A method for temperature drift compensation in hemispherical resonant gyroscopes

Abstract

A temperature drift compensation method for hemispherical resonant gyroscopes (CRGs) is proposed. Through on-orbit data analysis, a temperature estimation model for the CRG is designed, using parameters such as the heater switching state, heater on / off action time, cabin temperature, and bulkhead temperature as inputs from the thermal control subsystem. A neural network model is constructed to learn the nonlinear uncertain descent process of the CRG temperature under heater-off conditions. A compensation model between the CRG temperature and constant drift is established. This invention solves the on-orbit temperature compensation problem for hemispherical resonant gyroscopes, improves the measurement accuracy of the CRG, and eliminates the need for additional frequency measurement and temperature control circuits, thus saving costs.

Description

Technical Field

[0001] This invention relates to a temperature drift compensation method for hemispherical resonant gyroscopes, applicable to temperature drift compensation of hemispherical resonant gyroscopes in spacecraft control systems, and can also be extended to other similar inertial measurement units for temperature drift compensation. Background Technology

[0002] Hemispherical resonator gyroscopes are among the most commonly used inertial sensors on spacecraft and are crucial components for achieving spacecraft attitude control. During the operation of a spacecraft control system, the angular velocity information output by the hemispherical resonator is used to implement closed-loop angular velocity control or to integrate the spacecraft's attitude angles to determine the attitude angles. In applications requiring high control accuracy, temperature drift compensation of the hemispherical resonator is essential.

[0003] Temperature drift compensation methods for hemispherical resonant gyroscopes can generally be divided into two main categories:

[0004] (1) Method based on frequency measurement circuit. The relationship between the resonant frequency of a hemispherical resonant gyroscope and temperature is linear. Therefore, the resonant frequency of the hemispherical resonant gyroscope can be used as an indirect measurement of the ambient temperature of the hemispherical resonator to compensate for the scaling factor and zero bias of the gyroscope. However, it is necessary to design a frequency measurement circuit with certain resolution and sensitivity, which will increase the complexity and development cost of the single unit.

[0005] (2) Temperature control circuit-based method. By adding temperature measurement and control circuits, the temperature of the gyroscope can be controlled within a certain operating range or operating point, and compensation values ​​can be designed for specific operating ranges or operating points. However, this requires designing temperature measurement and control circuits, which will also increase the complexity and development cost of the unit. Summary of the Invention

[0006] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a temperature drift compensation method for hemispherical resonant gyroscopes. This method realizes temperature drift compensation for hemispherical resonant gyroscopes in the absence of direct and effective measurement data in orbit. It also effectively improves the accuracy of angular velocity measurement of hemispherical resonant gyroscopes when frequency measurement circuits and temperature control circuits cannot be introduced.

[0007] The technical solution of this invention:

[0008] A method for temperature drift compensation in a hemispherical resonant gyroscope includes the following steps:

[0009] 1) Establish a feedforward neural network for temperature estimation of hemispherical resonant gyroscopes;

[0010] 2) Use ground test data as sample data and target data to train the feedforward neural network;

[0011] 3) During in-orbit flight, determine the switching state of the hemispherical resonant gyroscope heater and record the corresponding on-time t. ON / Closing time t ON If the heater is off, the temperature estimate T′ of the hemispherical resonant gyroscope is obtained using the feedforward neural network described in step 2). HRG If the heater is on, the temperature estimate T′ of the hemispherical resonant gyroscope is obtained using the on-orbit estimation model. HRG ;

[0012] 4) Using the temperature estimate T′ of the hemispherical resonant gyroscope obtained in step 3), HRG The scaling factor and zero bias of the hemispherical resonant gyroscope are compensated sequentially to obtain the angular velocity pulse output by the hemispherical resonant gyroscope after compensation.

[0013] Preferably, the input to the feedforward neural network in step 1) includes: the star time t corresponding to the heater shutdown time. OFF The star time t corresponding to the sampling time, and the cabin plate temperature T inside the cabin where the hemispherical resonant gyroscope is located. deck and partition temperature T isolate The output of the feedforward neural network is an estimate of the temperature T' of the hemispherical resonant gyroscope. HRG .

[0014] Preferably, the sample data in step 2) is: the star time t corresponding to the heater shutdown time. OFF The current sampling time corresponds to the star time t, and the temperature T of the cabin plate inside the hemispherical resonant gyroscope is located. deck and partition temperature T isolate .

[0015] Preferably, the target data in step 2) is the real-time temperature measurement value T of the hemispherical resonant gyroscope. HRG .

[0016] Preferably, step 2) involves training the feedforward neural network, specifically as follows:

[0017] 21) Let the sampling period be T s For seconds, N sets of ground test data are collected as a sample set, N is not less than 10, each set has M sampling points, M≥100; the sampling data of each sampling point contains {T deck ,T isolaTe ,Δt,T HRG}, T HRG This represents the temperature of the hemispherical resonant gyroscope as measured in real time by a thermistor attached to its surface during ground testing. This thermistor does not accompany the hemispherical resonant gyroscope into orbit. Δt = tt OFF ;

[0018] 22) Select 70% of the data in the sample set to form the training dataset for training the feedforward neural network; then take 15% from each of the remaining sample sets for verification and testing of the training results of the feedforward neural network; the training ends when:

[0019]

[0020] Where i is the label of the data in the training dataset, and the value of i is [1, 0.7NM].

[0021] Preferably, the on-orbit estimation model described in step 3) is specifically:

[0022] T′ HRG =T ON +k·(tt ON )

[0023] Among them, T ON The threshold temperature for heater activation is indicated by the input from the preceding task; k represents the temperature gradient at which the heater is activated; t represents the star time corresponding to the current sampling moment; t ON This indicates the star time corresponding to the moment the heater is turned on.

[0024] Preferably, the temperature gradient k when the heater is turned on ranges from 0.005 to 0.1 °C / second.

[0025] Preferably, step 4) obtains the angular velocity pulse D output by the hemispherical resonant gyroscope after compensation. HRG Specifically:

[0026]

[0027] Among them, B2, B1 and B0 are zero-bias compensation coefficients, which are inherent parameters of the hemispherical resonant gyroscope. Their specific values ​​can be determined by polynomial fitting based on the laboratory test data of the hemispherical resonant gyroscope.

[0028] D″ HRG and D′ HRG A1 and A0 are the angular velocity pulses output by the hemispherical resonator before and after scaling factor compensation, respectively. A1 and A0 are the scaling factor compensation coefficients, which are inherent parameters of the hemispherical resonator. Their specific values ​​can be determined by polynomial fitting based on the laboratory test data of the hemispherical resonator.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] 1) This invention uses telemetry data from non-attitude and orbit control systems, such as the temperature of the entire satellite cabin panel and bulkhead, to improve the performance of individual control subsystems, making full use of the information flow advantages under the integrated electronic architecture.

[0031] 2) This invention utilizes neural networks to learn the nonlinear characteristics of temperature changes when the heater is off, thus solving the modeling difficulties caused by uncertainty.

[0032] 3) This invention does not rely on frequency measurement circuits and temperature control circuits, resulting in low cost and high reliability. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0034] To better describe the present invention, the present invention will be described in detail below with reference to schematic diagrams and examples.

[0035] like Figure 1 As shown, the present invention provides a method for temperature drift compensation in a hemispherical resonant gyroscope, comprising the following steps:

[0036] 1) Establish a feedforward neural network for temperature estimation of hemispherical resonant gyroscopes and determine the specific structure of the feedforward neural network;

[0037] Hemispherical resonant gyroscopes are typically installed in a compartment within the spacecraft's service module. This compartment is an enclosed space defined by the satellite's main body panels and bulkheads. When the heaters are turned off, the rate of temperature decrease of the hemispherical resonant gyroscope is primarily influenced by the temperatures of the main body panels and bulkheads at the gyroscope's installation location. Therefore, the input to the feedforward neural network includes the satellite time t corresponding to the heater shutdown moment. OFF The star time t corresponding to the sampling time, and the cabin plate temperature T inside the cabin where the hemispherical resonant gyroscope is located. deck and partition temperature T isolate The output of the feedforward neural network is an estimate of the temperature T' of the hemispherical resonant gyroscope. HRG .

[0038] 2) Use ground test data as sample data and target data to train the parameters of each layer of the feedforward neural network;

[0039] Step 2) The sample data used in the training is: the star time t corresponding to the heater shutdown time. OFF The current sampling time corresponds to the star time t, and the temperature T of the cabin plate inside the hemispherical resonant gyroscope is located. deck and partition temperature T isolate .

[0040] Step 2) The target data used in the training is the real-time temperature measurement value T of the hemispherical resonant gyroscope. HRGBy training a feedforward neural network, the network can learn the nonlinear uncertain relationship between the real-time measurements of satellite bulkhead temperature, satellite partition temperature, and hemispherical resonant gyroscope temperature when the heater is off. This enables the feedforward neural network to have an acceptable temperature estimation capability, and thus it is considered that a feedforward neural network that can be used in orbit has been obtained.

[0041] 3) During in-orbit flight, determine the switching state of the hemispherical resonant gyroscope heater and record the corresponding on-time t. ON / Closing time t ON If the heater is off, the temperature estimate T′ of the hemispherical resonant gyroscope is obtained using the feedforward neural network described in step 2). HRG If the heater is on, the temperature estimate T′ of the hemispherical resonant gyroscope is obtained using the on-orbit estimation model. HRG ;

[0042] The on-orbit estimation model mentioned in step 3) is specifically as follows:

[0043] T′ HRG =T ON +k·(tt ON )

[0044] Among them, T ON The threshold temperature for heater activation is indicated by the input from the preceding task; k represents the temperature gradient when the heater is activated, ranging from 0.005 to 0.1℃ / second; t represents the star time corresponding to the current sampling moment; t ON This indicates the star time corresponding to the moment the heater is turned on.

[0045] 4) Using the temperature estimate T′ of the hemispherical resonant gyroscope obtained in step 3), HRG As input parameters, the scaling factor and zero bias of the hemispherical resonant gyroscope are compensated sequentially to obtain the angular velocity pulse output by the hemispherical resonant gyroscope after compensation.

[0046] The feedforward neural network designed in step 1) is as follows:

[0047] 11) The input layer is designed with 3 neurons, namely x = (x1, x2, x3), and satisfies the following: x1 is taken as the temperature T of the entire satellite module of the compartment where the hemispherical resonant gyroscope is located. deck x2 is taken as the temperature T of the entire satellite bulkhead of the compartment where the hemispherical resonant gyroscope is located. isolate x3 is taken as the star time difference between the current sampling time and the heater shutdown time, i.e., x3 = Δt = tt OFF , let u i Let represent the output of the i-th neuron in the input layer, then:

[0048] u i=x i ,,i∈[1,3]

[0049] 12) Design a single hidden layer containing 10 neurons. The output h of the j-th neuron is... j for:

[0050]

[0051] Among them, f j (·) represents the transfer function of the hidden layer neurons, h j ω represents the output of the j-th neuron in the hidden layer. i,j b is the connection weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer, with a value range of [0,1]. j The j-th bias value can be any small real number, typically ranging from 0.01 to 1.

[0052] 13) The output layer is designed to be a single neuron, and the output y is the estimated temperature T′ of the hemispherical resonant gyroscope. HRG The calculation method is as follows:

[0053]

[0054] Where, ω j is the connection weight between the j-th neuron in the hidden layer and the neuron in the output layer, with a value range of [0,1]; b is the output layer bias value, which can be any small real number initially, typically ranging from 0.01 to 1.

[0055] Step 2) involves training the parameters of each layer of the feedforward neural network, as detailed below:

[0056] 21) Let the sampling period be T s For seconds, N sets (N not less than 10) of ground test data (each set containing M sampling points, M ≥ 100) are collected as a sample set. The sampling data of each sampling point contains {T deck ,T isolate ,Δt,T HRG}, T HRG This indicates the gyroscope temperature measured in real time by a thermistor attached to the surface of the hemispherical resonant gyroscope during ground testing. This thermistor does not follow the hemispherical resonant gyroscope into orbit. In one embodiment of the present invention, the sampling period is 8 seconds, and 36 groups are collected, with 1000 sampling points in each group.

[0057] 22) Select 70% of the data from the sample set to form the training dataset for training the neural network. Take another 15% from each of the remaining sample sets for verification and testing of the neural network training results. The neural network training method is a known technique. The training ends under the following conditions:

[0058]

[0059] That is, the sum of squared errors between the estimated and measured values ​​is less than 10. -5 i is the label of the data in the training dataset, and i takes the value [1, 0.7NM].

[0060] Step 4) of the hemispherical resonant gyroscope scaling factor compensation model and zero bias compensation model is as follows:

[0061] 41) The scaling factor compensation model for a hemispherical resonant gyroscope is as follows:

[0062]

[0063] Where: D″ HRG and D′ HRG A1 and A0 are the angular velocity pulses output by the hemispherical resonator before and after scaling factor compensation, respectively. A1 and A0 are the scaling factor compensation coefficients, which are inherent parameters of the hemispherical resonator. Their specific values ​​can be determined by polynomial fitting based on the laboratory test data of the hemispherical resonator.

[0064] 42) The zero-bias compensation model for a hemispherical resonant gyroscope is as follows:

[0065]

[0066] Where: D GRG B2, B1, and B0 are the angular velocity pulses output by the hemispherical resonant gyroscope after zero bias compensation. B2, B1, and B0 are the zero bias compensation coefficients, which are inherent parameters of the hemispherical resonant gyroscope. Their specific values ​​can be determined by polynomial fitting based on the laboratory test data of the hemispherical resonant gyroscope.

[0067] In one embodiment of the present invention, the corresponding values ​​of the hemispherical resonant gyroscope are A1 = 1.4552, A0 = 19991.31, B2 = 0.0000048, B1 = 0.00057, and B0 = 0.00061.

[0068] This invention addresses hemispherical resonator gyroscopes lacking frequency measurement and temperature control circuits, or those whose frequency measurement (temperature control) circuits malfunction during on-orbit testing. It comprehensively utilizes prior knowledge and the learning capabilities of feedforward neural networks to construct an on-orbit temperature estimation model for the hemispherical resonator gyroscope. By integrating on-orbit and laboratory data, it identifies model parameters and compensation parameters. This achieves temperature drift compensation for hemispherical resonator gyroscopes in situations where direct and effective on-orbit measurement data is unavailable, effectively improving the accuracy of angular velocity measurements when frequency and temperature control circuits cannot be introduced.

[0069] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any person skilled in the art can make possible variations and modifications to the technical solutions of the present invention using the disclosed methods and techniques without departing from the spirit and scope of the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the content of the technical solutions of the present invention, shall fall within the protection scope of the present invention. Where there is no conflict, the embodiments of this application and the technical features thereof can be combined with each other.

[0070] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for temperature drift compensation in a hemispherical resonant gyroscope, characterized in that, The steps include the following: 1) Establish a feedforward neural network for temperature estimation of hemispherical resonant gyroscopes; 2) Use ground test data as sample data and target data to train the feedforward neural network; 3) During in-orbit flight, determine the on / off state of the hemispherical resonator gyroscope heater and record the corresponding on and off times; if the heater is off, use the feedforward neural network described in step 2) to obtain the temperature estimate T′ of the hemispherical resonator gyroscope. HRG If the heater is on, the temperature estimate T′ of the hemispherical resonant gyroscope is obtained using the on-orbit estimation model. HRG ; 4) Using the temperature estimate T′ of the hemispherical resonant gyroscope obtained in step 3), HRG The scaling factor and zero bias of the hemispherical resonant gyroscope are compensated sequentially to obtain the angular velocity pulse output by the hemispherical resonant gyroscope after compensation. Step 1) The input to the feedforward neural network includes: the star time t corresponding to the heater shutdown time. OFF The star time t corresponding to the sampling time, and the cabin plate temperature T inside the cabin where the hemispherical resonant gyroscope is located. deck and partition temperature T isolate The output of the feedforward neural network is an estimate of the temperature T' of the hemispherical resonant gyroscope. HRG ; Step 2) The sample data is: the star time t corresponding to the heater shutdown time. OFF The current sampling time corresponds to the star time t, and the temperature T of the cabin plate inside the hemispherical resonant gyroscope is located. deck and partition temperature T isolate ; Step 4) obtains the angular velocity pulse D output by the hemispherical resonant gyroscope after compensation. HRG Specifically: Among them, B2, B1 and B0 are zero-bias compensation coefficients, which are inherent parameters of the hemispherical resonant gyroscope. Their specific values ​​can be determined by polynomial fitting based on the laboratory test data of the hemispherical resonant gyroscope. D″ HRG and D′ HRG A1 and A0 are the angular velocity pulses output by the hemispherical resonator before and after scaling factor compensation, respectively. A1 and A0 are the scaling factor compensation coefficients, which are inherent parameters of the hemispherical resonator. Their specific values ​​can be determined by polynomial fitting based on the laboratory test data of the hemispherical resonator.

2. The method for temperature drift compensation in a hemispherical resonant gyroscope according to claim 1, characterized in that, Step 2) The target data is the real-time temperature measurement value T of the hemispherical resonant gyroscope. HRG .

3. The method for temperature drift compensation in a hemispherical resonant gyroscope according to claim 2, characterized in that, Step 2) involves training the feedforward neural network, as detailed below: 21) Let the sampling period be T s For seconds, N sets of ground test data are collected as a sample set, N is not less than 10, each set has M sampling points, M≥100; the sampling data of each sampling point contains {T deck ,T isolate ,Δt,T HRG }, T HRG This represents the temperature of the hemispherical resonant gyroscope as measured in real time by a thermistor attached to its surface during ground testing. This thermistor does not accompany the hemispherical resonant gyroscope into orbit. Δt = tt OFF ; 22) Select 70% of the data in the sample set to form the training dataset for training the feedforward neural network; then take 15% from each of the remaining sample sets for verification and testing of the training results of the feedforward neural network; the training ends when: Where i is the label of the data in the training dataset, and the value of i is [1, 0.7NM].

4. A method for temperature drift compensation in a hemispherical resonant gyroscope according to any one of claims 1 to 3, characterized in that, The on-orbit estimation model mentioned in step 3) is specifically as follows: T′ HRG =T ON +k·(t-t ON ) Among them, T ON The threshold temperature for heater activation is indicated by the input from the preceding task; k represents the temperature gradient at which the heater is activated; t represents the star time corresponding to the current sampling moment; t ON This indicates the star time corresponding to the moment the heater is turned on.

5. A method for temperature drift compensation in a hemispherical resonant gyroscope according to claim 4, characterized in that, The temperature gradient k when the heater is turned on ranges from 0.005 to 0.1℃ / second.

Images