can reduce the dimensionality of flight parameter data from high to an appropriate dimension, thereby improving the applicability of traditional statistical models and helping to establish a more reliable prediction and inference model. Applying this method to flight load prediction based on neural network algorithms, single-aircraft fatigue life prediction, and helicopter HMS health monitoring can greatly improve the learning accuracy and prediction speed of neural networks, save computational resources, and has a very broad application prospect. Attached Figure Description

[0065] Figure 1 This is a schematic diagram showing the data jump point before it is cleared in a certain channel.

[0066] Figure 2 A schematic diagram showing the data jump point clearing process for a certain channel.

[0067] Figure 3 This is a schematic diagram of sample point extraction;

[0068] Figure 4 Correlation plot of mergeable flight parameter data;

[0069] Figure 5 This is a comparison chart of mean squared errors;

[0070] Figure 6 This is a comparison chart of training time;

[0071] Figure 7 The flow chart is realized for the flight parameter-dynamic load prediction algorithm based on the BP neural network. DETAILED DESCRIPTION

[0072] For the purpose, technical scheme and advantages of the implementation of the present application, the technical scheme in the embodiment of the present application will be described in more detail below in combination with the drawings in the embodiment of the present application. In the drawings, the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The described embodiments are part of the embodiments of the present application, not all the embodiments. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. The embodiments of the present application will be described in detail below in combination with the drawings.

[0073] (1) The pre-processed flight parameter channels are determined as shown in Table 1, and the component load channel is taken as the main blade main blade 3639 flap bending moment dynamic load.

[0074] Table 1 is a flight parameter table of a certain type of aircraft

[0075]

[0076]

[0077]

[0078] (2) First, the data is pre-processed, i.e. removing jump points, limiting amplitude and smoothing processing, and the processing effect is shown in Figure 1 、 Figure 2 .

[0079] (3) According to the blade azimuth angle file, the peak value and the valley value of the pitch rod load in a rotation period are extracted, the static load and the dynamic load are obtained, and the average value of each flight parameter in the corresponding time period is read as the corresponding flight parameter, as shown in Figure 3 ;

[0080] (4) The following formula is used to normalize the data of each flight parameter and load.

[0081]

[0082] After such processing, the value range of the parameter is converted to the range of (-1, 1).

[0083] (5) According to Spearman correlation coefficient, the correlation coefficient between each input flight parameter and flight parameter is calculated, see Table 2. The variables with absolute value of Spearman correlation coefficient greater than or equal to 0.7 are merged, and the merging schematic diagram is shown in Figure 4 Accordingly, the flight parameter matrix is preliminarily screened and reduced in dimension, and 31 flight parameters can be preliminarily reduced to 15 flight parameters, and the dimension reduction effect is shown in Table 3.

[0084] Table 2 Correlation coefficient of each flight parameter and main blade 3639 flap bending moment dynamic load

[0085]

[0086]

[0087]

[0088] The preliminarily screened flight parameters are shown in the following table.

[0089] Table 3 Flight parameter table after preliminary screening

[0090]

[0091]

[0092] (6) According to the correlation relationship between each flight parameter in (5), if the absolute value of the correlation coefficient of any two variables of the above preliminarily screened flight parameters is between (0.5, 0.7), it is considered that there is significant correlation between the two variables. Assuming that there is a variable A, and a variable B and several other variables are significantly correlated, and the correlation coefficient variable A> variable B>…, then the following operation is performed: the variable A with large correlation coefficient is the priority variable, its correlation coefficient remains unchanged, the rest of the significantly correlated variables are subordinate variables, their correlation coefficients are multiplied by 0.5, and for any variable, its correlation coefficient is multiplied by 0.5 only once.

[0093] (7) According to the significant correlation relationship processing and sorting according to the correlation coefficient ratio, the flight parameters with cumulative proportion of 90% are screened out, and the rest of the flight parameters are discarded, so as to further reduce the flight parameters. The screened flight parameters are shown in the following table.

[0094] Table 4 Correlation coefficient of flight parameter and load

[0095]

[0096]

[0097] (8) The prediction results of the main blade 3639 flap bending moment dynamic load are selected for display. The calculated working conditions include two, working condition 1 is that 35 flight parameter data are all used for neural network training, and working condition 2 is that 13 effective flight parameters are selected according to a cumulative contribution rate of 90% for neural network training, as shown in Table 4. The BP neural network model is based on the Matlab neural network toolbox, the training function is selected as the Lecenberg-Marquardt function, the learning efficiency is set to 0.1, the target error is 1e-4, and the hidden layer neuron node is set to 5, 10, 20 and 50.

[0098] (9) The smaller the prediction mean square error is, the higher the prediction accuracy of the network is; the shorter the running time is, the higher the network calculation efficiency is. Table 5 and Figure 5 、 Figure 6 The prediction results of the BP neural network for the normalized dynamic load under the typical flight state are compared by using the full flight parameter input and the flight parameter input based on the cumulative contribution rate screening. As can be seen from the chart, the prediction accuracy and calculation efficiency of the network using the flight parameters screened based on the cumulative contribution rate as the training input of the BP neural network are obviously better than those of the neural network using all flight parameters as the training input. This shows that the method of screening flight parameters based on the Spearman correlation coefficient and the cumulative contribution rate proposed in the present application can effectively improve the accuracy and calculation efficiency of the neural network load prediction model.

[0099] Table 5 Dynamic load prediction results

[0100]

[0101]

[0102] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless otherwise defined. The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application, and it should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any person skilled in the art can make changes or modifications to the above disclosed technical contents to obtain equivalent embodiments applied to other fields within the spirit and principles of the present application, but any simple modification, equivalent change and improvement made to the above embodiments in accordance with the technical essence of the present application should be included in the protection scope of the present application.

Claims

1. A method for screening key flight parameters based on the correlation between flight parameters and payload, characterized in that, This is achieved based on the Pearson correlation coefficient. The operating steps are as follows: Step (1) Preprocess the flight data collected by the sensors; preprocessing removes jump points; Step (2) Determine the standard blade azimuth signal: Process the blade azimuth signal into a standard blade azimuth signal, that is, convert the azimuth data into a peak signal that jumps to only one point each time; the blade azimuth data recorded in the flight test is P j If j = 1, ..., n, where n is the total number of test points, then the standard blade azimuth signal data is: Step (3) Determine the fundamental frequency period and the corresponding number of data points: the time T corresponding to each standard azimuth signal TOP. i The time difference between adjacent standard blade azimuth angle signals is the fundamental frequency period T. i+1 -T i T i and T i+1 The time points corresponding to two consecutive standard azimuth angle signals; the number of data points within the fundamental frequency period m = f × (T) i+1 -T i ), where f, in Hz, is the sampling rate; Step (4) Process the flight parameter data and component load data within the i-th fundamental frequency cycle: Calculate the time start point T corresponding to the i-th fundamental frequency cycle. i To the finish line T i+1 The maximum value S of the original data within the time period max,i Minimum value S min,i and average value S mean,i For flight parameter data, take the starting point T. i To the finish line T i+1 Average value over the time period: For component load data, take the starting point T. i To the finish line T i+1 Static load over time period Dynamic load Step (5) involves normalizing the parameters using the following formula: After this normalization process, the range of parameter values ​​is transformed to the range of (-1, 1); Step (6) Calculate the correlation coefficient between each input flight parameter and the flight parameters based on the Pearson correlation coefficient; for a two-dimensional random variable vector (X,Y), the Pearson correlation coefficient is defined as... (X i ,Y i Let i = 1, ..., n be the sample data of (X, Y). Then the formula for calculating the Pearson correlation coefficient is: The closer the absolute value of ρ is to 1, the stronger the monotonic relationship between the two entities. The absolute value of the correlation coefficient ρ indicates a slight correlation between variables X and Y if it is in the range of (0, 0.3), a real correlation if it is in the range of (0.3, 0.5), a significant correlation if it is in the range of (0.5, 0.8), and a high correlation if it is in the range of (0.8, 1). If the absolute value of the Pearson correlation coefficient between variables X and Y is greater than or equal to 0.7, then variables X and Y are considered to be highly correlated and mutually substitutable. If the absolute value of the correlation coefficient is between (0, 0.7), then variables X and Y are considered to have a certain correlation pattern. Based on this, the flight parameter matrix can be initially screened and its dimensionality reduced. Step (7) Calculate the correlation coefficient between the component load and the flight parameters after initial screening according to the Pearson correlation coefficient calculation formula in (6); calculate the proportion of the correlation coefficient of each flight parameter using the following formula: The flight parameters are sorted according to their correlation coefficient percentage, and those with a cumulative percentage of 90% are selected. The remaining flight parameters are discarded, thereby further reducing the dimensionality of the flight parameters.

2. The method as described in claim 1, characterized in that, Before implementing step (1), data conversion and channel decomposition should be performed.

3. The method as described in claim 2, characterized in that, The data transformation specifically refers to: The measured voltage signal is converted into corresponding channel data using a data processing equation consistent with the calibration equation. A typical data calibration equation and its corresponding data processing equation are shown below: S=k×V+b In the formula: S—Load value; f b —Calibrate voltage amplification factor; f c —Measuring voltage amplification factor; k — slope of the calibration equation; b—the intercept of the calibration equation; V — Record voltage; V0—The average value of the initial voltage; S0 – Initial load, determined using calculation and analysis methods.

4. The method as described in claim 3, characterized in that, When implementing step (1), in addition to manually removing outliers by judging them manually, the jump point removal can also be carried out in batches by using MATLAB or Python programming based on statistical methods. If there is noise or pulse peaks in the data, filtering methods are used to smooth the data and remove outliers. Commonly used filtering methods include moving average, median filtering, and Gaussian filtering.

5. The method as described in claim 4, characterized in that, In step (2), the standard blade azimuth angle signal can be determined by generating an azimuth angle file based on rotor speed integration.

6. The method as described in claim 5, characterized in that, The specific implementation of generating azimuth file based on rotor speed integration is as follows: helicopter rotor speed data is extracted through the flight parameter acquisition and processing system. The rotor speed data is recorded using a series of data points and displayed as a percentage.

7. The method as described in claim 6, characterized in that, The normalization process in step (5) can be performed using the mean-variance method.

8. The method as described in claim 7, characterized in that, The mean-variance method is described using the following formula: x k =(x k -x mean ) / x var In the formula x k For flight data, x mean Let x be the mean of the flight data series. var The variance of the flight data.

9. The method as described in claim 8, characterized in that, Step (6) calculates the correlation coefficient between each input flight parameter and the flight parameters using the Spearman correlation coefficient.

10. The method as described in claim 9, characterized in that, The specific implementation of the Spearman correlation coefficient: Similarly, for a random sample (X) i ,Y i ), i = 1,...n, and its sample estimation form is: in X represents i In {X1,...,X n The rank of} is similar. Y represents i In {Y1,...,Y n rank in}; After step (7) is completed, the flight parameter screening method is verified by the flight parameter-dynamic load prediction algorithm based on BP neural network. The training samples were selected to predict the same payload using both unselected and selected flight parameters. The effectiveness of the flight parameter selection method was verified by comparing the accuracy and speed of the neural network payload prediction method. The mean squared error of the validation set samples is used to measure the prediction accuracy of the neural network; the formula for calculating the mean squared error is:

Images