Piezoelectric ceramic adaptive control method and system based on hybrid model
By constructing a hybrid adaptive control method based on the PI inverse model of the asymmetric Play operator and the DRL feedback model, the large error and uncontrollability problems of piezoelectric ceramics in the open-loop situation are solved, and precise movement and rapid response are achieved.
Patent Information
- Application Number
- CN202510992807.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-17
- Publication Date
- 2025-10-17
AI Technical Summary
Piezoelectric ceramics have large errors and uncontrollable problems in open-loop conditions.
A PI inverse model with an asymmetric Play operator as the core is constructed for feedforward compensation, and feedback compensation is performed through a DRL feedback model, including obtaining asymmetric thresholds, fractional-order creep compensation, and reward function optimization to achieve adaptive control.
After feedforward compensation, the error is reduced to less than 5%, and further reduced to 0.5% through feedback compensation, improving the system's rapid response capability and robustness.
Smart Images
Figure CN120802623A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of piezoelectric ceramic adaptive control, in particular to a piezoelectric ceramic adaptive control method and system based on a hybrid model. BACKGROUND
[0002] Piezoelectric ceramics are a kind of electronic ceramic materials with piezoelectric properties. The main difference from typical piezoelectric quartz crystals that do not contain ferroelectric components is that the crystal phase that constitutes the main component is a grain with ferroelectricity. Since the ceramic is a polycrystalline aggregate with random orientation of grains, the spontaneous polarization vectors of each ferroelectric grain are also randomly oriented. In order to make the ceramic exhibit macroscopic piezoelectric properties, it is necessary to polarize the piezoelectric ceramic after sintering and complex electrode on the end face, so that the original randomly oriented spontaneous polarization vectors are preferentially oriented along the electric field direction. After polarization treatment, the piezoelectric ceramic will retain a certain macroscopic residual polarization intensity after the electric field is removed, so that the ceramic has certain piezoelectric properties. However, piezoelectric ceramics have large errors and are uncontrollable in open loop conditions, so compensation is needed for piezoelectric ceramics in open loop conditions. SUMMARY
[0003] In order to solve the above problems, the purpose of the present application is to provide a piezoelectric ceramic adaptive control technology based on a hybrid model, which aims to solve the problems of large error and uncontrollability of piezoelectric ceramics in open loop conditions.
[0004] In order to achieve the above technical purpose, the present application provides a piezoelectric ceramic adaptive control method based on a hybrid model, comprising the following steps:
[0005] A PI inverse model with an asymmetric Play operator as the core and a DRL feedback model are constructed to control the piezoelectric ceramic adaptively, wherein the PI inverse model is used for feedforward compensation, and the DRL feedback model is used for feedback compensation of the remaining error.
[0006] Preferably, when the asymmetric Play operator is obtained, an asymmetric threshold value is obtained as the asymmetric Play operator according to the boost circuit and the step-down path.
[0007] Preferably, when the PI inverse model is constructed, a standard PI model is constructed according to the asymmetric Play operator, and the PI inverse model is obtained by introducing fractional order creep compensation and using Grünwald-Letnikov approximation.
[0008] Preferably, when the DRL feedback model is constructed, the DRL network is constructed by PyTorch to generate the DRL feedback model, wherein the parameter training is performed by a proximal policy optimization algorithm.
[0009] Preferably, in the feedback compensation, the state variables including tracking error, error change rate, temperature disturbance and comprehensive disturbance are obtained and the reward function is defined through the DRL feedback model to complete the feedback compensation.
[0010] Preferably, in the reward function, the reward function is constructed through the state variables.
[0011] Preferably, in the adaptive control of the piezoelectric ceramic, the temperature drift influence of the piezoelectric ceramic is controlled to 0.25 mu m, the vibration interference displacement fluctuation is ±5 nm, and the recovery time is 40 ms.
[0012] The application discloses a piezoelectric ceramic adaptive control system based on a hybrid model.
[0013] The feedforward compensation module is used for performing feedforward compensation on the piezoelectric ceramic according to the constructed PI inverse model with the asymmetric Play operator as the core to obtain a feedforward compensation result.
[0014] The feedback compensation module is used for performing feedback compensation on the residual error according to the feedforward compensation result through the constructed DRL feedback model.
[0015] The adaptive control module is used for performing adaptive control on the piezoelectric ceramic according to the feedback compensation result.
[0016] Preferably, the feedforward compensation module is further used for obtaining an asymmetric threshold value as the asymmetric Play operator according to the step-up circuit and the step-down path, and constructing a standard PI model according to the asymmetric Play operator, and obtaining the PI inverse model by introducing fractional order creep compensation and adopting Grunwald-Letnikov approximation.
[0017] Preferably, the feedback compensation module is further used for implementing DRL network construction by using PyIorch, completing feedback compensation by obtaining state variables and defining a reward function, wherein the state variables include tracking error, error change rate, temperature disturbance and comprehensive disturbance, and performing parameter training through a proximal policy optimization algorithm when constructing the DRL network.
[0018] The application discloses the following technical effects:
[0019] The application solves the problems of large error and uncontrollability of the piezoelectric ceramic in an open loop by constructing a Prandtl-Ishlinskii (PI) inverse model with a play operator as the core; in the case of only relying on feedforward compensation, the error between the reference signal and the actual output is reduced to within 5%, and then the remaining error is compensated through feedback by training a DRL feedback model through multi-modal data acquisition, so that the error is further reduced, and finally precise movement (error within 0.5%) is realized, and the rapid response ability of the system is improved and the robustness is enhanced. BRIEF DESCRIPTION OF DRAWINGS
[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0021] Figure 1 is an integrated piezoelectric ceramic actuator described in the embodiments of the present application;
[0022] Figure 2 is a dSPACE control platform described in the embodiments of the present application;
[0023] Figure 3 is a piezoelectric ceramic driving power supply described in the embodiments of the present application;
[0024] Figure 4 is a method flowchart described in the embodiments of the present application. DETAILED DESCRIPTION
[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0026] The present application provides a piezoelectric ceramic adaptive control method based on a hybrid model, comprising the following steps:
[0027] The PI inverse model with the asymmetric Play operator as the core is constructed, and the DRL feedback model is constructed, so as to perform adaptive control on the piezoelectric ceramic.
[0028] Further preferably, the piezoelectric ceramic adaptive control method based on the hybrid model provided by the application comprises the following steps: obtaining the asymmetric threshold as the asymmetric Play operator according to the step-up circuit and the step-down path when the asymmetric Play operator is obtained.
[0029] Further preferably, the piezoelectric ceramic adaptive control method based on the hybrid model provided by the application comprises the following steps: constructing the standard PI model according to the asymmetric Play operator when the PI inverse model is constructed, and obtaining the PI inverse model by introducing the fractional order creep compensation and using the Grünwald-Letnikov approximation.
[0030] Further preferably, the piezoelectric ceramic adaptive control method based on the hybrid model provided by the application comprises the following steps: constructing the DRL feedback model by realizing the DRL network construction through PyTorch when the DRL feedback model is constructed, and generating the DRL feedback model, wherein the parameter training is performed through the proximal policy optimization algorithm.
[0031] Further preferably, the piezoelectric ceramic adaptive control method based on the hybrid model provided by the application comprises the following steps: obtaining the state variables and defining the reward function through the DRL feedback model when the feedback compensation is performed, and completing the feedback compensation, wherein the state variables include the tracking error, the error change rate, the temperature disturbance and the comprehensive disturbance.
[0032] Further preferably, the piezoelectric ceramic adaptive control method based on the hybrid model provided by the application comprises the following steps: constructing the reward function through the state variables when the reward function is obtained.
[0033] Further preferably, the piezoelectric ceramic adaptive control method based on the hybrid model provided by the application comprises the following steps: when the piezoelectric ceramic is adaptively controlled, the temperature drift influence of the piezoelectric ceramic is controlled to 0.25 μm, the vibration interference displacement fluctuation is ±5 nm, and the recovery time is 40 ms.
[0034] The application discloses a piezoelectric ceramic adaptive control system based on a hybrid model.
[0035] The feedforward compensation module is configured to perform feedforward compensation on the piezoelectric ceramic according to the PI inverse model with the asymmetric Play operator as the core, and obtain a feedforward compensation result.
[0036] The feedback compensation module is configured to perform feedback compensation on the residual error according to the feedforward compensation result through the constructed DRL feedback model.
[0037] The adaptive control module is configured to perform adaptive control on the piezoceramics according to the feedback compensation result.
[0038] Further preferably, the feedforward compensation module of the piezoceramics adaptive control system based on a hybrid model is further configured to obtain an asymmetric threshold value as an asymmetric Play operator according to the boost circuit and the step-down path, and to construct a standard PI model according to the asymmetric Play operator, and to obtain a PI inverse model by introducing fractional order creep compensation and using Grünwald-Letnikov approximation.
[0039] Further preferably, the feedback compensation module of the piezoceramics adaptive control system based on a hybrid model is further configured to implement DRL network construction by using PyTorch, to complete feedback compensation by obtaining state variables and defining a reward function, wherein the state variables include tracking error, error change rate, temperature disturbance and comprehensive disturbance, and to perform parameter training by using a proximal policy optimization algorithm when constructing the DRL network.
[0040] Embodiment: As shown in the following table 1, the piezoceramics adaptive control technology based on a hybrid model is provided. Figures 1-4
[0041] Table 1
[0042]
[0043]
[0044] The piezoceramics adaptive control technology based on a hybrid model provided by the present application reduces the error between the reference signal and the actual output to within 5% by constructing a Prandtl-Ishlinskii (PI) inverse model with a play operator as the core, only relying on feedforward compensation, and then further reduces the residual error by feedback compensation through training a DRL feedback model by multi-modal data acquisition, and finally realizes precise movement (error within 0.5%).
[0045] Specifically, the improved segmented asymmetric dynamic hysteresis PI inverse model is adopted for feedforward compensation, and the standard PI model is improved in view of three major defects, i.e., insufficient compensation at high frequency, asymmetric boost / step-down path and creep effect.
[0046] Asymmetric Play operator definition:
[0047] Boost path (pressurization):
[0048]
[0049] Buck path (decompression) :
[0050]
[0051] Asymmetric threshold value :
[0052] (General case according to multiple experiments, the boost threshold is 15-30% larger than the buck threshold) ;
[0053] Model construction :
[0054]
[0055] In order to solve the creep effect under low frequency state, we introduce the fractional order creep compensation,
[0056] Creep model :
[0057]
[0058] Where, k c The creep gain is α, and the fractional order is α
[0059] In order to adapt to computer control, Grünwald-Letnikov approximation is used, and the discrete result is:
[0060]
[0061] In the formula, k represents, α represents, u (t-kh) represents, h represents;
[0062] Therefore, the feedforward PI inverse model compensation is:
[0063]
[0064] The piezoceramic adaptive control method based on the mixed model comprises the following steps:
[0065] Step 1: hardware platform: the host computer (PC) is connected with the dspace platform, the dspace is connected with the piezoceramic driving power supply device, and the driver is connected with the piezoceramic actuator.
[0066] Step 2: First, we pass a sinusoidal sweep signal with an amplitude of 5v and a frequency of 0.1, 1, 5, 10, 20, 30, 40, 50 Hz to the piezoelectric ceramic, each frequency band is maintained for 3s, and the displacement data of the piezoelectric ceramic actuator is collected (for fitting the PI model), then the experimental platform is turned off and cooled to normal, then a step signal with an amplitude of 10v is passed in, maintained for 300s, and the displacement data of the piezoelectric ceramic actuator is collected (for fitting the fractional order creep model).
[0067] Step 3: The collected data is identified by matlab inverse model, using least squares estimation, the model formula is given above.
[0068] Identification results: weights: And j is a positive integer, as the data quantity is large, it is not directly shown here.
[0069] Threshold: And j is a positive integer, as the data quantity is large, it is not directly shown here.
[0070] Creep parameters: k c , α; k c = 0.05, α = 0.3;
[0071] Step 4: The hardware platform remains unchanged, the control adopts the previously constructed feedforward compensator and the standard pid feedback controller, and the reference input yref(t) and the real output y actual (t), pid parameters, temperature, vibration acceleration are prepared for real-time collection.
[0072] Step 5: The piezoelectric ceramic is heated from 25℃ to 60℃, the pid parameters are adjusted to achieve the best state, and the data is recorded; different degrees of vibration disturbance are made, and the pid parameters are adjusted to achieve the best state, repeated multiple times, and the data is recorded; temperature and vibration are applied together, with random disturbance, and the pid parameters are adjusted to be stable, and the relevant data is recorded.
[0073] Step 6:
[0074] DRL feedback control implementation
[0075] State space and action definition:
[0076] Define state quantity:
[0077] Tracking error: e(t) = y ref (t) - y actual (t);
[0078] Error rate:
[0079] Temperature disturbance:
[0080] Integrated disturbance:
[0081] Reward function: R(t) = -10|e(t)| - 0.1|u(t)| - 2T(t) - 1.5U(t);
[0082] Step 7:
[0083] Network structure:
[0084] Implement DRL network construction through PyTorch.
[0085] Training parameters:
[0086] Algorithm: PPO (Proximal Policy Optimization)
[0087] Number of iterations: 100,000;
[0088] Batch size: 64;
[0089] Learning rate: 1e-4;
[0090] Step 8: Training is complete, export ONNX model.
[0091] Step 9:
[0092] Real-time control flow:
[0093] (1) Control signal generation:
[0094]
[0095] Feedback compensation: DRL output action ΔK p , ΔK i , dynamically update PID parameters.
[0096] K p (t) = K p (t-1) + ΔK p , K i (t) = K i (t-1) + ΔK i ;
[0097] Feedback control variable:
[0098]
[0099] Total control variable:
[0100]
[0101] Unmodeled disturbance estimation:
[0102] e(t) = y actual (t) - y P1 (t) + y temp (t);
[0103] Wherein, y P1 (t) is PI model predicted displacement;y temp (t) is temperature drift compensation.
[0104] The results are compared, as shown in table 2:
[0105] Table 2
[0106] Indicator Conventional PID The present scheme Boost rate Temperature drift influence 1.2 μm 0.25 μm 79.1％ Vibration disturbance displacement fluctuation ± 12 nm ± 5 nm 58.3％ Recovery time 200 ms 40 ms 80％
[0107] The application solves the problems of large error and uncontrollability of piezoelectric ceramics in open loop by constructing a Prandtl-Ishlinskii (PI) inverse model with a play operator as the core;In the case of only relying on feedforward compensation, the error between the reference signal and the actual output is reduced to within 5%, then the remaining error is compensated by feedback through the training of the DRL feedback model, and the remaining error is further reduced, and finally the precise movement (error within 0.5%) is realized, and the rapid response ability of the system is improved and the robustness is enhanced.
[0108] The application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of the flows and / or blocks in the flowchart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a machine that implements the functions specified in the flowchart and / or block diagram. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks
[0109] In the description of the application, it should be understood that the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.
[0110] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.
Claims
1. A piezoelectric ceramic adaptive control method based on a hybrid model, characterized in that: The following steps are involved: A PI inverse model with an asymmetric Play operator as the core and a DRL feedback model are constructed to perform adaptive control of piezoelectric ceramics. Feedforward compensation is performed through the PI inverse model, and feedback compensation of the residual error is performed through the DRL feedback model.
2. The piezoelectric ceramic adaptive control method based on a hybrid model according to claim 1, characterized in that: When obtaining the asymmetric Play operator, an asymmetric threshold is obtained according to the boost circuit and the buck path, and is used as the asymmetric Play operator.
3. The piezoelectric ceramic adaptive control method based on a hybrid model according to claim 2, characterized in that: When constructing the PI inverse model, a standard PI model is constructed according to the asymmetric Play operator, and the PI inverse model is obtained by introducing fractional-order creep compensation and adopting the Grünwald–Letnikov approximation.
4. The piezoelectric ceramic adaptive control method based on a hybrid model according to claim 3, characterized in that: When constructing the DRL feedback model, the DRL network construction is implemented through PyTorch to generate the DRL feedback model, wherein parameter training is performed through the proximal strategy optimization algorithm.
5. The piezoelectric ceramic adaptive control method based on a hybrid model according to claim 4, characterized in that: When performing feedback compensation, the DRL feedback model is used to obtain state variables and define a reward function to complete feedback compensation, wherein the state variables include: tracking error, error change rate, temperature interference and comprehensive interference.
6. The piezoelectric ceramic adaptive control method based on a hybrid model according to claim 5, characterized in that: When obtaining the reward function, the reward function is constructed using the state variables.
7. The piezoelectric ceramic adaptive control method based on a hybrid model according to claim 6, characterized in that: When the piezoelectric ceramics are adaptively controlled, the temperature drift of the piezoelectric ceramics is controlled to be 0.25 μm; the vibration interference displacement fluctuation is ±5 nm; and the recovery time is 40 ms.
8. A piezoelectric ceramic adaptive control system based on a hybrid model, used to implement a piezoelectric ceramic adaptive control method based on a hybrid model as claimed in any one of claims 1 to 7, characterized in that: include: The feedforward compensation module is used to perform feedforward compensation on the piezoelectric ceramics based on the constructed PI inverse model with the asymmetric Play operator as the core to obtain the feedforward compensation result; The feedback compensation module is used to perform feedback compensation on the residual error based on the feedforward compensation result through the constructed DRL feedback model; The adaptive control module is used to perform adaptive control on the piezoelectric ceramic according to the feedback compensation result.
9. The piezoelectric ceramic adaptive control system based on a hybrid model according to claim 8, characterized in that: The feedforward compensation module is also used to obtain an asymmetric threshold value as an asymmetric Play operator based on the boost circuit and the buck path; and to construct a standard PI model based on the asymmetric Play operator, and to obtain the PI inverse model by introducing fractional-order creep compensation and adopting the Grünwald–Letnikov approximation.
10. The piezoelectric ceramic adaptive control system based on a hybrid model according to claim 9, characterized in that: The feedback compensation module is also used to implement DRL network construction using PyTorch, and completes feedback compensation by obtaining state variables and defining a reward function, wherein the state variables include: tracking error, error change rate, temperature interference and comprehensive interference; when constructing the DRL network, parameter training is performed through the proximal policy optimization algorithm.