A method and system for automatically calibrating operating parameters of a MEMS gyroscope

Abstract

The present invention belongs to the technical field of MEMS gyroscopes, and discloses a method and system for automatically calibrating the operating parameters of a MEMS gyroscope. The present invention introduces a deep deterministic policy gradient reinforcement learning algorithm, which can automatically calibrate the PID parameters of the four closed loops of the phase-locked loop, the amplitude stabilization loop, the orthogonal suppression loop, and the force balance loop required by the gyroscope system, as well as the tuning voltage combination value of the driving mode and the sensitive mode, effectively suppress the frequency difference between the two modes, improve the signal-to-noise ratio of the gyroscope's effective signal, reduce the phase error generated by the gyroscope when performing signal demodulation, and improve the accuracy and reliability of angular velocity measurement. At the same time, the present invention constructs an experience pool, which can provide an efficient learning solution for gyroscope parameter calibration, and effectively solve the problems of low efficiency and lack of exploration in the process of reinforcement learning strategy learning. The present invention solves the problem that traditional measurement methods are difficult to handle multiple key parameters of gyroscopes, and does not require human intervention, is time-consuming and efficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of MEMS gyroscopes, and particularly relates to a method and system for automatically calibrating operating parameters of an MEMS gyroscope. Background Art

[0002] An MEMS (Micro Electro Mechanical Systems) gyroscope is a sensor element based on the principle of the Coriolis force effect, which can detect the external rotational angular velocity. Compared with large-sized gyroscopes such as laser gyroscopes, hemispherical gyroscopes, and fiber optic gyroscopes, MEMS gyroscopes have advantages such as small volume and low cost, and thus play an important role in some specific application scenarios, such as individual north-seeking instruments, small unmanned aerial vehicles, and underwater robots, especially in the short-range inertial navigation field of small devices, where they have been widely used.

[0003] Before an MEMS gyroscope is put into use, it generally needs to be tested and calibrated. Through testing and calibration, the error parameters of the MEMS gyroscope can be obtained, so as to perform compensation and calibration on it to improve the application accuracy. The calibration of an MEMS gyroscope can be divided into two main aspects: basic parameter calibration. When leaving the factory, the gyroscope needs to calibrate the basic parameters, which are currently mainly completed through manual intervention, such as testing the resonance frequency and quality factor, etc.; operating parameter calibration. In order to improve the detection accuracy of the gyroscope, it is also necessary to calibrate the parameters in its operating state. Traditional solutions usually rely on manual adjustment, including tuning the voltage to achieve mode matching, adjusting the drive phase amplitude to maximize the drive efficiency, adjusting the detection mode drive phase to suppress mode coupling, etc.

[0004] Currently, there have also emerged methods of automatic calibration. Automatic calibration of gyro parameters is a process of optimally calibrating the control parameters of an MEMS gyroscope measurement and control system by using automation technology. By applying different control parameters to the gyroscope and collecting the state data of the gyro measurement and control system under different drives, the optimal control parameters in the best output state of the gyro system are found. However, automatic calibration basically only targets individual sensors and cannot achieve batch automatic calibration. When calibrating the parameters of batch-produced MEMS gyroscopes, a large amount of manpower and material resources need to be invested, which hinders the application and industrial development of MEMS gyroscopes.

[0005] Although the current MEMS gyroscope parameter calibration methods can achieve the parameter calibration task, there are still the following disadvantages:

[0006] (1) Manual calibration of basic parameters is time-consuming and laborious: The traditional calibration process relies on manual operation, which not only increases the calibration time and cost but also affects the large-scale application of MEMS gyroscopes; manual intervention is not only cumbersome but also difficult to ensure high efficiency and high accuracy.

[0007] (2) The traditional calibration method has accuracy problems: Due to the phase error of the actual measurement and control process and the gyro itself, the signal will produce phase deviation during demodulation, which will affect the scale factor of angular rate detection. In addition, the tuning voltage of the gyro needs to be adjusted during the debugging process to suppress the frequency difference between the two modes. The traditional calibration method may cause the gyro mode to mismatch, thereby affecting the measurement accuracy and stability. At the same time, the existence of frequency difference will reduce the signal-to-noise ratio of the gyro's effective signal, which will seriously reduce the gyro's angular rate measurement accuracy.

[0008] Therefore, it is necessary to propose a method and system for automatically calibrating the operating parameters of a MEMS gyroscope to solve the above-mentioned technical problems existing in the prior art. Summary of the invention

[0009] The purpose of the present invention is to provide a method and system for automatic calibration of MEMS gyroscope operating parameters, introduce a deep deterministic policy gradient reinforcement learning algorithm, and automatically calibrate the motion control drive signal required by the gyroscope system without human intervention, so as to ensure that the gyroscope can operate within a reasonable working range.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] A method for automatically calibrating operating parameters of a MEMS gyroscope comprises the following steps:

[0012] Step 1. By scanning the resonant frequency, performing pull-in detection, and calculating the quality factor, the basic parameters of the gyroscope are measured to determine the initial values ​​of the gyroscope closed-loop control, including the initial frequency of the phase-locked loop, the initial locked phase of the phase-locked loop, the frequency output range of the phase-locked loop, and the initial locked amplitude of the amplitude stabilization loop;

[0013] Step 2. Set the initial PID parameters of the four closed loops of the phase-locked loop, amplitude stabilization loop, orthogonal inhibition loop, and force balance loop in the gyroscope state solution unit, as well as the initial value of the tuning voltage combination of the tuning voltage controller; output the environmental state parameters, namely the driving mode amplitude error, driving mode phase error, sensitive mode orthogonal coupling amplitude error, sensitive mode balance force amplitude error, and the driving mode and sensitive mode frequency difference; introduce the deep deterministic policy gradient DDPG reinforcement learning algorithm, input the environmental state parameters to the state policy network in the intelligent agent for analysis, so as to output the action control drive signal, namely the PID parameters of the four closed loops of the gyroscope's phase-locked loop, amplitude stabilization loop, orthogonal inhibition loop, and force balance loop, as well as the tuning voltage combination value of the driving mode and the sensitive mode; the action evaluation network in the intelligent agent evaluates the performance value of the action control drive signal, and at the same time inputs the action control drive signal into the gyroscope state solution unit to obtain new environmental state parameters;

[0014] Finally, the error is calculated based on the performance value of the previous stage and the current stage, and the action is updated and coordinated through the state strategy network and the action evaluation network to achieve automatic calibration of the MEMS gyroscope operating parameters.

[0015] A MEMS gyroscope operating parameter automatic calibration system includes the following modules:

[0016] The gyroscope basic parameter calculation module is used to scan the resonant frequency, perform pull-in detection, and calculate the quality factor to realize the gyroscope basic parameter calculation to determine the initial value of the gyroscope closed-loop control, including the initial frequency of the phase-locked loop, the initial locked phase of the phase-locked loop, the frequency output range of the phase-locked loop, and the initial locked amplitude of the amplitude stabilization loop;

[0017] The gyroscope operating parameter automatic calibration module is used to set the initial PID parameters of the four closed loops of the phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop in the gyroscope state solution unit, as well as the initial value of the tuning voltage combination of the tuning voltage controller, so as to output the environmental state parameters; and introduce the deep deterministic policy gradient DDPG reinforcement learning algorithm to input the environmental state parameters into the state policy network in the intelligent agent for analysis to output the action control drive signal; the action evaluation network in the intelligent agent evaluates the performance value of the action control drive signal, and at the same time inputs the action control drive signal into the gyroscope state solution unit to obtain new environmental state parameters; finally, the error is calculated based on the performance value of the previous stage and the current stage, and the action is updated and coordinated through the state policy network and the action evaluation network to realize the automatic calibration of the MEMS gyroscope operating parameters.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] As described above, the present invention relates to a method and system for automatically calibrating the operating parameters of a MEMS gyroscope, and introduces a deep deterministic policy gradient reinforcement learning algorithm based on the debugging experience of gyroscope experts. Without human intervention, the PID parameters of the phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop of the gyroscope required by the gyroscope system can be automatically calibrated, as well as the tuning voltage combination of the driving mode and the sensitive mode, effectively suppressing the frequency difference between the two modes, improving the signal-to-noise ratio of the gyroscope's effective signal, and reducing the phase error generated by the gyroscope when performing signal demodulation, thereby significantly improving the accuracy and reliability of angular velocity measurement. At the same time, by constructing an experience pool, an efficient learning solution can be provided for gyroscope parameter calibration, effectively solving the problems of low efficiency and insufficient exploration in the process of reinforcement learning strategy learning. The present invention can not only improve the efficiency of basic parameter calculation, but also solve the problem that the traditional manual measurement method is difficult to effectively process multiple key parameters of the gyroscope; while realizing the correction of gyroscope system errors and improving its measurement accuracy, it solves the problem that the traditional parameter calibration process requires a lot of manual intervention, is time-consuming and inefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required to be used in the embodiments are briefly introduced below.

[0021] Figure 1 A flowchart of a method for automatically calibrating operating parameters of a MEMS gyroscope in an embodiment;

[0022] Figure 2 is a flow chart of a resonant frequency scanning unit in an embodiment;

[0023] Figure 3 is a flow chart of the suction detection unit in the embodiment;

[0024] Figure 4 is a flow chart of a quality factor calculation unit in an embodiment;

[0025] Figure 5 The structure diagram of the interaction between the gyroscope state solving unit and the intelligent agent in the embodiment is shown in FIG. Figure 1 ;

[0026] Figure 6 The structure diagram of the interaction between the gyroscope state solving unit and the intelligent agent in the embodiment is shown in FIG. Figure 2 ;

[0027] Figure 7 A schematic diagram of the structure of an intelligent agent in an embodiment;

[0028] Figure 8 Schematic diagram of the structure of the experience pool in the embodiment. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0030] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0031] Example 1

[0032] like Figure 1As shown, this embodiment describes a method for automatic calibration of MEMS gyroscope operating parameters, and introduces a deep deterministic policy gradient (DDPG) reinforcement learning algorithm based on the debugging experience of gyroscope experts. Without human intervention, the PID parameters of the four closed loops of the gyroscope's phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop, as well as the tuning voltage combination of the driving mode and the sensitive mode required by the gyroscope system can be automatically calibrated, effectively suppressing the frequency difference between the two modes, improving the signal-to-noise ratio of the gyroscope's effective signal, and reducing the phase error generated by the gyroscope during signal demodulation, thereby significantly improving the accuracy and reliability of angular velocity measurement.

[0033] In this embodiment, a method for automatically calibrating operating parameters of a MEMS gyroscope includes the following steps:

[0034] Step 1. Calculate the basic parameters of the gyroscope by scanning the resonant frequency, performing pull-in detection, and calculating the quality factor to determine the initial values ​​of the gyroscope closed-loop control, including the initial frequency of the phase-locked loop, the initial locked phase of the phase-locked loop, the frequency output range of the phase-locked loop, and the initial locked amplitude of the amplitude stabilization loop, to ensure the normal and stable operation of the gyroscope.

[0035] Step 1.1. Scan the resonant frequency and perform the gyro phase-locked loop (PLL) phase-lock operation to ensure that the gyroscope can stably track its inherent resonant frequency during operation;

[0036] First, set the initial resonant frequency range and set the scanning step length. Then, within the initial resonant frequency range, perform automatic frequency scanning according to the scanning step length to find the resonance phenomenon. If the resonance phenomenon is detected, record the resonant frequency values ​​of the two modes of the gyroscope, calculate the frequency difference between the driving mode and the sensitive mode, and record the phase value at the resonant frequency. If the resonance phenomenon is not detected within the initial resonant frequency range, adjust the initial resonant frequency range, rescan, until the resonance phenomenon is detected, and record the data.

[0037] Step 1.2. After completing the gyro phase-locked loop adjustment, perform a pull-in test;

[0038] First, turn on the amplitude stabilization loop and lock the modal amplitude of the gyro to a stable and appropriate value by adjusting the amplitude control loop; then, design the tuning voltage step length, starting from the initial tuning voltage, and increase the tuning voltage step length each time, monitor and record the reading of the pull-in current detector when the gyro system is in the non-pull-in state, and when the current reaches the uA level, record the tuning voltage value at this time and set it as the maximum value of the subsequent tuning voltage output;

[0039] Step 1.3. Calculation of quality factor;

[0040] After resonant frequency scanning and pull-in detection, the gyro system is allowed to operate stably near the resonant frequency. At a certain moment, the excitation signal amplitude of the driving mode is set to 0, and the driving amplitude is observed and recorded to decay freely to the initial amplitude. The time constant t is recorded as the time constant; then the quality factor Q is calculated, the formula is , where is the resonant frequency of the driven mode.

[0041] This embodiment introduces an intelligent calculation method for gyroscope basic parameters to achieve automatic calculation of gyroscope resonant frequency, attraction state, and quality factor basic parameters, and automatically scans out the optimal basic parameter combination within a specified parameter range. Compared with the traditional manual measurement method with human intervention, the optimal basic parameter combination can be obtained in a larger range, faster time, and more accurately.

[0042] The calculation of the basic parameters of resonant frequency, engagement state, and quality factor here is to better calibrate the PID parameters of the optimal phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop required for the gyro system, as well as the tuning voltage combination of the driving mode and the sensitive mode, effectively suppress the frequency difference between the two modes, improve the signal-to-noise ratio of the gyro's effective signal, and reduce the phase error generated by the gyro during signal demodulation, thereby significantly improving the accuracy and reliability of angular velocity measurement, providing a stable gyro operating environment, and ensuring that the gyro can operate within a reasonable working range.

[0043] Step 2. First, set the initial PID parameters of the four closed loops of the phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop in the gyroscope state solver, as well as the initial value of the tuning voltage combination of the tuning voltage controller, to output the environmental state parameters, namely, the driving mode amplitude error, the driving mode phase error, the sensitive mode orthogonal coupling amplitude error, the sensitive mode balance force amplitude error, and the driving mode and sensitive mode frequency difference;

[0044] In this embodiment, the initial PID parameters of the four closed loops, namely, the phase-locked loop, the amplitude stabilization loop, the orthogonal suppression loop, and the force balance loop, are set to fixed values ​​according to the system and manual debugging experience, P is 30, I is 0.1, and D is 0; the initial value of the tuning voltage combination of the tuning voltage controller is set to 0;

[0045] Then, the deep deterministic policy gradient DDPG reinforcement learning algorithm is introduced to input the environmental state parameters into the state policy network in the intelligent agent for analysis to output the action control drive signal, namely the PID parameters of the four closed loops of the gyroscope's phase-locked loop, amplitude stabilization loop, orthogonal inhibition loop, and force balance loop, as well as the tuning voltage combination value of the driving mode and the sensitive mode; the action evaluation network in the intelligent agent evaluates the performance value of the action control drive signal, and at the same time inputs the action control drive signal into the gyroscope state solution unit to obtain new environmental state parameters;

[0046] Finally, the error is calculated based on the performance value of the previous stage and the current stage, and the action is updated and coordinated through the state strategy network and the action evaluation network to achieve automatic calibration of the MEMS gyroscope operating parameters.

[0047] The gyroscope state solver includes an orthogonal suppression loop, a force balance loop, a phase-locked loop, an amplitude stabilization loop, and a tuning voltage controller. Figure 5 , Figure 6 As shown;

[0048] The orthogonal inhibition loop receives the PID action parameters of the orthogonal inhibition loop output by the intelligent agent, performs PID control on the error between the orthogonal coupling output amplitude of the gyro sensitive mode and the target amplitude, and reduces the orthogonal coupling output error of the sensitive mode, wherein the target amplitude of the orthogonal coupling of the gyro sensitive mode is 0; at the same time, the orthogonal inhibition loop inputs the updated amplitude error of the orthogonal coupling of the gyro sensitive mode as a new environmental state parameter back to the intelligent agent;

[0049] The force balance loop receives the force balance PID action parameters output by the intelligent agent, and performs PID control on the error between the gyro sensitive modal force balance output amplitude and the target amplitude, wherein the gyro sensitive modal force balance target amplitude is 0; at the same time, the force balance loop inputs the updated gyro sensitive modal force balance amplitude error as a new environmental state parameter back to the intelligent agent;

[0050] The phase-locked loop receives the PID action parameters of the phase-locked loop output by the intelligent agent, and performs PID control on the phase error between the steady-state phase of the gyro phase-locked loop drive mode output and the target phase, wherein the gyro drive mode target phase is the phase size at the drive mode resonance peak obtained by frequency sweeping; at the same time, the phase-locked loop inputs the updated gyro drive mode phase error back to the intelligent agent as a new environmental state parameter;

[0051] The amplitude stabilization loop receives the PID action parameters of the amplitude stabilization loop output by the intelligent agent, and performs PID control on the steady-state error of the driving mode amplitude between the output amplitude of the gyro amplitude stabilization loop driving mode and the target amplitude, wherein the target amplitude of the gyro phase-locked loop driving mode is the resonance peak amplitude obtained during the frequency sweep; at the same time, the amplitude stabilization loop inputs the updated gyro driving mode amplitude error as a new environmental state parameter back to the intelligent agent;

[0052] A tuning voltage controller receives the tuning voltage combination of the driving mode and the sensitive mode output by the intelligent agent, and performs tuning control on the gyro driving mode and the sensitive mode; at the same time, the tuning voltage controller inputs the updated frequency difference between the gyro driving mode and the sensitive mode back to the intelligent agent as a new environmental state parameter;

[0053] like Figure 5 , Figure 6As shown, the phase-locked loop and the amplitude stabilization loop are both aimed at the driving mode. The phase-locked loop determines the driving frequency, that is, the phase, and the amplitude stabilization loop determines the driving amplitude. The results determined by the two act simultaneously to adjust the excitation signal of the driving mode; the orthogonal suppression loop and the force balance loop respectively determine the cosine amplitude and sin amplitude of the sensitive mode drive, and the results determined by the two act simultaneously to adjust the excitation signal of the sensitive mode.

[0054] In this embodiment, the DDPG agent includes a state policy network and an action evaluation network. Figure 7 As shown;

[0055] The state strategy network receives the driving mode amplitude error, driving mode phase error, sensitive mode orthogonal coupling amplitude error, sensitive mode balance force amplitude error and driving mode and sensitive mode frequency difference output by the phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, force balance loop and tuning voltage controller in the gyroscope state solution unit as environmental state parameters, and transmits them to the state strategy network. The weight matrix The input environmental state parameters are analyzed and processed to obtain the gyro motion control drive signal, that is, the PID parameters of the four closed loops of the gyro's phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop, as well as the tuning voltage combination value of the driving mode and the sensitive mode; this motion control drive signal, that is, the action information, is not only used as the input of the gyro state solver unit, but also as the current action evaluation network The action evaluation network calculates the performance value of the action based on the input action information. , that is, the value estimate of taking this action in the current state;

[0056] In the next stage, the gyroscope state solver receives the motion control drive signal output by the state strategy network. The gyroscope state solver processes the motion control drive signal received by the gyroscope system phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, force balance loop and tuning voltage controller to output the updated drive mode amplitude error, drive mode phase error, sensitive mode orthogonal coupling amplitude error, sensitive mode balance force amplitude error and drive mode and sensitive mode frequency difference of the gyroscope system, which are input back to the state strategy network in the DDPG agent as new environmental state parameters. ;

[0057] The agent uses the state strategy network based on the newly received state information. Calculate new motion control drive signals and the motion evaluation network Evaluating the performance value of new motion control actuation signals ; By comparing the performance value of the two time points before and after and , calculate the error, and use this error to reversely update the current state policy network and the current action evaluation network The weight of .

[0058] In this implementation, the intelligent agent also includes an experience pool, which includes expert experience units and historical experience units, such as Figure 8 shown.

[0059] The expert experience unit is used to collect and organize various experience data in the process of manual calibration of gyroscope parameters, including parameter settings under different working environments and different calibration conditions; the historical experience unit is used to collect historical process data generated by the intelligent agent and the gyroscope state solver during the learning and use process.

[0060] This embodiment requires the gyro expert to use the experience data of the gyro operating parameter calibration as the input of the expert experience unit to complete the record of the expert calibration experience; the specific process of the data collected by the historical experience unit is as follows:

[0061] The state policy network takes the environment state parameters it inputs Perform analytical processing to obtain the gyro motion control drive signal , namely the PID parameters of the four closed loops of the gyroscope's phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop, as well as the tuning voltage combination value of the driving mode and the sensitive mode; the gyroscope state solver receives the gyroscope action control drive signal output by the intelligent agent The gyroscope state solver controls the input action drive signal Processing to output updated environmental state parameters of the gyro system , namely the driving mode amplitude error, driving mode phase error, sensitive mode orthogonal coupling amplitude error, sensitive mode balance force amplitude error and driving mode and sensitive mode frequency difference, the new environmental state parameters are input back to the state strategy network in the DDPG agent; at the same time, the gyroscope state solver unit is based on the updated environmental state parameters of the gyroscope system Get reward results , and will reward Return the agent; finally, the environment state parameters , motion control drive signal , Environmental status parameters and rewards Multidimensional array Deposit into experience pool As historical experience.

[0062] Introducing the experience pool After that, the execution process of the agent is as follows:

[0063] Step 2.1. Initialize the experience pool , initialize random noise , get the initialization environment status parameters ;

[0064] Step 2.2. State Policy Network and action evaluation network Randomly initialize the network weight matrix and ;

[0065] From the experience pool The initial experience is obtained from the expert experience in the state policy network The environmental state parameters input based on the initial experience obtained Perform analysis to obtain motion control drive signals , the formula is as follows,

[0066] ;

[0067] in, represents the environmental state parameter, Representing the state policy network The weight matrix Environmental status parameters Analyze and process the obtained motion control drive signal;

[0068] Action Assessment Network For motion control drive signal Perform performance value assessment to obtain performance value ;

[0069] Step 2.3. The gyroscope state solver receives the motion control drive signal , and control the driving signal for the action Processing to output updated environmental state parameters of the gyro system , environmental state parameters Input back to the state policy network in the agent At the same time, the gyroscope state solver is based on the environmental state parameters Get Rewards , and will reward Return agent;

[0070] Step 2.4. Set the environment status parameters , motion control drive signal , Environmental status parameters and rewards Multidimensional array Deposit into experience pool Historical experience unit in

[0071] Step 2.5. Use the stateful policy network and action evaluation network The network weight matrix and Stateful Policy Network and action evaluation network The weight matrix and Initialize, that is , ;

[0072] Step 2.6. From the experience pool A multidimensional array of randomly sampled n batches of values ​​from two mixed experiences: expert experience and historical experience , and input the sampled n batches of multidimensional arrays into the state policy network and action evaluation network , to calculate the performance value of the next stage of n batches , thereby obtaining n batches of rewards in the current stage and the performance value of the next stage in the weight The cumulative result under the formula is as follows:

[0073] ;

[0074] In the formula, is the weight distribution value between the current stage and the next stage, Indicates the current stage Rewards With the next stage Performance value In weight The accumulation of the following;

[0075] Step 2.7. State Policy Network Based on the obtained mixed experience, the environmental state parameters Perform analysis to obtain motion control drive signals ;

[0076] ;

[0077] in, represents new random noise;

[0078] The action control drive signal Follow step 2.3 to obtain the updated environmental state parameters of the gyroscope state solver. At the same time, the gyroscope state solver is based on the environmental state parameters Get reward results , and will reward Return agent;

[0079] Finally, the environmental state parameter , motion control drive signal , Environmental status parameters and rewards Multidimensional array Deposit into experience pool Historical experience unit in

[0080] Step 2.8. Calculate the minimum loss function L, the formula is as follows:

[0081] ;

[0082] Where n represents the number of samples from the experience pool in step 2.6;

[0083] If the minimized loss function L is lower than the preset threshold, the calculation is stopped; otherwise, return to step 2.5 for the next stage of adjustment, and continuously update the weight parameters of the state policy network and the action evaluation network until the minimized loss function L is lower than the preset threshold.

[0084] In this embodiment, the core of reinforcement learning is to optimize the strategy through repeated trial and error. The agent selects actions, evaluates and updates the strategy at each time step, gradually approaching the optimal strategy. In this process, the gyroscope state solver acts as a bridge between the current state and the next state. It converts the decision results of the agent into new state information for the next strategy decision. Finally, the agent forms an efficient and robust control system through continuous learning and adjustment, which can cope with various complex situations in the MEMS gyroscope system.

[0085] The gyroscope state solving unit interacts with the intelligent agent. After completing phase locking, amplitude stabilization, and pull-in detection to put the gyroscope environment in a stable working state, the output end of the gyroscope state solving unit is connected to the input end of the state policy network. According to the environmental state parameters output by the gyroscope state solving unit, the state policy network outputs the corresponding action control drive signal; at the same time, the output end of the state policy network is connected to the input end of the gyroscope state solving unit. According to the action control drive signal output by the state policy network, the gyroscope state solving unit adjusts the state of the gyroscope to obtain the environmental state parameters of the next stage, and obtains the reward result based on the updated environmental state parameters of the gyroscope system (the environmental state parameters of the next stage), and returns the reward to the intelligent agent; the output end of the state policy network is connected to the input end of the action evaluation network, and the action evaluation network completes the value evaluation of the action solved by the state policy network, thereby guiding the improvement of the state policy network.

[0086] The state policy network and the action evaluation network form an agent. The behavior of the agent is based on the interaction between the outputs of its state policy network and action evaluation network and the gyro environment to learn a policy to maximize the cumulative reward.

[0087] Embodiment 2

[0088] As Figures 2 to 8 shown, Embodiment 2 of the present invention describes an automatic calibration system for the operating parameters of a MEMS gyroscope. This system is based on the same inventive concept as the automatic calibration method for the operating parameters of a MEMS gyroscope described in Embodiment 1 above.

[0089] An automatic calibration system for the operating parameters of a MEMS gyroscope includes the following modules:

[0090] A gyroscope basic parameter measurement module, which is used to scan the resonance frequency, perform pull-in detection, and calculate the quality factor to realize the measurement of the gyroscope basic parameters to determine the initial values for the closed-loop control of the gyroscope, including the initial frequency of the phase-locked loop, the initial locked phase of the phase-locked loop, the frequency output range of the phase-locked loop, and the initial locked amplitude of the amplitude stabilization loop;

[0091] An automatic calibration module for the operating parameters of the gyroscope, which is used to set the initial PID parameters of the four closed loops of the phase-locked loop, amplitude stabilization loop, quadrature rejection loop, and force balance loop in the gyroscope state solution unit, as well as the initial value of the tuning voltage combination of the tuning voltage controller to output environmental state parameters; and introduce the deep deterministic policy gradient (DDPG) reinforcement learning algorithm, input the environmental state parameters into the state policy network in the agent for analysis to output an action control drive signal; the action evaluation network in the agent performs a performance value evaluation on the action control drive signal, and at the same time inputs the action control drive signal into the gyroscope state solution unit to obtain new environmental state parameters; finally, calculate the error based on the performance values at two time points of the previous stage and the current stage, and update and allocate actions through the state policy network and the action evaluation network to realize the automatic calibration of the operating parameters of the MEMS gyroscope.

[0092] The embodiments of the present invention are only used to illustrate the technical solutions of the present invention and not to limit them. For those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically calibrating operating parameters of a MEMS gyroscope, characterized in that: The steps include: Step 1. By scanning the resonant frequency, performing pull-in detection, and calculating the quality factor, the basic parameters of the gyroscope are measured to determine the initial values ​​of the gyroscope closed-loop control, including the initial frequency of the phase-locked loop, the initial locked phase of the phase-locked loop, the frequency output range of the phase-locked loop, and the initial locked amplitude of the amplitude stabilization loop; Step 2. Set the initial PID parameters of the four closed loops of the phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop in the gyroscope state solver, as well as the initial value of the tuning voltage combination of the tuning voltage controller; Output environmental state parameters, namely, driving mode amplitude error, driving mode phase error, sensitive mode orthogonal coupling amplitude error, sensitive mode balance force amplitude error, and driving mode and sensitive mode frequency difference; introduce deep deterministic policy gradient DDPG reinforcement learning algorithm, input environmental state parameters to the state policy network in the intelligent agent for analysis, so as to output action control drive signals, namely, the PID parameters of the four closed loops of the gyroscope's phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop, as well as the tuning voltage combination value of the driving mode and the sensitive mode; the action evaluation network in the intelligent agent evaluates the performance value of the action control drive signal, and at the same time inputs the action control drive signal into the gyroscope state solution unit to obtain new environmental state parameters; Finally, the error is calculated based on the performance value of the previous stage and the current stage, and the action is updated and coordinated through the state strategy network and the action evaluation network to achieve automatic calibration of the MEMS gyroscope operating parameters.

2. The method for automatically calibrating the operating parameters of a MEMS gyroscope according to claim 1, characterized in that: The intelligent agent includes a state policy network and an action evaluation network; The state strategy network receives the environmental state parameters output by the gyroscope state solver and passes the weight matrix θ of the state strategy network V at the current stage V The input environmental state parameters are analyzed to obtain the motion control drive signal; this motion control drive signal is not only used as the input of the gyroscope state solver unit, but also as the input of the current stage motion evaluation network u for real-time evaluation; At the current stage, the action evaluation network u calculates the performance value Value corresponding to the action based on the input action control drive signal; In the next stage, the gyroscope state solving unit receives the action control driving signal output by the state strategy network V of the current stage, processes the action control driving signal, and outputs the updated environment state parameters of the gyroscope; the updated environment state parameters of the gyroscope are input back to the next stage state strategy network V' in the intelligent agent; The agent calculates the new action control driving signal through the next stage state strategy network V', and the next stage action evaluation network u' evaluates the performance value Value' of the new action; By comparing the performance values ​​Value and Value′ at two time points before and after, the error is calculated, and this error is used to reversely update the weights of the current stage state strategy network V and the current stage action evaluation network u.

3. The method for automatically calibrating the operating parameters of a MEMS gyroscope according to claim 1, characterized in that: The intelligent agent also includes an experience pool, which includes expert experience units and historical experience units; The expert experience unit is used to collect and organize various experience data in the process of manual calibration of gyroscope parameters, including parameter settings under different working environments and different calibration conditions; the historical experience unit is used to collect historical process data generated by the intelligent agent and the gyroscope state solver during the learning and use process.

4. The method for automatically calibrating the operating parameters of a MEMS gyroscope according to claim 3, characterized in that: In step 2, the execution process of the agent after introducing the experience pool is as follows: Step 2.

1. Initialize the experience pool R and random noise N1, and obtain the initial environment state parameter s1; Step 2.

2. Randomly initialize the network weight matrix θ of the current stage state policy network V and the current stage action evaluation network u V and θ u ; Initial experience is obtained from the expert experience in R. V parses and processes its input s1 based on the obtained initial experience to obtain the action control drive signal a1; u evaluates the performance value of a1 to obtain the performance value Value; Step 2.

3. The gyroscope state solving unit receives a1 and processes a1 to output the updated environmental state parameter s2 of the gyroscope system, and s2 is input back to the next stage state strategy network V' in the intelligent agent; At the same time, the gyroscope state solver obtains the reward r1 based on s2 and returns the reward r1 to the agent; Step 2.

4. Store s1, a1, s2 and r1 into the historical experience unit in R as a multidimensional array (s1, a1, r1, s2); Step 2.

5. Use θ V and θ u The weight matrix θ for V' and u' V' and θ u' Initialize; Step 2.

6. Randomly sample a multidimensional array of n batches of values ​​from the mixed experience of R's expert experience and historical experience (s i ,a i ,r i ,s i+1 ), and input the sampled n batches of multidimensional arrays into V' and u' to calculate the performance value Value' of the next stage of n batches, and obtain the cumulative result of the rewards of the current stage and the performance value of the next stage under the weight γ of n batches; Step 2.

7. V' analyzes s2 based on the obtained mixed experience to obtain the action control drive signal a2; According to step 2.3, a2 is used to obtain the updated environmental state parameter s3 of the gyroscope state solver; at the same time, the gyroscope state solver obtains the reward result r2 based on s3, and returns the reward r2 to the agent; Finally, s2, a2, s3, and r2 are stored in the historical experience unit in R as a multidimensional array (s2, a2, r2, s3); Step 2.

8. Calculate the minimum loss function L; If L is lower than the preset threshold, the calculation is stopped; otherwise, return to step 2.5 for the next stage of adjustment, and continuously update the weight parameters of the state policy network and the action evaluation network until the minimized loss function L is lower than the preset threshold.

5. A method for automatically calibrating operating parameters of a MEMS gyroscope according to claim 4, characterized in that: In step 2.2, the calculation formula of the action control drive signal a1 is as follows: a1=V(s1|θ v )+N1; Among them, a1 represents the weight matrix θ of V V Analyze s1 to obtain the action control drive signal; In step 2.6, the calculation formula for the cumulative result is as follows: y i =r i +γu'(s i+1 ,V'(s i+1 |θ V' )|θ u' ); In the formula, y i Indicates the current stage i Rewards i With the next stage i+1 Performance Value′ i+1 Accumulation under weight γ; γ is the weight distribution value between the current stage and the next stage; In step 2.7, the calculation formula of the action control drive signal a2 is as follows: <h2 style=";text-align:left;direction:ltr">a2 = V′(s2|θ<h2 style=";text-align:left;direction:ltr"> V′ <h2 style=";text-align:left;direction:ltr"> )+N2; Among them, N2 represents new random noise; In step 2.8, the formula for minimizing the loss function L is as follows: Where n represents the number of samples sampled from the experience pool in step 2.

6.

6. A method for automatically calibrating operating parameters of a MEMS gyroscope according to claim 1, characterized in that: The specific process of resonant frequency scanning in step 1 is as follows: First, set the initial resonant frequency range and the scanning step size; Then, within the initial resonant frequency range, the frequency is automatically scanned according to the scanning step length to find the resonance phenomenon; If resonance is detected, the resonant frequency values ​​of the two modes of the gyroscope are recorded, the frequency difference between the driving mode and the sensitive mode is calculated, and the phase value at the resonant frequency is recorded; If no resonance phenomenon is detected within the initial resonance frequency range, the initial resonance frequency range is adjusted, and the scan is performed again until a resonance phenomenon is detected, and the data is recorded.

7. The method for automatically calibrating the operating parameters of a MEMS gyroscope according to claim 1, characterized in that: The specific process of the pull-in detection in step 1 is as follows: First, turn on the amplitude stabilization loop and lock the modal amplitude of the gyroscope to a stable and appropriate value by adjusting the amplitude control loop. Then, design the tuning voltage step. Starting from the initial tuning voltage, increase the tuning voltage step each time. Monitor and record the reading of the pull-in current detector when the gyroscope system is in the un-engaged state. When the current reaches the uA level, record the tuning voltage value at this time and set it as the maximum value of the subsequent tuning voltage output.

8. The method for automatically calibrating the operating parameters of a MEMS gyroscope according to claim 1, characterized in that: The specific process of quality factor calculation in step 1 is as follows: After resonant frequency scanning and pull-in detection, the gyro system is allowed to operate stably near the resonant frequency. At a certain moment, the excitation signal amplitude of the driving mode is set to 0, and the driving amplitude is observed and recorded to decay freely to the initial amplitude. The moment, then calculate the quality factor Q, the formula is as follows, Q = 2πf; Where f is the resonant frequency of the driving mode.

9. The method for automatically calibrating the operating parameters of a MEMS gyroscope according to claim 1, characterized in that: The gyroscope state solving unit includes an orthogonal suppression loop, a force balancing loop, a phase-locked loop, an amplitude stabilization loop and a tuning voltage controller; The orthogonal inhibition loop receives the PID action parameters of the orthogonal inhibition loop output by the intelligent agent, and performs PID control on the error between the orthogonal coupling output amplitude of the gyro sensitive mode and the target amplitude; at the same time, the orthogonal inhibition loop inputs the updated amplitude error of the orthogonal coupling of the gyro sensitive mode back to the intelligent agent as a new environmental state parameter; The force balance loop receives the force balance PID action parameters output by the intelligent agent, and performs PID control on the error between the gyro sensitive modal force balance output amplitude and the target amplitude; at the same time, the force balance loop inputs the updated gyro sensitive modal force balance amplitude error back to the intelligent agent as a new environmental state parameter; The phase-locked loop receives the PID action parameters of the phase-locked loop output by the agent, and performs PID control on the phase error between the steady-state phase of the gyro phase-locked loop drive mode output and the target phase; at the same time, the phase-locked loop inputs the updated gyro drive mode phase error back to the agent as a new environmental state parameter; The amplitude stabilization loop receives the PID action parameters of the amplitude stabilization loop output by the intelligent agent, and performs PID control on the steady-state error of the driving modal amplitude between the output amplitude of the gyro amplitude stabilization loop driving modal amplitude and the target amplitude; at the same time, the amplitude stabilization loop inputs the updated gyro driving modal amplitude error as a new environmental state parameter back to the intelligent agent; A tuning voltage controller receives a tuning voltage combination of a driving mode and a sensitive mode output by the intelligent agent, and performs tuning control on the gyro driving mode and the sensitive mode; At the same time, the tuned voltage controller inputs the updated frequency difference between the gyro driving mode and the sensitive mode back to the intelligent agent as the new environmental state parameter.

10. A MEMS gyroscope operating parameter automatic calibration system, characterized in that: Includes the following modules: The gyroscope basic parameter calculation module is used to scan the resonant frequency, perform pull-in detection, and calculate the quality factor to realize the gyroscope basic parameter calculation to determine the initial value of the gyroscope closed-loop control, including the initial frequency of the phase-locked loop, the initial locked phase of the phase-locked loop, the frequency output range of the phase-locked loop, and the initial locked amplitude of the amplitude stabilization loop; The gyroscope operating parameter automatic calibration module is used to set the initial PID parameters of the four closed loops of the phase-locked loop, amplitude stabilization loop, orthogonal suppression loop, and force balance loop in the gyroscope state solution unit, as well as the initial value of the tuning voltage combination of the tuning voltage controller, so as to output the environmental state parameters; and introduce the deep deterministic policy gradient DDPG reinforcement learning algorithm to input the environmental state parameters into the state policy network in the intelligent agent for analysis to output the action control drive signal; the action evaluation network in the intelligent agent evaluates the performance value of the action control drive signal, and at the same time inputs the action control drive signal into the gyroscope state solution unit to obtain new environmental state parameters; finally, the error is calculated based on the performance value of the previous stage and the current stage, and the action is updated and coordinated through the state policy network and the action evaluation network to realize the automatic calibration of the MEMS gyroscope operating parameters.

Images