CN114398827B - A method for constructing a virtual gyroscope based on deep learning - Google Patents

Abstract

According to the virtual gyroscope construction method based on deep learning, a magneto-resistance sensor of a high rotator is utilized to obtain geomagnetic vectors on a high rotator coordinate system; acquiring an acceleration vector on a high-rotation body coordinate system by using an accelerometer of the high-rotation body; inputting geomagnetic vectors at the current moment and the last moment on the high-rotation body coordinate system and acceleration vectors into a BILSTM network to obtain the high-rotation body posture change quaternary parameters; and obtaining the angular rate of the high rotator according to the conversion relation between the rotation vector of the high rotator and the gesture change quaternary parameter of the high rotator. The stable sensing of the high rotator angular rate after the gyroscope fails can be realized.

Description

A method for constructing a virtual gyroscope based on deep learning

Technical Field

The present invention belongs to the technical field of angular rate acquisition in complex, highly dynamic and strong interference environments, and in particular relates to a method for constructing a virtual gyroscope based on deep learning.

Background Art

Complex battlefield environments such as high dynamics and strong interference require high-speed rotating bodies to have the ability of real-time accurate positioning, intelligent autonomous decision-making and flexible motion control, and a single navigation system is difficult to meet the requirements. Therefore, a multi-sensor combined navigation strategy is often adopted, that is, multiple sensors are carried at the same time, and high-precision navigation is achieved through multi-modal combination. The motion perception of high-speed rotating bodies mainly uses gyroscopes, accelerometers and magnetoresistive sensors to measure angular velocity, acceleration and geomagnetic information. The high-speed spin of the high-speed rotating body (≥10r/s) will cause the gyroscope range to be saturated, resulting in the inability to accurately obtain the angular velocity information of the high-speed rotating body; the high overload of the high-speed rotating body (≥10000g) is easy to cause the gyroscope to fail during the working process, and it is impossible to ensure that the gyroscope can operate normally during the flight of the high-speed rotating body; the existence of elastic vibration will generate high-frequency vibration noise to the sensor strapped to the high-speed rotating body, thereby reducing the measurement accuracy of the angular velocity of the high-speed rotating body. Therefore, how to ensure the continuous, reliable, robust and available angular velocity measurement of high-speed rotating bodies in complex environments has become an urgent technical demand.

At present, there are mainly multiple accelerometer combination schemes, magnetoresistive sensor signal analysis schemes, and magnetoresistive sensor and accelerometer combination sensing schemes for high-speed rotating bodies. For the multiple accelerometer combination scheme, on the one hand, a specific geometric distribution is required, and the installation requirements of the accelerometer are high. On the other hand, for high-speed rotating bodies with too high spin rates such as shells, the complex cross-coupling terms in the angular rate solution process will cause the angular rate solution accuracy to decrease. The magnetoresistive sensor signal analysis scheme uses the magnetoresistive sensor strapped on the high-speed rotating body to make its output signal present a sine and cosine waveform as the high-speed rotating body rotates, and performs signal analysis on the waveform. This scheme can effectively and accurately obtain the roll angular rate, but it cannot obtain the pitch angular rate and yaw angular rate. When the gyroscope fails completely, it is still impossible to accurately obtain the full angular rate of the high-speed rotating body. For the magnetoresistive sensor and accelerometer combination sensing scheme, Bernerd et al. of the French-German Saint Louis Institute used this scheme to accurately obtain the rotation speed and acceleration information of the flying body for high-speed rotating flying bodies with a rotation speed of up to 170r/s. Fiot et al. also used a combination of acceleration and geomagnetic sensors to measure the motion information of a high-speed flying object. Although this method can obtain all the angular velocity information of a high-speed flying object, it relies on accurate kinematic equations. In practical applications, it is difficult to obtain the parameters in the kinematic equations, which leads to angular velocity estimation errors.

At the same time, the rapid development of information science has led to the widespread application of intelligent computing technologies such as machine learning, pattern recognition, and neural networks. Driven by these technologies, deep learning-assisted carrier motion perception has emerged as an important auxiliary means, and has penetrated into all aspects of motion perception measurement and is widely used in the field of INS/GNSS navigation and positioning. Most modern deep learning models are based on several basic structures of artificial neural networks, namely multilayer perceptron (MLP), convolutional neural network (CNN), recurrent neural network (RNN), and graph neural network (GNN). RNN can effectively process the time series output by INS. The internal unit structure of the long short-term memory network (LSTM) has been improved compared to RNN, adding forgetting and preservation mechanisms. It has also been applied to the field of INS/GNSS integrated navigation. In addition, two variants of LSTM: gated recurrent unit (GRU) and bidirectional long short-term memory network (BILSTM) have also become more widely used deep learning models. Therefore, this paper applies BILSTM to the field of high-rotating body angular rate prediction and designs a virtual gyroscope to provide stable and reliable angular rate information for high-rotating body navigation and guidance.

Summary of the invention

The present invention overcomes one of the deficiencies of the prior art and provides a method for constructing a virtual gyroscope based on deep learning, thereby achieving stable perception of high angular rates of a rotating body after gyroscope failure.

According to one aspect of the present disclosure, the present invention provides a method for constructing a virtual gyroscope based on deep learning, the method comprising:

Using the magnetoresistance sensor of the high-speed rotating body, the geomagnetic vector on the high-speed rotating body coordinate system is obtained;

Using an accelerometer of a high-speed rotating body, obtaining an acceleration vector on a coordinate system of the high-speed rotating body;

Inputting the geomagnetic vector at the current moment and the previous moment and the acceleration vector on the coordinate system of the high-speed rotating body into the BILSTM network to obtain the quaternion parameters of the posture change of the high-speed rotating body;

The angular velocity of the high-speed rotating body is obtained according to the conversion relationship between the rotation vector of the high-speed rotating body and the quaternary parameter of the posture change of the high-speed rotating body.

In a possible implementation, after acquiring the geomagnetic vector on the coordinate system of the high-speed rotating body by using the magnetoresistive sensor of the high-speed rotating body, the method includes:

According to the attitude conversion matrix of the high-rotating body coordinate system and the geographic coordinate system, a conversion relationship between the three-axis components of the geomagnetic vector of the high-rotating body in the high-rotating body coordinate system and the three-axis components in the geographic coordinate system is obtained;

According to the conversion relationship between the three-axis components of the geomagnetic vector in the high-rotating body coordinate system and the three-axis components in the geographic coordinate system and the chain rule, the relationship between the geomagnetic vector at the current moment and the geomagnetic vector at the previous moment in the high-rotating body coordinate system is obtained.

In a possible implementation manner, after acquiring the acceleration vector on the high-rotating body coordinate system by using the accelerometer of the high-rotating body, the method further includes:

Obtaining the conversion relationship between the component of the acceleration of the high-speed rotating body in the high-speed rotating body coordinate system and the component in the geographic coordinate system according to the velocity differential equation of the accelerometer;

According to the conversion relationship between the components of the acceleration in the high-rotating body coordinate system and the components in the geographic coordinate system and the chain rule, the relationship between the acceleration vector at the current moment and the acceleration vector at the previous moment in the high-rotating body coordinate system is obtained.

In a possible implementation, the geomagnetic vector at the current moment and the previous moment on the high-rotating body coordinate system and the acceleration vector are input into the BILSTM network to obtain the quaternion parameters of the high-rotating body posture change, including:

The geomagnetic vector conversion relationship between the current moment and the previous moment on the high-rotating body coordinate system and the acceleration vector conversion relationship between the current moment and the previous moment on the high-rotating body coordinate system are combined, and the quaternion parameters of the high-rotating body posture change are obtained according to the transfer matrix on the high-rotating body coordinate system.

In a possible implementation, obtaining the angular velocity of the high-rotating body according to the conversion relationship between the rotation vector of the high-rotating body and the quaternion parameter of the posture change of the high-rotating body includes:

Obtaining an angular increment of the high-speed rotating body according to a conversion relationship between a rotation vector of the high-speed rotating body and a quaternion parameter of a posture change of the high-speed rotating body;

The angular increment of the high-speed rotating body is extracted according to the angular increment single-sample extraction method to obtain the angular velocity of the high-speed rotating body.

The virtual gyroscope construction method based on deep learning of the present invention uses a magnetoresistive sensor of a high-speed rotating body to obtain a geomagnetic vector on a high-speed rotating body coordinate system; uses an accelerometer of a high-speed rotating body to obtain an acceleration vector on the high-speed rotating body coordinate system; inputs the geomagnetic vector of the current moment and the previous moment on the high-speed rotating body coordinate system, and the acceleration vector into a BILSTM network to obtain the quaternary parameters of the high-speed rotating body posture change; obtains the angular velocity of the high-speed rotating body according to the conversion relationship between the rotation vector of the high-speed rotating body and the quaternary parameters of the high-speed rotating body posture change. The stable perception of the angular velocity of the high-speed rotating body after the gyroscope fails can be achieved.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are used to provide a further understanding of the technical solution of the present application or the prior art, and constitute a part of the specification. Among them, the accompanying drawings expressing the embodiments of the present application are used together with the embodiments of the present application to explain the technical solution of the present application, but do not constitute a limitation on the technical solution of the present application.

FIG1 shows a flow chart of a method for constructing a virtual gyroscope based on deep learning according to an embodiment of the present disclosure;

FIG2 shows a block diagram of a virtual gyroscope based on deep learning according to another embodiment of the present disclosure;

FIG3 shows a schematic diagram of a simulation of a virtual gyroscope based on deep learning according to another embodiment of the present disclosure.

DETAILED DESCRIPTION

The following will describe the implementation methods of the present invention in detail with reference to the accompanying drawings and embodiments, so that the implementation process of how the present invention applies technical means to solve technical problems and achieve corresponding technical effects can be fully understood and implemented accordingly. The embodiments of the present application and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the protection scope of the present invention.

In addition, the steps shown in the flowchart of the accompanying drawings can be executed in a computer such as a set of computer executable instructions. Also, although a logical sequence is shown in the flowchart, in some cases, the steps shown or described can be performed in a sequence different from that here.

FIG1 shows a block diagram of a principle of constructing a virtual gyroscope based on deep learning according to another embodiment of the present disclosure.

As shown in Figure 1, before the high-spinning object is launched, it is necessary to calculate the trajectory elements of the high-spinning object based on the ammunition model, meteorological conditions and geographical location of the launch point, design a trajectory generator based on the trajectory elements of the high-spinning object, and use the data of the trajectory generator as a training set for solving the deep learning model, and then learn and train to obtain the learning network model of the trajectory elements of the high-spinning object.

The operation process of the trajectory generator is the inverse process of the strapdown inertial navigation calculation. The gyroscope data can be obtained according to the attitude differential equation; the magnetoresistive sensor data can be obtained according to the three-axis geomagnetic component of the local geomagnetic field combined with the attitude information; the accelerometer data can be obtained according to the speed and attitude combined with the specific force equation. That is, after the high-speed rotating body is launched, the output values of the accelerometer and magnetoresistive sensor strapped on the high-speed rotating body are used as the input of the deep learning model to obtain the attitude change quaternion and further extract the angular velocity of the high-speed rotating body.

FIG2 shows a flow chart of a method for constructing a virtual gyroscope based on deep learning according to an embodiment of the present disclosure. As shown in FIG2 , the method may include:

Step S1: using the magnetoresistive sensor of the high-speed rotating body to obtain the geomagnetic vector in the coordinate system of the high-speed rotating body. After obtaining the geomagnetic vector in the coordinate system of the high-speed rotating body, according to the attitude conversion matrix between the coordinate system of the high-speed rotating body and the geographic coordinate system, the conversion relationship between the three-axis components of the geomagnetic vector of the high-speed rotating body in the coordinate system of the high-speed rotating body and the three-axis components in the geographic coordinate system is obtained; according to the conversion relationship between the three-axis components of the geomagnetic vector in the coordinate system of the high-speed rotating body and the three-axis components in the geographic coordinate system and the chain rule, the relationship between the geomagnetic vector at the current moment and the geomagnetic vector at the previous moment in the coordinate system of the high-speed rotating body is obtained.

For example, as shown in FIG1 , during the flight of a high-speed spinning body, the magnetoresistive sensor strapped on the high-speed spinning body can measure the three-axis components of the geomagnetic vector in the high-speed spinning body coordinate system in real time. The three-axis components of the geomagnetic vector in the geographic coordinate system are Since the range of the high-speed rotating body is very small relative to the radius of the earth, it is assumed that the magnitude and direction of the geomagnetic vector remain unchanged in the geographic coordinate system, which can be obtained from the longitude and latitude of the launch site and the time according to the International Geomagnetic Reference Field (IGRF12) model.

According to Figure 1, the coordinate system transformation relationship can be obtained:

In the formula, is the attitude conversion matrix between the high-speed body coordinate system and the geographic coordinate system, and [θγψ] are the pitch angle, roll angle and yaw angle respectively.

For both time k and time k-1, equation (1) holds true, that is:

In the formula, ^Hn(k) = ^Hn(k-1) .

Since the high-speed rotating body has a short flight time and a short landing distance, it can be considered that the navigation coordinate system n does not change, then:

Where I _3×3 is a 3×3 identity matrix.

Further expansion of formula (2) according to the chain rule yields:

Step S2: using the accelerometer of the high-rotating body to obtain the acceleration vector on the high-rotating body coordinate system. After obtaining the acceleration vector on the high-rotating body coordinate system, the conversion relationship between the components of the acceleration of the high-rotating body on the high-rotating body coordinate system and the components on the geographic coordinate system is obtained according to the velocity differential equation of the accelerometer; according to the conversion relationship between the components of the acceleration on the high-rotating body coordinate system and the components on the geographic coordinate system and the chain rule, the relationship between the acceleration vector at the current moment and the acceleration vector at the previous moment on the high-rotating body coordinate system is obtained.

For example, according to the velocity differential equation of SINS:

In the formula, _Re is the average radius of the earth, _ωie is the angular rate of the earth's rotation; [L λ h] ^T is the position of the high rotation body, representing latitude, longitude and altitude respectively; are the high rotation speeds, representing the east, north and celestial speeds respectively; ^fb is the three-axis specific force measured by the accelerometer, and ^gn is the local gravitational acceleration.

By transforming formula (5), we can get:

In the formula,

For both time k and time k-1, equation (6) holds true, that is:

Further expansion of formula (7) according to the chain rule yields:

Combining equation (6) and equation (8), we can get:

in,

10 ^-5 order of magnitude.

Between [t _k-1 , t _k ], the speed of the high-speed rotating body will not change dramatically, so the angular velocity of the high-speed rotating body relative to the earth can be approximately equal within the time [t _k-1 , t _k ], that is, 10 ^-4 order of magnitude.

Between [t _k-1 , t _k ], use a constant to fit the high-speed rotating body, that is, v = a, then Right now Among them, Δv _ων =0.

Between [t _k-1 , t _k ], if a linear function is used to fit the high-speed rotating body, that is, v = at + b, then but Among them, the accelerometer sampling time is T _s = 0.001s,

Between [t _k-1 , t _k ], a quadratic function is used to fit the high-speed rotating body, that is, v = at ² + bt + c, then but

In summary, between [t _k-1 ,t _k ], Δf≈0, and equation (8) can be written as:

As shown in FIG3 , a simulation diagram of a virtual gyroscope based on deep learning is shown, where Δf and f ^b differ by three orders of magnitude. As can be seen from FIG3 , it can be shown that the approximation of formula (11) is scientific.

Step S3: inputting the geomagnetic vector at the current moment and the previous moment and the acceleration vector on the high-speed rotating body coordinate system into the BILSTM network to obtain the quaternion parameters of the high-speed rotating body posture change.

In one example, the conversion relationship between the geomagnetic vector at the current moment and the previous moment on the high-rotating body coordinate system and the conversion relationship between the acceleration vector at the current moment and the previous moment on the high-rotating body coordinate system are combined, and the quaternion parameters of the high-rotating body posture change are obtained by solving the transfer matrix on the high-rotating body coordinate system.

For example, combining equation (4) and equation (11), we can get:

In the formula,

[q ₀ q ₁ q ₂ q ₃ ] = q is the attitude change quaternion.

According to formula (12), when the gyroscope fails, the attitude change quaternion can be calculated based on the output vector of the magnetoresistive sensor and accelerometer at the previous moment and the output vector at the current moment, so as to continue to realize attitude update.

By transforming formula (12), we can get:

q＝f(H ^b(k) ,H ^b(k-1) ,f ^b(k) ,f ^b(k-1) ) Formula (13),

In the formula, f(*) represents a nonlinear mapping relationship, and its specific mapping relationship f(*) can be obtained through deep learning.

Step S4: obtaining the angular velocity of the high-speed rotating body according to the conversion relationship between the rotation vector of the high-speed rotating body and the quaternary parameters of the posture change of the high-speed rotating body.

In one example, the angular increment of the high-rotating body can be obtained according to the conversion relationship between the rotation vector of the high-rotating body and the quaternary parameters of the posture change of the high-rotating body; the angular increment of the high-rotating body is extracted according to the angular increment single sample extraction method to obtain the angular velocity of the high-rotating body.

For example, after obtaining the attitude change quaternion, the angle increment can be obtained according to the relationship between the rotation vector and the attitude change quaternion:

According to the single sample extraction method of angle increment, the angular rate can be obtained:

According to formula (15), the high rotation angular rate obtained by the combination of the accelerometer and the magnetoresistive sensor can be obtained. Compared with the introduction of the virtual gyroscope sensor (i.e., the data predictor), on the one hand, it provides new data observations for the combined attitude measurement, and on the other hand, it ensures the continuous operation of the navigation system when the gyroscope fails, ensuring the continuity, availability and adaptability of attitude fusion in complex environments.

The rotation vector method can be used to convert the gyroscope data into the attitude change quaternion. The specific process is the inverse process of equations (14) and (15). BILSTM is used to solve the model of equation (13), which is regarded as a regression problem. The current and previous data of the magnetoresistive sensor and accelerometer are input: [H ^b(k) H ^b(k-1) f ^b(k) f ^b(k-1) ] _12×1 ; the output is the attitude change quaternion: q _4×1 . Because BILSTM is composed of forward LSTM and backward LSTM, it can make full use of data in both directions and effectively improve the network accuracy.

Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims (1)

1. A method for constructing a virtual gyroscope based on deep learning, characterized in that the method comprises: Using the magnetoresistance sensor of the high-speed rotating body, the geomagnetic vector on the high-speed rotating body coordinate system is obtained; Using an accelerometer of a high-speed rotating body, obtaining an acceleration vector on a coordinate system of the high-speed rotating body; Inputting the geomagnetic vector at the current moment and the previous moment and the acceleration vector on the coordinate system of the high-speed rotating body into the BILSTM network to obtain the quaternion parameters of the posture change of the high-speed rotating body; Obtaining the angular velocity of the high-speed rotating body according to the conversion relationship between the rotation vector of the high-speed rotating body and the quaternary parameter of the posture change of the high-speed rotating body; After the magnetoresistive sensor of the high-speed rotating body is used to obtain the geomagnetic vector on the coordinate system of the high-speed rotating body, the method includes: According to the attitude conversion matrix of the high-rotating body coordinate system and the geographic coordinate system, a conversion relationship between the three-axis components of the geomagnetic vector of the high-rotating body in the high-rotating body coordinate system and the three-axis components in the geographic coordinate system is obtained; According to the conversion relationship between the three-axis components of the geomagnetic vector in the high-rotating body coordinate system and the three-axis components in the geographic coordinate system and the chain rule, the relationship between the geomagnetic vector at the current moment and the geomagnetic vector at the previous moment in the high-rotating body coordinate system is obtained; After the accelerometer of the high-speed rotating body is used to obtain the acceleration vector on the high-speed rotating body coordinate system, the method further includes: Obtaining the conversion relationship between the component of the acceleration of the high-speed rotating body in the high-speed rotating body coordinate system and the component in the geographic coordinate system according to the velocity differential equation of the accelerometer; According to the conversion relationship between the components of the acceleration in the high-rotating body coordinate system and the components in the geographic coordinate system and the chain rule, the relationship between the acceleration vector at the current moment and the acceleration vector at the previous moment in the high-rotating body coordinate system is obtained; The geomagnetic vector at the current moment and the previous moment on the coordinate system of the high-speed rotating body, and the acceleration vector are input into the BILSTM network to obtain the quaternary parameters of the posture change of the high-speed rotating body, including: Combine the geomagnetic vector conversion relationship between the current moment and the previous moment on the high-rotating body coordinate system, and the acceleration vector conversion relationship between the current moment and the previous moment on the high-rotating body coordinate system, and solve the high-rotating body attitude change quaternion parameters according to the transfer matrix on the high-rotating body coordinate system; The step of obtaining the angular velocity of the high-speed rotating body according to the conversion relationship between the rotation vector of the high-speed rotating body and the quaternion parameter of the posture change of the high-speed rotating body comprises: Obtaining an angular increment of the high-speed rotating body according to a conversion relationship between a rotation vector of the high-speed rotating body and a quaternion parameter of a posture change of the high-speed rotating body; The angular increment of the high-speed rotating body is extracted according to the angular increment single-sample extraction method to obtain the angular velocity of the high-speed rotating body.