1. A method of generating a dithered turbulence response signal to an impulse response signal, comprising the steps of:

step 1: the turbulent flow response signal acquired by the flutter flight test is subjected to band-pass filtering in a frequency analysis interval to generate a band-pass filtered turbulent flow response signal;

step 2: constructing a flutter-based impulse response generation system;

step 2-1: generating a turbulence response signal and a corresponding impulse response signal through simulation, and taking the turbulence response signal and the corresponding impulse response signal as training data of an impulse response generation system;

step 2-2: taking a turbulence response signal generated by simulation as an input, sequentially passing through an encoder, a midle and a decoder, and finally outputting an impulse response corresponding to the turbulence response generated by the simulation for error calculation;

the encoder and the decoder are formed by taking one-dimensional convolution, a ReLU activation function and BatchNormal as basic structures, and the middle is formed by an LSTM network structure and used for extracting frequency domain characteristics of the impulse response signal;

step 2-3: carrying out network structure optimization on the impulse response generating system by adopting a Pythrch deep learning frame according to the system training data generated in the step 2-1 to obtain the final impulse response generating system composition and parameter information;

and step 3: and the final impulse response generating system is used for generating the impulse response signal of the turbulent response data tested by the flutter test in the actual engineering.

Background

Because the application environment requirement of an aircraft or an aeroelastic structure requires that atmospheric turbulence excitation always acts on a structural system, in the test flight of the aeroelastic system, modal analysis of the structure is carried out through a response signal of the atmospheric turbulence excitation, which is a common structural modal analysis method, however, since the turbulence response signal-to-noise ratio is low, the modal parameter estimation needs complex calculation; the most ideal modal parameter estimation is realized through the impulse response, but in the actual flutter test flight, the problems of high difficulty in obtaining the impulse response, high risk, poor data validity and the like are solved, so that the modal parameter identification based on the impulse response signal is difficult to be used as a unique modal parameter identification method in the actual engineering test. Further, atmospheric turbulence excitation usually represents random excitation, which cannot be measured in practical engineering, and therefore, the method of analyzing the structural system by using excitation and response signals cannot be realized under the condition of turbulence excitation, and usually requires the modal parameter estimation of the structural system under the condition of only knowing the response.

The common natural excitation signal modal parameter identification method at the present stage mainly takes a random decrement technology, a random subspace, an autoregressive modeling and an expansion method thereof and the like as main components, and the method mainly has the following problems:

(1) for a turbulent response signal in a single mode, the random decrement technology can effectively calculate an impulse response signal corresponding to the signal, but the problem of dense modes often exists in an actual flutter test flight test, and the random decrement technology is difficult to obtain an effective impulse response signal;

(2) the stochastic subspace method needs to traverse modal parameters of multiple orders, and then determines stable modal parameters (frequency and damping) by taking the modal parameters of each order as features, however, as the traverse orders increase, the time consumption of an algorithm for calculating the modal parameters is very high, real-time processing is difficult to achieve, and the time complexity and the space complexity of the algorithm have a great deal of difficulty in analyzing the real-time modal parameters, which is a main disadvantage that the method cannot process flutter test flight turbulence response data on line. On the other hand, since the stable mode needs to be analyzed through the mode parameters of a plurality of orders, the problem of the stable mode at the present stage is mainly based on a clustering method, but for the determination of the stable point, the order determination based on the response signal is the main difficulty of the method at present, and the order determination needs to be realized through artificial determination.

(3) The main problem of the related methods such as autoregressive modeling is how to determine the order, the autoregressive modeling method firstly needs to determine the order of modeling, then a coefficient equation is established based on the order to carry out parameter optimization, the first step of the process is to determine the order to be modeled, and at present, the step is usually manually realized by related data processing personnel.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides a method for generating a flutter turbulence response signal to an impulse response signal, which comprises the steps of firstly constructing an impulse response generating system based on a flutter test, wherein the system is a deep learning model and comprises an encoder, a midle and a decoder, the encoder and the decoder are formed by taking one-dimensional convolution, a ReLU activation function and BatchNormal as basic structures, and the midle is formed by an LSTM network structure and is used for extracting the frequency domain characteristics of the impulse response signal; then, taking the new turbulence response signal and the corresponding impulse response signal as training data of an impulse response generation system, and obtaining a final impulse response generation system by adopting a Pythrch deep learning framework; and finally, deploying the final impulse response generation system in actual engineering. The invention can obtain a better modal parameter estimation result on the premise of ensuring the real-time performance of the algorithm.

The technical scheme adopted by the invention for solving the technical problem comprises the following steps:

step 1: the turbulent flow response signal acquired by the flutter flight test is subjected to band-pass filtering in a frequency analysis interval to generate a band-pass filtered turbulent flow response signal;

step 2: constructing a flutter-based impulse response generation system;

step 2-1: generating a turbulence response signal and a corresponding impulse response signal through simulation, and taking the turbulence response signal and the corresponding impulse response signal as training data of an impulse response generation system;

step 2-2: taking a turbulence response signal generated by simulation as an input, sequentially passing through an encoder, a midle and a decoder, and finally outputting an impulse response corresponding to the turbulence response generated by the simulation for error calculation;

the encoder and the decoder are formed by taking one-dimensional convolution, a ReLU activation function and BatchNormal as basic structures, and the middle is formed by an LSTM network structure and used for extracting frequency domain characteristics of the impulse response signal;

step 2-3: carrying out network structure optimization on the impulse response generating system by adopting a Pythrch deep learning frame according to the system training data generated in the step 2-1 to obtain the final impulse response generating system composition and parameter information;

and step 3: and the final impulse response generating system is used for generating the impulse response signal of the turbulent response data tested by the flutter test in the actual engineering.

The invention has the following beneficial effects:

according to the method, the turbulence response signal is converted into the corresponding impulse response signal through a data driving method, the impulse response signal of the natural excitation response is obtained through a data driving model, and then the modal parameter of the system is obtained through the impulse response signal, so that a better modal parameter estimation result can be obtained on the premise of ensuring the real-time performance of the algorithm.

Drawings

FIG. 1 is a flow chart of an impulse response signal generating system according to the method of the present invention.

FIG. 2 is a block diagram of an impulse response generation system of the method of the present invention.

Detailed Description

The invention is further illustrated with reference to the following figures and examples.

The invention provides an impulse response signal generation method, which is characterized in that a convolution process is considered in the mathematical principle between a turbulence response signal and an impulse response signal of a structural system, and the impulse response signal is difficult to measure, so that the impulse response signal is generated based on a generation model through the measurable turbulence response signal, and for a generation model which is trained, the generation of the impulse response signal for the turbulence response signal of actual test flight can be realized only through the loading and calculation of the model.

As shown in fig. 1, a method for generating a flutter turbulent flow response signal to an impulse response signal includes the following steps:

step 1: the turbulent flow response signal acquired by the flutter flight test is subjected to band-pass filtering in an analysis frequency interval to obtain a band-pass filtered turbulent flow response signal, and the step is used as a common means of signal processing in actual engineering and mainly aims to filter information of irrelevant frequency bands and obtain useful information related to analysis frequency;

step 2: constructing an impulse response generating system based on flutter, and calculating an impulse response signal corresponding to a turbulent response signal, wherein the impulse response signal generating system is used as a deep learning model, needs model pre-training, and is convenient for practical engineering application, and the specific steps are as follows:

step 2-1: generating a turbulent flow response signal and a corresponding impulse response signal through simulation, and taking the simulated turbulent flow response signal and the corresponding impulse response signal as training data of an impulse response generation system;

step 2-2: as shown in fig. 2, a simulated turbulent flow response signal is taken as an input and sequentially passes through an encoder, a midle and a decoder, and finally an impulse response corresponding to the turbulent flow response generated by the simulation is output for error calculation;

the encoder and the decoder are formed by taking one-dimensional convolution, a ReLU activation function and BatchNormal as basic structures, and the middle is formed by an LSTM network structure and used for extracting frequency domain characteristics of the impulse response signal;

step 2-3: carrying out network structure optimization on the impulse response generation system by using the simulation training data generated in the step 2-1 by using a Pythrch deep learning frame to obtain the final impulse response generation system composition and parameter information;

and step 3: and the final impulse response generating system is used for generating an impulse response signal of turbulent response data of flutter test flight in actual engineering, and the method comprises the steps of deploying the impulse response generating system and processing engineering test data.

The flow of the impulse response signal generation system is shown in fig. 1, in the figure, the impulse response signal generation system trained based on the simulation signal is established according to the content described in step 2, and the model parameters of the impulse response signal generation system are determined through the simulation data because the engineering test data often cannot obtain an ideal impulse excitation response signal in practice and cannot be used as a data set of a deep learning model for optimizing and training the model, so that the model is optimized based on the simulation data, and then the model is deployed for generating the impulse response signal corresponding to the turbulent response signal of the flutter test flight.

The specific implementation mode is divided into two parts and comprises network parameter optimization of the impulse response signal generation system and deployment application of the model.

Firstly, establishing a data set aiming at a designed impulse response signal generation system network model according to a training and deployment principle of a deep learning algorithm, and optimizing and parameter optimizing the model, wherein the data set is established through simulation to establish an impulse response and turbulence response construction data set for network parameter optimization of an impulse response generation system, and the network optimization process is realized based on a Pythrch deep learning framework.

Secondly, the application of the impulse response signal generation system is mainly to perform deployment to be used on a pre-trained network, and the step mainly comprises the step of generating impulse response to a turbulent response signal of actual flutter test flight for modal parameter estimation in the flutter test flight.