1. An aviation equipment vibration environment analysis and prediction method is characterized by comprising the following steps:

s1, for the aviation equipment installed on the aircraft platform, the mapping relation between the aircraft platform state parameter U and the vibration environment magnitude value G is expressed as follows:

G＝f(U)

s2, dividing the mission profile of the aircraft into n states according to the fact that the vibration response of a measuring point of the aircraft platform where the aviation equipment is located is different along with the difference of flight states, wherein the vibration environment quantity value set G of the measuring point of the aircraft platform is expressed as:

G＝(g₁，g₂，g₃，.....g_n)

wherein G is a set of vibration environment quantity values with the capacity of n, G_iThe vibration environment value of a certain measuring point of the aircraft platform in the ith flight state is represented by i ═ 1-n;

meanwhile, the aircraft state parameter set U of the aircraft platform is as follows:

U＝(u₁，u₂，u₃，.....u_n)

wherein U is a set of aircraft state parameters with a capacity of n, U_iThe status parameter is the ith status parameter, i is 1-n;

s3, normalization processing:

carrying out normalization processing on the aircraft state parameter set U to obtain a normalized aircraft state parameter set U', wherein:

U′＝(u′₁，u′₂.....u′_n)

in formula (II) u'_iThe normalized ith state parameter is i ═ 1-n;

normalizing the vibration environment quantity value set G to obtain a normalized vibration environment quantity value set G', wherein:

G′＝(g′₁，g′₂.....g′_n)

in formula (II), g'_iThe ith vibration magnitude value after normalization is 1-n;

s4, constructing a group of deep learning networks R₁Carrying out feature classification on the normalized aircraft state parameter set U', wherein R comprises two levels of a visible layer and a hidden layer, the visible layer comprises n nodes, the hidden layer comprises k nodes, and the visible layer and the hidden layer are not connected in the layers and are fully connected in the layers; the hidden layer node is provided with an activation function f for converting an input state parameter set U 'into a new characteristic parameter set U'; w is the weight between layers, a_iFor visible layer node biasing, b_kBiasing for hidden nodes;

S5repeatedly constructing a plurality of groups of deep learning network classifiers R₂--R_XR is adjusted according to the requirements of the previous steps₁--R_XStacking up and R_XThe hidden layer output node is connected with a neural network with an input node k and an output node m;

s6, collecting S groups U' and a feature classifier R₁Connected to a feature classifier R₁--R_XParameter (w, a)_i，b_k) Assigning an initial value;

s7, inputting the result G 'after the S groups U' are collected, analyzing and calculating the error of G 'and G', and adjusting the classifier R₁--R_XParameter (w, a)_i，b_k) And the structural parameters of the neural network are calculated until the error meets the requirement, and then the modeling of the vibration environment prediction model is completed; setting an error threshold value delta E, calculating the error E of U 'and U', and adjusting the parameters (w, a) of the classifier R when E is larger than delta E_i，b_k) And the structural parameters of the neural network until the error E is less than delta E;

and S8, inputting any group of aircraft platform state parameters U to the model established in the step S7, calculating the vibration environment magnitude value G of the corresponding aircraft platform, and realizing analysis and prediction of the vibration environment magnitude value.

2. The method for analyzing and predicting the vibration environment of aircraft equipment according to claim 1, wherein the aircraft state parameters are parameters affecting the magnitude of the vibration environment of the aircraft platform, and comprise the altitude of the aircraft, the flight speed, the attack angle of the aircraft and the engine speed.

3. The method for analyzing and predicting the vibration environment of the aviation equipment according to claim 1, wherein the vibration environment quantity value is a numerical value representing the vibration degree of the platform and is expressed by a vibration root mean square value (RMS).

4. The method of claim 1, wherein the feature classifier is a deep network having one or more of a self-encoder, a constrained boltzmann machine, and/or a recurrent neural network.

5. The method of analyzing and predicting the vibratory environment of aerospace equipment of claim 1, wherein the feature classifier R₁--R_XParameter (w, a)_i，b_k) Is set to 0, a_i0 and b_k＝0。

Background

The combined action of multiple vibration sources such as power devices, rotating parts and the like inside the platform and complex aerodynamic loads outside the platform on the aeronautical equipment installed on various aircraft platforms makes the vibration environment experienced in the service life very harsh, and is an important cause of structural damage, performance degradation and functional failure of the aeronautical equipment. Therefore, the extreme value of the vibration environment in the life of the aviation equipment needs to be accurately determined, and design and test verification work is carried out according to the extreme value, so that the normal and safe use of the aviation equipment is ensured.

At present, an effective means for determining the extreme value of the vibration environment of the aviation equipment is to carry out actual measurement work of the vibration environment on an installation platform of the aviation equipment, and the actual measurement data of the vibration environment of the platform is effectively analyzed to determine the extreme value of the vibration environment.

The vibration environment quantity value of the aircraft platform is directly determined by the state parameters of the aircraft, but all limit state parameters of the aircraft cannot be measured generally due to the restriction of some factors in the actual measurement process of the vibration environment, so that the result of analysis based on the obtained measured data cannot represent the vibration environment limit quantity value in the life cycle of the aviation equipment, and the subsequent design and test verification of the aviation equipment are influenced.

The state parameter of the aircraft is a multi-factor set, and certain coupling exists among all factors; in addition, flight states are different, and weights of all state parameters are also different, so that a highly complex nonlinear relationship is formed between the aircraft state parameters and the platform vibration response, and in order to ensure the accuracy of the result of the vibration response prediction, a plurality of state parameters need to be classified firstly, but the classification of the plurality of state parameters by means of manual methods depends on human experience, so that the prediction results are different from person to person, and great difference exists.

Disclosure of Invention

The invention aims to provide a method for analyzing and predicting a vibration environment of aviation equipment.

In order to achieve the above object, the present invention provides an aviation equipment vibration environment analysis and prediction method, which includes the following steps:

s1, for the aviation equipment installed on the aircraft platform, the mapping relation between the aircraft platform state parameter U and the vibration environment magnitude value G is expressed as follows:

G＝f(U)

s2, dividing the mission profile of the aircraft into n states according to that the vibration response of a measuring point of the aircraft platform where the aviation equipment is located is different along with the difference of flight states, wherein the vibration environment magnitude value G of the measuring point of the aircraft platform is represented as:

G＝(g₁,g₂,g₃,…g_n)

wherein G is a set of vibration environment quantity values with the capacity of n, G_iThe vibration environment value of the jth area of the aircraft platform is represented by i which is 1-n;

meanwhile, the aircraft state parameter set U of the aircraft platform is as follows:

U＝(u₁,u₂,u₃,…u_n)

wherein U is a set of aircraft state parameters with a capacity of n, U_iThe status parameter is the ith status parameter, i is 1-n;

s3, normalization processing:

carrying out normalization processing on the aircraft state parameter set U to obtain a normalized aircraft state parameter set U', wherein:

U'＝(u'₁,u'₂…u'_n)

in formula (II) u'_iThe normalized ith state parameter is i ═ 1-n;

normalizing the vibration environment quantity value set G to obtain a normalized vibration environment quantity value set G', wherein:

G'＝(g'₁,g'₂…g'_n)

in formula (II), g'_jThe ith vibration magnitude value after normalization is 1-n;

s4, constructing a group of deep learning networks R₁Carrying out feature classification on the normalized aircraft state parameter set U', wherein R comprises two levels of a visible layer and a hidden layer, the visible layer comprises n nodes, the hidden layer comprises k nodes, and the visible layer and the hidden layer are not connected in the layers and are fully connected in the layers; the hidden layer node is provided with an activation function f for converting an input state parameter set U 'into a new characteristic parameter set U'; w is the weight between layers, a_iFor visible layer node biasing, b_kBiasing for hidden nodes;

s5, repeatedly constructing multiple groups of deep learning network classifiers R₂--R_XR is adjusted according to the requirements of the previous steps₁--R_XPiling up and R_XThe hidden layer output node is connected with a neural network with an input node k and an output node m;

s6, collecting S groups U' and a feature classifier R₁Connected to a feature classifier R₁--R_XParameter (w, a)_i，b_k) Assigning an initial value;

s7, inputting the result G 'after the set of S groups U', analyzing and calculating the error between G 'and G', and adjusting the classifier R₁--R_XParameter (w, a)_i，b_k) And the structural parameters of the neural network are calculated until the error meets the requirement, and then the modeling of the vibration environment prediction model is completed; setting an error threshold value delta E, calculating the error E of U 'and U', and adjusting the parameters (w, a) of the classifier R when E is larger than delta E_i，b_k) And the structural parameters of the neural network until the error E is less than delta E;

and S8, inputting any group of aircraft platform state parameters U to the model established in the step S7, calculating the vibration environment magnitude value G of the corresponding aircraft platform, and realizing prediction of the vibration environment magnitude value.

In a preferred embodiment, the aircraft state parameter refers to a parameter which affects the magnitude of the vibration environment of the aircraft platform, and includes aircraft altitude, flight speed, aircraft angle of attack, and engine speed.

In a preferred embodiment, the vibration environment quantity value refers to a numerical value representing the vibration degree of the platform, and is expressed by a vibration root mean square value RMS.

In a preferred embodiment, the feature classifier is a deep network, and the network architecture is one or more of a self-encoder, a constrained boltzmann machine and/or a recurrent neural network.

Preferably, the feature classifier R₁--R_XParameter (w, a)_i，b_k) Is set to 0, a_i0 and b_k＝0。

The invention has the following effects: the method adopts a combination mode of the feature classifier and the neural network, solves the nonlinear problem in the vibration environment prediction, and has higher prediction precision compared with the traditional linear regression prediction method. Meanwhile, the characteristic classifier adopted by the invention can automatically and effectively extract characteristic parameters from a plurality of aircraft state parameters, and adopts fewer parameters to describe the influence factors of the platform vibration environment, thereby eliminating the dependence of manual classification on prior data or knowledge of operators and effectively optimizing the classification result.

Drawings

FIG. 1 is a schematic flow diagram of a method for analyzing and predicting a vibration environment of an aviation equipment according to the present invention;

FIG. 2 is a schematic diagram of the vibration prediction modeling process based on the deep learning network according to the present invention;

FIG. 3 is a schematic diagram of a set of deep learning networks R1 constructed by the present invention;

FIG. 4 is a schematic diagram of the adjustment of structural parameters of the neural network of the present invention;

FIG. 5 is a schematic diagram of the results of predicting the RMS values of the vibrations of an aircraft.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to examples, but it will be understood by those skilled in the art that the following examples are only illustrative of the present invention and should not be construed as limiting the scope of the present invention.

As shown in fig. 1, the method for analyzing and predicting the vibration environment of the aviation equipment provided by the invention comprises the following steps:

s1, establishing a mapping relation between the aircraft platform state parameter U and the vibration environment magnitude value G for the aviation equipment installed on the aircraft platform, specifically expressed as:

G＝f(U)

s2, according to the fact that the vibration response of a certain measuring point of an aircraft platform where aviation equipment is located is different along with the difference of flight states, dividing an aircraft task section into n flight states, and determining a vibration environment magnitude value G and an aircraft state parameter set U of the aircraft platform, wherein the vibration environment magnitude value G of the aircraft platform is represented as:

G＝(g₁,g₂,g₃,…g_n)

wherein G is a set of vibration environment quantity values with the capacity of m, G_iThe vibration environment value of a certain measuring point of the aircraft platform in the ith flight state is represented by i ═ 1-n;

meanwhile, the aircraft state parameter set U of the aircraft platform is as follows:

U＝(u₁,u₂,u₃,…u_n)

wherein U is a set of aircraft state parameters with a capacity of n, U_iThe status parameter is the ith status parameter, i is 1-n;

s3, normalization processing:

carrying out normalization processing on the aircraft state parameter set U to obtain a normalized aircraft state parameter set U', wherein:

U'＝(u'₁,u'₂.....u'_n)

in formula (II) u'_iThe normalized ith state parameter is i ═ 1-n;

normalizing the vibration environment quantity value set G to obtain a normalized vibration environment quantity value set G', wherein:

G'＝(g'₁,g'₂…g'_n)

in formula (II), g'_iThe ith vibration magnitude value after normalization is 1-n; the purpose of normalization is to allow the preprocessed data to be limited to a certain range, such as [0,1]]Thereby eliminating the adverse effects caused by singular sample data.

S4, constructing a group of deep learning networks R₁Carrying out feature classification on the normalized aircraft state parameter set U', wherein R comprises two levels of a visible layer and a hidden layer, the visible layer comprises n nodes, the hidden layer comprises k nodes, and the visible layer and the hidden layer are not connected in the layers and are fully connected in the layers; the hidden layer node is provided with an activation function f for converting an input state parameter set U 'into a new characteristic parameter set U'; w is the weight between layers, a_iFor visible layer node biasing, b_kHidden nodes are biased. The purpose of the deep learning network R as a feature classifier is to transform the feature representation of the sample in the original space to a new feature space by layer-by-layer feature transformation, thereby making classification or prediction easier.

S5, repeatedly constructing multiple groups of deep learning network classifiers R₂--R_XR is adjusted according to the requirements of the previous steps₁--R_XPiling up and R_XK is between the hidden layer output node and the input nodeThe neural networks with m output nodes are connected;

s6, collecting S groups U' and a feature classifier R₁Connected to a feature classifier R₁--R_XParameter (w, a)_i，b_k) Assigning an initial value;

s7, inputting the result G 'after the set of S groups U', analyzing and calculating the error between G 'and G', and adjusting the classifier R₁--R_XParameter (w, a)_i，b_k) And the structural parameters of the neural network are calculated until the error meets the requirement, and then the modeling of the vibration environment prediction model is completed; setting an error threshold value delta E, calculating the error E of U 'and U', and adjusting the parameters (w, a) of the classifier R when E is larger than delta E_i，b_k) And the structural parameters of the neural network until the error E is less than delta E;

and S8, inputting any group of aircraft platform state parameters U to the model established in the step S7, calculating the vibration environment magnitude value G of the corresponding aircraft platform, and realizing analysis and prediction of the vibration environment magnitude value.

The method for analyzing and predicting the vibration environment of the aviation equipment is specifically described below with reference to a specific example, which is specifically shown in fig. 2.

The method comprises the steps of taking certain type of aviation equipment arranged on a fixed-wing jet aircraft as an object, selecting the flight height and flight speed (Mach number) of the aircraft as input parameters of a vibration prediction model, and taking the root mean square value of a vibration environment standard spectrum of a certain type of aviation equipment installation area as a prediction model for output.

A prediction model consisting of a plurality of classifiers stacked by limiting Boltzmann machines (Retnicted Boltzmann Mach ines) and a regression layer or classification layer of a neural network is constructed. Because the input parameters of the prediction model are the flight altitude and the flight speed (mach number), the number of nodes of the input layer of the model is 2; the output parameter of the prediction model is the root mean square value, so the number of nodes of the output layer of the model is 1.

Selecting a root mean square value corresponding to a vibration specification spectrum of certain type of aviation equipment within the range of the flight altitude of 670-11920m and the flight speed of 0.25-1.52 Mach number of the fixed-wing jet aircraft, wherein the root mean square maximum value is 10g, and totaling 252 groups of samples.

And carrying out normalization processing on 252 groups of samples according to a maximum and minimum normalization method, and projecting attribute data to a [0,1] interval in proportion.

Taking 220 groups of normalized samples as a training set, and taking the rest 32 groups of normalized samples as a test set;

as shown in fig. 3, a group is constructed, which includes a deep learning network R1, where R1 is composed of 1 DBN network (feature extractor) and 1 NN network (classifier), where an input layer includes 2 nodes (representing altitude and flight parameters), an output layer includes 1 node (representing vibration root mean square value), and the DBN deep learning network further includes 2 hidden layers, each layer includes 10 nodes. Thus, the DBN model is formed by stacking 2 RBMs, namely RBM1 and RBM2, with the output of the former serving as the input to the latter.

Referring to fig. 4, the abscissa represents the number of model training iterations, the number of training iterations is set to 100 for the RBM1 model, a normalized state parameter vector U' (220 rows and 2 columns) is input, and 3 parameters (w) of the model are passed_ijIs the weight between any two connected neuron nodes between layers, a_iFor visible layer node biasing, b_kHidden layer node bias), the initial values of the model parameters are all 0, the output vector h1(220 rows and 1 columns) of the RBM1 model is calculated, the maximum likelihood function is optimized by adopting a gradient ascending algorithm, the parameters in the model are solved, the hidden layer can more accurately display the characteristics of the display layer after the training upper limit (100 times) is reached, and meanwhile, the display layer can be restored, namely, the characteristic vector mapping is optimal;

taking the output of the RBM1 as the input of an RBM2 model, repeating the operation of the steps, and calculating the parameters of the DBN model as follows;

outputting the model in the DBN feature extractor to an NN model for regression fitting, verifying by adopting 32 groups of data in a test set, and respectively calculating RMSE (root mean square error), MAE (mean absolute error) and MRE (mean relative error);

adjusting the number of hidden layer nodes in the DBN feature extractor, setting 19 groups of models of 10,15,20,25, … … and 100, repeating the related steps, comparing three types of errors of the models in each group, and determining the optimal network size to be [60,60] after optimization.

And predicting the vibration root mean square value of the aircraft according to the prediction model after the network parameters are determined, and obtaining a result shown in the attached figure 5.

The method for analyzing and predicting the vibration environment of the aviation equipment, provided by the invention, is characterized in that on the basis of a large amount of obtained actual test data, the state parameters of the aircraft platform are classified based on deep learning, and fewer parameters are adopted to describe the influence factors of the vibration environment of the platform, so that the nonlinear problem in the prediction of the vibration environment is solved; and predicting the vibration response corresponding to the unknown extreme state parameter at the safety range boundary by establishing a mapping relation between the aircraft state parameter and the platform vibration response.

The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements made to the technical solution of the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope defined by the claims of the present invention.