A dynamic surface control method and device for an active vibration isolation system

Abstract

The application provides a kind of active vibration isolation system dynamic surface control method and equipment, belong to the field of precision equipment vibration isolation, establish system state space model;Collect and preprocess load position data;Train LSTM neural network to obtain prediction model;Real-time system state is obtained and load position is predicted;Based on the predicted position, design a virtual control law with advance compensation;Recursion is defined to follow-up control law through dynamic surface control method;Calculate the actuating force of throttle valve;Output control force and update state to execute in a cycle.By LSTM neural network, the load position is predicted in advance and smoothed, and the prediction information is deeply integrated into the design of the dynamic surface controller, achieving advance compensation of the system vibration, improving the control accuracy and dynamic response speed of the vibration isolation system, and ensuring the stability and robustness of the system.

Description

Technical Field

[0001] This invention belongs to the field of vibration isolation technology for precision equipment, and specifically relates to a dynamic surface control method and device for an active vibration isolation system that combines Long Short-Term Memory (LSTM) neural network position prediction with Dynamic Surface Control (DSC). Background Technology

[0002] In fields such as precision manufacturing, precision measurement, and aerospace inertial navigation, micro-vibrations in the working environment are one of the key factors affecting equipment performance and yield. Active vibration isolation systems, by applying counter-control forces to counteract vibrations, are the core devices that ensure the stable operation of these precision equipment.

[0003] Traditional active vibration isolation systems generally employ proportional-integral-derivative (PID) control. While PID controllers are simple in structure and easy to tune, their performance is often inadequate when dealing with systems like air-bearing vibration isolation platforms, which exhibit inherent nonlinearity, time-delay characteristics, and complex external disturbances. This often manifests as response lag, significant overshoot, and insufficient suppression of vibrations at specific frequencies. Although compensation can be achieved using feedforward control and notch filters, this increases the system's complexity and tuning difficulty.

[0004] Related technologies adjust control actions based on current or historical load position errors. When external disturbances occur, there is a time lag in the control response, making it impossible to promptly contain the spread of vibration. This leads to an increased deviation of the load position from the desired trajectory, affecting the operational stability of precision equipment. Single-degree-of-freedom air-bearing vibration isolation systems involve the coupling of multiple components such as throttle valves, air chambers, and loads, resulting in complex dynamic characteristics. It is difficult to construct a precise model that closely matches reality using control methods, and the control accuracy cannot meet the requirements of high-end precision equipment.

[0005] The position prediction models in related technologies mostly adopt single-point prediction modes, and the output results are easily affected by noise, have strong randomness, and cannot provide a smooth trend reflecting position changes over a future period of time, making them difficult to use directly for control strategy design. The control variables calculated by the control strategy often do not fully consider the physical constraints of the actuator, which may result in amplitudes exceeding the rated range or excessively rapid rates of change, causing the actuator to fail to respond accurately to commands, or even to be damaged due to long-term overload, interrupting the continuous operation of the system. Summary of the Invention

[0006] This invention provides a dynamic surface control method for an active vibration isolation system. By establishing an accurate dynamic model of the throttle valve-air chamber-load, and combining data-driven advance prediction with model-driven robust control, it solves the problems of response lag and insufficient nonlinear compensation in traditional methods, thereby improving the dynamic performance and control accuracy of the system.

[0007] The methods include:

[0008] S101: Establish the state-space model of the active vibration isolation system, and define the system state variables and control inputs;

[0009] S102: Collect time-series data of the load location and preprocess the time-series data to obtain smoothed and denoised load location data;

[0010] S103: Train an LSTM neural network using smoothed and denoised load location data to obtain a load location prediction model; during training, the input is a historical load location sequence, and the output is a Gaussian-weighted future load location trend value;

[0011] S104: During real-time control, the current system state variables are acquired, and the load position prediction model is used to predict the load position to obtain the predicted load position.

[0012] S105: Define the load position tracking error based on the predicted load position and the expected load position, and design the first virtual control law; the first virtual control law introduces the predicted load position for advance compensation;

[0013] S106: Based on the first virtual control law, subsequent virtual control laws and tracking errors are defined step by step through dynamic surface control methods; subsequent virtual control laws include load speed tracking surface control law, air chamber pressure tracking surface control law, throttle valve core displacement tracking surface control law, and throttle valve core speed tracking surface control law.

[0014] S107: Calculate the throttle valve as the final control input based on all virtual control laws and tracking errors;

[0015] S108: The throttle valve outputs power to the throttle valve actuator, and updates the system state variables and load position prediction in real time, repeatedly executing the control cycle.

[0016] According to another embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a dynamic surface control method for an active vibration isolation system.

[0017] As can be seen from the above technical solutions, the present invention has the following advantages:

[0018] The active vibration isolation system dynamic surface control method provided by this invention eliminates dimensional differences and measurement interference in load position data through normalization and robust quadratic regression smoothing and denoising, obtaining smooth and pure data to ensure the accuracy of subsequent model training and control calculations. An LSTM network uses historical smoothed data extracted via a sliding window as input, generates the output target through Gaussian weighted summation, accurately captures the temporal variation of load position, and outputs stable future position trend values, providing high-quality leading information for control. A primary virtual control law is integrated with the predicted load position, initiating compensation actions before the load deviates from the desired trajectory, improving the timeliness of vibration suppression and reducing the risk of position deviation expansion. Real-time control synchronously updates system state variables and predicted input data to maintain real-time performance, allowing the control method to adapt to changes in operating conditions and ensuring stable control effects. Control quantities undergo amplitude and rate of change verification and over-limit processing to match the physical capabilities of the actuators, avoiding overload damage to the actuators and extending equipment lifespan. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 Flowchart of dynamic surface control method for active vibration isolation system;

[0021] Figure 2 This is a flowchart illustrating an embodiment of a dynamic surface control method for an active vibration isolation system.

[0022] Figure 3 This is a schematic diagram of an electronic device. Detailed Implementation

[0023] The dynamic surface control method for an active vibration isolation system involved in this invention is based on an active vibration isolation system, which includes the following components:

[0024] Mechanical device: including load (vibration isolation object), base and multiple vibration isolators disposed between the two.

[0025] Sensing system: including position sensor, speed sensor, and pressure sensor.

[0026] Optionally, the position sensor can be a proximity sensor or a laser displacement sensor. The velocity sensor can be obtained based on the differentiation of the position signal or by using a detector, and the pressure sensor is used to detect the pressure inside the air chamber.

[0027] Actuator: A precision throttle valve is used to regulate the gas flow rate into / out of the gas chamber, thereby changing the gas chamber pressure.

[0028] Control system: A real-time target machine, i.e. the controller, is used to embed the control algorithm of this invention to process sensor data, perform LSTM prediction and DSC calculation, and generate control signals to drive the actuator, so as to achieve stable control of the load and vibration suppression.

[0029] The following describes in detail the dynamic surface control method for active vibration isolation systems according to this application. Specific details, such as particular system structures and technologies, are presented for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details.

[0030] It should be understood that, when used in this specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0031] The terms "one embodiment" or "some embodiments" used in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this application do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 and 2 The diagram shows a flowchart of a dynamic surface control method for an active vibration isolation system in a specific embodiment. The method includes:

[0034] S101: Establish the state-space model of the active vibration isolation system, and define the system state variables and control inputs.

[0035] The system state variables include load position, load speed, relative pressure of the air chamber, throttle valve core displacement, and throttle valve core speed. The control input is the throttle valve as the power source.

[0036] S101 specifically includes the following steps:

[0037] S1011: Determine the state variables of the system.

[0038] The load displacement is defined as a first state variable. The load speed is defined as a second state variable. The relative pressure of the air chamber, i.e., the difference between the air chamber pressure and the ambient pressure, is defined as the third state variable. The displacement of the throttle valve spool is defined as the fourth state variable. The throttle valve spool speed is defined as the fifth state variable. .

[0039] In some embodiments, by selecting a minimal set of variables that can describe the dynamic behavior of the system, namely state variables, a continuous physical system is abstracted into a mathematical representation, thus clarifying the controller's observations and target objects.

[0040] S1012: Establish the load dynamics equation and decouple the pressure term.

[0041] Based on Newton's second law, and according to the dynamic equation of the load... Rewrite it as:

[0042] .

[0043] Using the state variables defined in step S1011, Replace with ,Will Replace with ,Will Replace with Thus, the load acceleration equation expressed in terms of state variables is obtained.

[0044]

[0045] in .

[0046] S1013: Establish the gas chamber pressure dynamics equation and decouple the valve core displacement term.

[0047] Based on the law of conservation of mass and the equation of state for gases, and according to the dynamic equation of the pressure in the gas cavity...

[0048]

[0049] Among them Represented as ( + Using the state variable defined in step S1011, replace z with , load speed Replace with Replace (Pc-Pa) with Displace the throttle valve core Replace with Thus, the pressure change rate equation expressed in terms of state variables is obtained.

[0050]

[0051] in .

[0052] This embodiment describes the pressure change mechanism inside the air chamber based on fluid mechanics and the law of conservation of mass. The relationship between the pressure change rate, the mass flow rate into the air chamber, and the change in air chamber volume due to load motion is established. The above equations describe the aerodynamic process using state variables, and the decoupled... The cascaded control structure is clearly defined. The fourth state variable, namely the valve core displacement, is the third state variable. Virtual control input.

[0053] S1014: Establish the dynamic equation of the throttle valve and decouple the dynamic term.

[0054] Based on Newton's second law, the dynamic equation for configuring a throttle valve is...

[0055]

[0056] in: : Throttling valve spool displacement; Valve core equivalent mass; Damping coefficient of valve core movement; : Stiffness of the valve core's return spring; : The force (control quantity) acting on the valve core; The pressure-bearing area of ​​the valve core; Air chamber pressure Environmental pressure difference, Atmospheric pressure.

[0057] The dynamic equation of the throttle valve is rewritten as follows:

[0058] .

[0059] Using the state variables defined in step S1011, Replace with The speed of the throttle valve core Replace with ,Will Replace with Control input Defined as Thus, the valve core acceleration equation expressed in terms of state variables is obtained. .

[0060] in, .

[0061] In this embodiment, the throttle valve of the actuator is modeled as a second-order mechanical system, whose motion is determined by the working force, spring force, damping force, and air pressure. This system is then structurated through deformation. In this form, the actual physical control quantity is used as the driving force. With state variables The association completes the modeling of the control chain from the control input to the final controlled object.

[0062] S1015: A state-space model integrated into a strict feedback form.

[0063] Integrate all the equations obtained in steps S1011 to S1014 to form a system state-space model in strict feedback form:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] in, The vector representing the i-th state.

[0070] In this embodiment, the equations of all the aforementioned subsystems are arranged according to the derivatives of the state variables, forming a system of differential equations. Step S1015 allows the derivative of each state variable to be expressed as a function of itself and the preceding state variables, including the next state variable as input. The resulting rigorous feedback form is the direct basis for applying advanced nonlinear control methods such as dynamic surface control. This form allows the controller to be designed from the innermost load position loop to the outermost valve core force loop. By designing virtual control laws for each subsystem and ensuring their stability, the stability and tracking performance of the system are ultimately ensured.

[0071] S102: Collect time-series data of the load location and preprocess the time-series data to obtain smoothed and denoised load location data. The preprocessing includes normalization and robust quadratic regression smoothing and denoising.

[0072] S102 specifically includes the following steps:

[0073] S1021: Configure the data acquisition system and set the sampling parameters. Connect the high-precision position sensor to the real-time controller, and set the data sampling frequency to be more than twice the main vibration frequency of the system, optionally not less than 500Hz, to ensure that complete load position time-series data is acquired.

[0074] S1022: Perform data collection and build the raw dataset.

[0075] The data acquisition system continuously collects data on the load position of the vibration isolation object during operation, forming an original time-series data sequence. ,in This represents the load location value at the i-th sampling time.

[0076] In some embodiments, the mechanical displacement of the load is converted into an electrical signal by a sensor, and then quantized into a digital sequence by a data acquisition card. Continuous acquisition constitutes a time series, reflecting the dynamic behavior of the load position changing over time.

[0077] S1023: Normalize the original load location time series data.

[0078] The original dataset Z obtained in step S1022 is subjected to Min-Max normalization calculation, which linearly maps it to the interval [0,1].

[0079] Specifically, for each data point in the sequence, according to the formula...

[0080]

[0081] The calculation is performed to obtain the normalized sequence. .

[0082] S1024: Robust quadratic regression is used to smooth and denoise the normalized data.

[0083] The normalized sequence obtained in step S1023 A robust quadratic regression method is applied for smoothing. The method involves solving an optimization problem. This is achieved by using ρ, where ρ is the Huber loss function.

[0084] In some embodiments, the method can effectively suppress measurement noise and occasional outliers in the data, thereby extracting the true trend of load location changes.

[0085] S1025: Define the Huber loss function and perform iterative weighted fitting.

[0086] The Huber loss function is defined as:

[0087]

[0088] Where r is the residual and the threshold parameter δ is 1.345. The optimization problem is solved by iterative reweighted least squares method, with outliers assigned smaller weights.

[0089] S1026: Outputs the load position data after smoothing and denoising.

[0090] The numerical sequence corresponding to the smoothed curve obtained from the robust quadratic regression fitting in step S1025 is taken as the final smoothed and denoised load location data, denoted as... .

[0091] In some embodiments, the Huber loss function combines the advantages of squared loss and absolute value loss. For small residuals (|r|≤δ|r|≤δ), a squared term is used to ensure the fitted function is sufficiently smooth near normal data points. For large residuals (|r|>δ|r|>δ), a linear term is used to reduce the excessive influence of outliers on the overall optimization objective. Through iterative solving, the weight of each data point is dynamically adjusted according to the current residual, achieving intelligent processing of noise and outliers. This embodiment obtains the load position signal. . It preserves the true trend of the system, filters out noise and outliers, and improves the stability of the entire system.

[0092] S103: Train an LSTM neural network using smoothed and denoised load location data to obtain a load location prediction model. During training, the input is a sequence of historical load locations, and the output is a Gaussian-weighted trend value of future load locations.

[0093] Specifically, S103 includes the following steps:

[0094] S1031: Determine the data structure for network input and output.

[0095] Set the sliding window length L to be used to extract data from the smoothed and denoised load location data sequence. Extract historical sequences; set the total prediction step size P and Gaussian window parameters μ and σ to construct an output target representing the future trend.

[0096] Step S1032: Generate training input samples for the LSTM network.

[0097] Using a sliding window of length L, the sequence after smoothing and denoising is... Slide the slider up and down to extract continuous subsequences of historical load positions to form the training input feature set. Each sample .

[0098] S1033: Generate the training output target of the LSTM network.

[0099] For each input sample Normalize the original data of its corresponding future P points. Perform Gaussian weighted summation to calculate the output target. The weight .

[0100] This embodiment enables the LSTM network to learn an output that represents a future trend. The prediction results are stable and suitable as input to a feedforward controller, which can generate anticipatory control commands.

[0101] S1034: Construct the LSTM neural network model structure.

[0102] Stack the following network layers in sequence: an input layer with dimensions consistent with the sliding window length L; one or more LSTM layers, each containing a predetermined number of hidden units; a Dropout layer with a dropout rate set between 0.2 and 0.5; a fully connected layer for integrating features; and a regression output layer that outputs a single predicted value.

[0103] In some embodiments, LSTM layers selectively memorize and forget information through internal input, forget, and output gates, effectively capturing long-term dependencies in time series. Dropout layers randomly shut down a subset of neurons during training, forcing the network to learn more robust features by not relying on a few specific nodes. Fully connected layers are responsible for mapping the temporal features extracted by the LSTM layers to the final output values. Dropout layers perform regularization techniques, improving generalization ability.

[0104] S1035: Configure model training parameters and start the training process.

[0105] The mean squared error or root mean squared error is selected as the loss function of the network. The Adam optimizer is used to minimize the loss function, and early stopping is used during training to monitor the performance on the validation set to prevent overfitting.

[0106] S1036: Complete model training and save the final model.

[0107] Once the training reaches the early stopping condition or the maximum number of iterations, the trained LSTM network weights and structural parameters are saved to form a load location prediction model that can be used for real-time prediction.

[0108] In some embodiments, the mean squared error loss function measures the average squared difference between the network's predicted values ​​and the Gaussian-weighted target value, guiding the optimizer to update the network weights in a direction that reduces this difference. The Adam optimizer combines the advantages of momentum and adaptive learning rate, enabling stable convergence. The Adam optimizer accelerates the training process and improves convergence stability, ensuring that the final model has good performance in practical applications.

[0109] The model file saved in this embodiment encapsulates the system dynamic laws learned from the data, enabling it to be loaded and executed for forward propagation calculations without being placed in the training environment during the real-time control phase, thereby generating load location prediction values ​​in real time and ensuring the availability and independence of the prediction function.

[0110] S104: During real-time control, the current system state variables are obtained, and the load position prediction model is used to predict the load position to obtain the predicted load position.

[0111] S104 specifically includes the following steps:

[0112] S1041: Initialize the real-time control loop and data buffer.

[0113] Configure a first-in-first-out (FIFO) data buffer of length L in the controller to store the latest load position data, and initialize the buffer to the initial load position value when the system starts up.

[0114] It should be noted that LSTM models require input to be a fixed-length continuous historical sequence. The first-in, first-out buffer simulates the time points used in the training phase, ensuring that at any given time, the controller can provide smoothed positional data from the most recent L time steps as the model's input context.

[0115] S1042: Read sensor data and update system status.

[0116] At the beginning of each control cycle, the measured values ​​of the load position sensor, pressure sensor and valve core displacement sensor are read synchronously to update the current values ​​of state variables x1, x3 and x4, and the current values ​​of x2 and x5 are updated by differential calculation or direct measurement.

[0117] In some embodiments, state variables x1, x3, and x4 are derived from sensors, while load speed x2 and valve spool speed x5 are obtained by differential analysis of the position signals or by using an observer. This allows the controller and LSTM predictor to acquire the system's current instantaneous dynamic state information.

[0118] S1043: Preprocess the real-time load location data.

[0119] After obtaining the current load position x1 from step S1042, normalization is performed using the same normalization parameters as in the training phase, i.e., based on the training dataset's min(Z) and max(Z), using the following formula:

[0120]

[0121] The normalized value is smoothed in real time using the same robust quadratic regression algorithm as in the training phase, resulting in... .

[0122] S1044: Update the input buffer and construct the model input.

[0123] The smoothed real-time load location data obtained in step S1043 Push the FIFO buffer initialized in step S1041 onto the buffer, while discarding the oldest data point in the buffer, thus forming the latest historical load position sequence of length L. This serves as the current input to the load location prediction model.

[0124] This embodiment updates the buffer based on the arrival of new, smoothed positional data, ensuring that the buffer always contains the latest L data points, traced back from the current moment. This continuously updated sequence constitutes all the contextual information needed for the LSTM network to understand the system's recent dynamics and infer its future trends.

[0125] S1045: Perform forward propagation calculations for the LSTM model.

[0126] The constructed current input sequence The input is fed into a deployed, trained load location prediction model. This model performs forward propagation calculations, passing through an LSTM layer, a Dropout layer, and a fully connected layer, ultimately producing a Gaussian-weighted prediction of future load location trends at the regression output layer. .

[0127] S1046: Perform inverse normalization on the model's predicted output.

[0128] The prediction output at the normalized scale obtained in step S1045 Performing inverse normalization calculations to map it back to the original physical dimensions yields the final predicted load location in the physical world. .

[0129] In some embodiments, the trained LSTM model is invoked as a non-linear function mapping. Input sequence The input flows sequentially through each layer of the network. LSTM units update their state and produce an output based on their internally stored state and the current input. Dropout layers act as a scaling mechanism during inference to compensate for random dropouts during training. The fully connected layers ultimately map this high-dimensional temporal feature vector to a single predicted value. The prediction results based on LSTM networks can be used by dynamic surface controllers to predict load locations. Having the same dimensions and physical meaning as the desired trajectory yd and the state variable x1, it can be integrated into the calculation of the DSC control law, thus completing the transition from data-driven prediction to model-driven control.

[0130] S105: Define the load position tracking error based on the predicted load position and the desired load position, and design the first virtual control law. The first virtual control law introduces the predicted load position for advance compensation.

[0131] In some embodiments, the desired load position trajectory yd(t) is set according to the vibration isolation requirements. For precision vibration isolation, yd(t) = 0. The slowly varying trajectory is generated by linear interpolation, and the desired velocity is obtained by differentiating yd(t) with a step size equal to the sampling period. Extract x1 from the system state variables, and then... The system calculates position tracking errors in real time and stores time-series data. Treated as a virtual control variable, its expected value is defined. To stabilize Subsystem.

[0132] In this embodiment, the predictive control gain δ is set to 0.5-2.0, and a lead compensation term is constructed. Γ represents the predicted load location, and the location error gain K1 = 5 - 20 is set according to the error exponential convergence requirement. Substituting this into the formula... Calculate the first virtual control law.

[0133] Introduce a first-order low-pass filter, according to initialization Solving using the Euler method Calculate |α1 If the speed exceeds 0.005 m / s, adjust τ2 until the requirement is met.

[0134] S106: Based on the first virtual control law, subsequent virtual control laws and tracking errors are defined step by step using a dynamic surface control method. Subsequent virtual control laws include the load speed tracking surface control law, the air chamber pressure tracking surface control law, the throttle valve spool displacement tracking surface control law, and the throttle valve spool speed tracking surface control law.

[0135] In some embodiments, dynamic surface control logic is used to progressively pass control commands down from the position control target. Each level defines a corresponding tracking error and virtual control law, integrating the requirements of preceding error propagation with the current level's system dynamics compensation. A first-order low-pass filter avoids analytical differentiation, forming a progressive control chain of position, velocity, pressure, valve core displacement, and valve core velocity. This decomposes complex multivariate control tasks, reduces the design difficulty of control laws at each level, ensures that deviations in each link are compensated specifically, and the filter makes control commands continuous and smooth, improving control stability.

[0136] S107: Calculate the throttle valve power as the final control input based on all virtual control laws and tracking errors.

[0137] In some embodiments, the speed error of the end valve core and the error transmission requirements of each level at the front end are integrated to compensate for the dynamic characteristics of the throttle valve. The comprehensive control requirements are transformed into the throttle valve as the driving force through the control law formula. After amplitude and rate of change verification, it is ensured that the control quantity is within the physical capability range of the actuator, and the over-limit handling ensures the feasibility of the command.

[0138] S108: The throttle valve outputs power to the throttle valve actuator, and updates the system state variables and load position prediction in real time, repeatedly executing the control cycle.

[0139] In some embodiments, the digital control signal is converted into an analog drive signal recognizable by the actuator, which is then amplified to provide sufficient power to drive the throttle valve. System state variables and predicted input data can also be updated. Through cyclic monitoring and anomaly handling, real-time control and continuity are maintained, ensuring that the throttle valve adjusts the air chamber pressure as expected, guaranteeing the continuous and stable operation of the vibration isolation system, and improving overall control reliability.

[0140] In one embodiment of the present invention, based on step S105, the following is a possible embodiment and its specific implementation will be described in a non-limiting manner. S105 specifically includes the following steps:

[0141] S1051: Defines the load position tracking error.

[0142] Calculate the current load location The difference between the load position trajectory and the expected load position trajectory yd(t) is defined as the load position tracking error. , .

[0143] In some embodiments, by comparing the load position of the controlled object The control deviation of the system is obtained by comparing it with the desired reference signal yd(t). The error signal can drive the controller to generate corrective actions, reducing the position tracking error. The defined error... This enables the controller to react to the current control effect.

[0144] S1052: Calculate the derivative of the desired load location trajectory.

[0145] The desired load velocity can be obtained by analytical or numerical differentiation of the given desired load location trajectory yd(t). .

[0146] In some embodiments, during vibration isolation, yd(t) is a zero or slowly varying signal, and its derivative... This represents the desired trend of motion. This differential value can be obtained in real time using analytical or numerical methods.

[0147] S1053: Obtain the predicted load location and calculate the prediction deviation.

[0148] Obtain the predicted load location from step S104. And calculate its relationship with the desired load location. The difference is used to obtain the prediction bias. .

[0149] In some embodiments, the future load locations predicted by the LSTM network are used and compared with the current expected values ​​to obtain the prediction bias. It quantifies the trend of the system deviating from the expected trajectory in the short term, obtains potential disturbances or dynamic changes, and provides data support for controller adjustment.

[0150] S1054: Design the basic structure of the first virtual control law.

[0151] Constructing virtual control laws Virtual control laws include desired load speed. Proportional feedback term for load position tracking error and feedforward compensation terms based on prediction bias ,in and It is an adjustable positive control gain.

[0152] S1055: Combine the terms to form a complete virtual control law.

[0153] The results obtained in steps S1052, S1051, and S1053 are combined according to the structure designed in step S1054 to form the mathematical expression of the first virtual control law: .

[0154] S1056: Filter the first virtual control law.

[0155] The virtual control law obtained in step S1055 Through a time constant A first-order low-pass filter is used to obtain the filtered output signal. Its dynamics are determined by Description, initial conditions are set to .

[0156] In some embodiments, the desired speed provides a reference traction, and the error feedback term Provides damping and correction forces, and predicts compensation terms. Then, based on the predicted future deviation, a reverse control action is applied in advance.

[0157] The filter takes a rapidly changing signal as input and outputs a smooth, continuous signal. The filter's time constant... Very small, designed to ensure Capable of rapid tracking At the same time, filter out theoretically calculated High-frequency noise or numerical instability at that time. Easily obtainable derivatives. To replace the original and This simplifies the parameter configuration of the controller, improves the numerical stability of the control algorithm and its feasibility in real-time systems, and ensures the smoothness of the signal.

[0158] In one embodiment of the present invention, based on step S106, the following is a possible embodiment and its specific implementation is described in a non-limiting manner. Step S106: Based on the first virtual control law, subsequent virtual control laws and tracking errors are defined step by step through a dynamic surface control method. S106 specifically includes the following steps:

[0159] S1061: Define the load speed tracking error and design the second virtual control law.

[0160] Calculate load speed With the output of the first filter The difference is defined as the load speed tracking error. .

[0161] Based on this error, the system dynamics function and control gain Design the second virtual control law

[0162]

[0163] in The control gain is positive.

[0164] In some embodiments, error The actual load speed was measured against the filtered desired speed command. The deviation between them. The virtual control law is designed based on Lyapunov stability theory, and specifically includes: compensation for the system's own dynamics. Item, rate of change of instruction after tracking filter The item provides damping stability error The terms, and the error surface from the previous one. Coupling terms To ensure the stability of the entire system.

[0165] S1062: Filter the second virtual control law.

[0166] Virtual control law With a time constant A first-order low-pass filter is used to obtain the filtered output. Its dynamics are described as follows

[0167] Initial conditions are set as follows .

[0168] In some embodiments, a first-order low-pass filter is used to process the virtual control law. The time constant of the filter The output is very small, making it small. Capable of quickly tracking input The changes are filtered out, and high-frequency components are removed to obtain a smooth signal. and its differential This ensures the stability of control.

[0169] S1063: Define the air chamber pressure tracking error and design the third virtual control law.

[0170] Calculate the relative pressure of the air chamber With the output of the second filter The difference is defined as the air chamber pressure tracking error. Based on this error, the system dynamics function and control gain Design the third virtual control law

[0171]

[0172] in The control gain is positive.

[0173] In some embodiments, a gas chamber pressure tracking error is defined and a third virtual control law is designed. Error This reflects the actual pressure and expected pressure of the air chamber. The gap. Virtual control law. The structure is similar to the previous one, utilizing the dynamics of air chamber pressure in the system model. and input gain Through feedback and feedforward compensation To stabilize the error, and through the term Establishing a stability correlation with the previous error surface ensures the accuracy of pressure control.

[0174] S1064: Filter the third virtual control law.

[0175] Virtual control law With a time constant A first-order low-pass filter is used to obtain the filtered output. Its dynamics are described as follows

[0176]

[0177] Initial conditions are set as follows .

[0178] In some embodiments, filtering is primarily applied to the third virtual control law. The filter uses... As input, produce a smooth output. . Differential signal It can be conveniently obtained from the filter dynamics equations.

[0179] S1065: Define the valve core displacement tracking error and design the fourth virtual control law.

[0180] Calculate the throttle valve core displacement With the output of the third filter The difference is defined as the valve core displacement tracking error. Based on this error, a fourth virtual control law is designed.

[0181]

[0182] in The control gain is positive.

[0183] In some embodiments, a valve core displacement tracking error is defined and a fourth virtual control law is designed. Error Indicates the actual displacement and expected displacement of the valve core Deviation. Virtual control law. The configuration is simplified, and the corresponding subsystem dynamics It's an integral relationship. It's mainly achieved through proportional feedback. To stabilize the displacement error, a feedforward term is introduced. Improve tracking speed through coupling terms Maintaining stability with the pre-stage pressure ring enables precise control of the throttle valve position.

[0184] S1066: Filter the fourth virtual control law.

[0185] Virtual control law With a time constant A first-order low-pass filter is used to obtain the filtered output. Its dynamics are described as follows The initial conditions are set as follows: .

[0186] In some embodiments, the filter processes virtual control laws. Output smooth valve spool speed expected command This provides a smooth input signal and its derivative for calculating the final control law. This ensures that the control signal ultimately acting on the throttle valve actuator is smooth, improving the overall robustness and reliability of the system.

[0187] In one embodiment of the present invention, based on step S107, the following is a possible embodiment and its specific implementation is described in a non-limiting manner. Step S107: Calculate the throttle valve power as the final control input based on all virtual control laws and tracking errors. S107 specifically includes the following steps:

[0188] S1071: Defines the speed tracking error of the throttle valve spool.

[0189] Calculate the throttle valve spool speed With the output of the fourth filter The difference is defined as the valve core speed tracking error. .

[0190] In some embodiments, a throttle valve spool speed tracking error is defined. The actual movement speed of the throttle valve spool is configured to match the desired speed command obtained after layers of filtering. The difference between them. This error can reflect the accuracy with which the actuator tracks instructions.

[0191] S1072: Calculation element of the organization's final control law.

[0192] Prepare all the elements needed to calculate the final control law, including: system dynamics function. The derivative of the fourth filter output Valve core speed tracking error Valve core displacement tracking error Control gain and positive control gain .

[0193] S1073: Calculate the system dynamics function The real-time value.

[0194] Based on the current system state variables The value, according to the formula

[0195]

[0196] Real-time calculation of system dynamics functions The value.

[0197] In some embodiments, using real-time acquired or estimated air chamber pressure x3, valve core displacement x4, valve core velocity x5, and system damping, stiffness, area, and mass, nonlinear and coupling terms in the valve core dynamics are calculated in real-time through a mathematical model. This achieves precise compensation for the actuator's own dynamics. This is accomplished through real-time calculation and offsetting. Due to the effects of inertia, damping, elasticity, and pneumatic pressure, the controller can control the acceleration of the valve core.

[0198] S1074: Obtain the differential signal of the filter output. .

[0199] According to the dynamic equation of the fourth first-order low-pass filter and currently known and The value, obtained through algebraic operations Real-time calculation to obtain differential signal .

[0200] S1075: Combine the terms to form the final control law.

[0201] All elements prepared in steps S1072 to S1074 are combined according to the control law structure to calculate the final throttle valve power. .

[0202] In some embodiments, feedforward compensation, error feedback, and stability coupling terms are integrated into a single control force command. Used to compensate for the nonlinear dynamics of the system itself; As a feedforward term, it improves the tracking response speed; Provides damping to stabilize the error of the innermost loop; This is the coupling term with the previous error surface, ensuring the Lyapunov stability of the system. Finally, divide by... This is to counteract the effect of the control gain, thereby enabling the precise generation of the required driving force to drive the entire active vibration isolation system to achieve the desired dynamic performance.

[0203] S1076: Limit the calculated control input.

[0204] Based on the physical capacity of the throttle valve actuator and the system safety requirements, the throttle valve calculated in step S1075 is powered. Amplitude limiting is applied to ensure that the value is constrained within a preset reasonable range.

[0205] In some embodiments, the theoretically calculated control quantity is saturated and limited based on the maximum output force of the throttle valve actuator, power supply voltage limit, or system safety threshold to ensure that the instructions sent to the actuator are always within its physical linear operating range or safety range.

[0206] In one embodiment of the present invention, based on step S108, the following is a possible embodiment and its specific implementation is described in a non-limiting manner. Step S108: The throttle valve outputs power to the throttle valve actuator, and the system state variables and load position prediction are updated in real time, and the control loop is repeated. To achieve this step, S108 specifically includes the following steps:

[0207] S1081: Converts control signals into actuator drive signals.

[0208] The throttle valve digital quantity u calculated in step S107 is converted into the analog voltage or current drive signal required by the throttle valve actuator through the digital-to-analog conversion module of the real-time controller.

[0209] In some embodiments, the control force u calculated by the real-time controller is a digital quantity, and the throttle valve actuator requires a continuous analog voltage or current signal to drive it. A digital-to-analog converter linearly converts the digital quantity u into a corresponding analog signal according to a preset quantization ratio.

[0210] S1082: Outputs the drive signal to the throttle valve actuator.

[0211] The analog drive signal generated by S1081 is amplified by a power amplifier and then transmitted to the throttle valve actuator to generate the corresponding physical action force Fv.

[0212] In some embodiments, the power amplifier amplifies this signal to provide sufficient voltage and current, enabling the actuator to generate physical force Fv corresponding to the control quantity u, thereby driving the valve core to produce displacement and ensuring that the control command has sufficient energy to drive it.

[0213] S1083: Update the system state variable storage unit.

[0214] All system state variables collected and calculated by sensors during the current control cycle to The latest value is stored in the state variable storage unit of the real-time controller, overwriting the data of the previous cycle.

[0215] In some embodiments, during real-time control, state variables need to be updated over time. Storing the latest acquired and calculated state variable values ​​serves as the basis for all control law calculations in the next control cycle, providing initial conditions for the next control cycle.

[0216] S1084: Update the load location prediction input buffer.

[0217] The current smoothed load position data obtained in step S1043 after preprocessing is added to the end of the input buffer of the LSTM prediction model, while the oldest data point in the buffer is removed, keeping the buffer length L.

[0218] In some embodiments, the LSTM prediction model requires smoothed positional data from the most recent L time steps as input. A first-in, first-out (FIFO) queue management mechanism is employed, adding the latest data to the end of the queue and discarding the oldest data, ensuring that the buffer always stores data from the L consecutive sampling points traced back from the current time step, thus guaranteeing the real-time performance and accuracy of the prediction results.

[0219] S1085: Perform real-time control cycle timing management.

[0220] After all tasks in the current control cycle are completed, the timer of the real-time controller is started. After waiting for one sampling period T, the start of the next control cycle is triggered to ensure the stability of the control frequency.

[0221] In some embodiments, after all tasks of a control cycle are completed, the program enters a waiting state, where a hardware timer generates an interrupt after a precise sampling time T, triggering the start of the next control cycle. This timing management method ensures that the time intervals of each control cycle are equal.

[0222] S1086: Monitor system operating status and handle anomalies.

[0223] Throughout the entire control loop execution process, the rationality of sensor data, state variables, and control output values ​​is monitored in real time. When abnormal situations such as data exceeding limits or communication interruption are detected, a preset safety protection mechanism is activated.

[0224] In some embodiments, by monitoring in real time whether sensor readings are within a reasonable range, whether state variables undergo abrupt changes, and whether control output is saturated, when an anomaly is detected, the system switches to a preset safety strategy, such as maintaining the control output from the previous moment, gradually resetting to zero, or enabling backup control mode, thereby improving the robustness of the control system.

[0225] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0226] like Figure 3 As shown, this application also provides an electronic device, including a display module 103, a memory 102, a processor 101, a communication module 104, and a computer program stored in the memory and executable on the processor 101. When the processor 101 executes the program, it implements the steps of a dynamic surface control method for an active vibration isolation system.

[0227] In embodiments of the present invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments described and / or claimed herein.

[0228] In this embodiment, processor 101 may be implemented using at least one of an application-specific integrated circuit, a programmable logic device, a field-programmable gate array, a processor, a controller, a microcontroller, a microprocessor, or an electronic unit designed to perform the functions described herein. In some cases, such an implementation may be implemented within a controller. For software implementation, implementations such as processes or functions may be implemented with separate software modules that allow the performance of at least one function or operation. Software code may be implemented by a software application (or program) written in any suitable programming language, and the software code may be stored in memory and executed by the controller.

[0229] The display module 103 is used to display information input by the user or information provided to the user. The display module 103 may include a display panel, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like.

[0230] The memory 102 can be used to store software programs and various data. The memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0231] The communication module 104 transmits radio signals to and / or receives radio signals from at least one of a base station, an external terminal, and a server. Such radio signals may include voice call signals, video call signals, or various types of data sent and / or received according to text and / or multimedia messages.

[0232] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dynamic surface control method for an active vibration isolation system, characterized in that, The methods include: S101: Establish the state-space model of the active vibration isolation system, and define the system state variables and control inputs; S102: Collect time-series data of the load location and preprocess the time-series data to obtain smoothed and denoised load location data; S103: Train an LSTM neural network using smoothed and denoised load location data to obtain a load location prediction model; during training, the input is a historical load location sequence, and the output is a Gaussian-weighted future load location trend value; S104: During real-time control, the current system state variables are acquired, and the load position prediction model is used to predict the load position to obtain the predicted load position. S105: Define the load position tracking error based on the predicted load position and the expected load position, and design the first virtual control law; the first virtual control law introduces the predicted load position for advance compensation; Calculate the current load location The difference between the load position trajectory and the expected load position trajectory yd(t) is defined as the load position tracking error. , ; This is the current load location. For the desired load location; The desired load velocity can be obtained by analytical or numerical differentiation of the given desired load location trajectory yd(t). ; Obtain the predicted load location from step S104. And calculate the location relative to the desired load. The difference is used to obtain the prediction bias. ; Constructing virtual control laws Virtual control laws include desired load speed. Proportional feedback term for load position tracking error and feedforward compensation terms based on prediction bias ,in, The results are combined according to the basic structure of the first virtual control law to form the mathematical expression of the first virtual control law: ; The obtained virtual control law Through a time constant of A first-order low-pass filter is used to obtain the filtered output signal. kinetics by Description, initial conditions are set to ; S106: Based on the first virtual control law, subsequent virtual control laws and tracking errors are defined step by step using dynamic surface control methods; Subsequent virtual control laws include load speed tracking surface control law, air chamber pressure tracking surface control law, throttle valve core displacement tracking surface control law, and throttle valve core speed tracking surface control law; S107: Calculate the throttle valve as the final control input based on all virtual control laws and tracking errors; S108: The throttle valve outputs power to the throttle valve actuator, and updates the system state variables and load position prediction in real time, repeatedly executing the control cycle.

2. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S101 specifically includes the following steps: Define the load displacement as the first state variable. The load speed is defined as the second state variable. The relative pressure of the air chamber, i.e., the difference between the air chamber pressure and the ambient pressure, is defined as the third state variable. The displacement of the throttle valve spool is defined as the fourth state variable. The throttle valve spool speed is defined as the fifth state variable. ; Based on Newton's second law, according to the dynamic equation of the load Rewritten as: ； Using the defined state variables, Replace with ,Will Replace with ,Will Replace with Thus, the load acceleration equation expressed in terms of state variables is obtained. in ; Based on the law of conservation of mass and the equation of state for gases, and according to the dynamic equation of pressure in the gas chamber... Among them Represented as ( + ); Using the defined state variables, replace z with , load speed Replace with Replace (Pc-Pa) with Displace the throttle valve core Replace with Thus, the pressure change rate equation expressed in terms of state variables is obtained. in Based on Newton's second law, the dynamic equation for configuring a throttle valve is... in: : Throttling valve spool displacement; Valve core equivalent mass; Damping coefficient of valve core movement; : Stiffness of the valve core's return spring; The force acting on the valve core; The pressure-bearing area of ​​the valve core; Air chamber pressure Environmental pressure difference, Atmospheric pressure; The dynamic equation of the throttle valve is rewritten as follows: Using the defined state variables, Replace with The speed of the throttle valve core Replace with ,Will Replace with Control input Defined as The valve core acceleration equation in terms of state variables is obtained. ; in, Integrate all the obtained equations to form the system state-space model: in, The vector representing the i-th state.

3. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S102 specifically includes the following steps: Configure the data acquisition system and set the sampling parameters; The data acquisition system continuously collects the load position of the vibration isolation object during operation, forming the original time-series data sequence. ,in This represents the load location value at the i-th sampling time. The original dataset Z is subjected to Min-Max normalization to linearly map it to the interval [0,1]. Robust quadratic regression was used to smooth and denoise the normalized data. Define the Huber loss function and perform iterative weighted fitting to output smoothed and denoised load location data.

4. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S103 specifically includes the following steps: Set the sliding window length L to be used to extract data from the smoothed and denoised load location data sequence. Extract historical sequences from the data; set the total prediction step size P and Gaussian window parameters μ and σ; Using a sliding window of length L, the sequence after smoothing and denoising... Slide the slider up and down to extract continuous subsequences of historical load positions to form the training input feature set. Each sample ; For each input sample Normalize the original data of its corresponding future P points. Perform Gaussian weighted summation to calculate the output target. The weight ; Construct the LSTM neural network model structure, configure the model training parameters and start the training process, complete the model training and save the final model.

5. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S104 specifically includes the following steps: Configure a first-in-first-out data buffer of length L in the controller; At the beginning of each control cycle, the measured values ​​of the load position sensor, pressure sensor and valve core displacement sensor are read, the current values ​​of state variables x1, x3 and x4 are updated, and the current values ​​of x2 and x5 are updated by differential calculation or direct measurement. Starting from the current load position x1, normalization is performed using the same normalization parameters as during the training phase, i.e., based on the training dataset, min(Z) and max(Z) are normalized using the following formula: The normalized values ​​are smoothed in real time using the same robust quadratic regression algorithm as in the training phase, resulting in... ; The obtained smoothed real-time load location data Push the initialized FIFO buffer onto the data, discard the oldest data point in the buffer, and thus form the latest historical load position sequence of length L. , which serves as the current input to the load location prediction model; The constructed current input sequence The data is fed into a deployed, trained load location prediction model. The load location prediction model performs forward propagation calculations, passing through an LSTM layer, a Dropout layer, and a fully connected layer in sequence, ultimately producing a Gaussian-weighted prediction of the future load location trend at the regression output layer. ; The predicted output at the normalized scale is obtained. Performing inverse normalization calculations to map it back to the original physical dimensions yields the final predicted load location in the physical world. .

6. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S106 specifically includes the following steps: Calculate load speed With the output of the first filter The difference is defined as the load speed tracking error. ; Based on error and system dynamics function and control gain Design the second virtual control law in A positive control gain; Virtual control law With a time constant A first-order low-pass filter is used to obtain the filtered output. Its dynamics are described as The initial conditions are set as follows: ; Calculate the relative pressure of the air chamber With the output of the second filter The difference is defined as the air chamber pressure tracking error. Based on this error, the system dynamics function and control gain Design the third virtual control law in A positive control gain; Virtual control law With a time constant A first-order low-pass filter is used to obtain the filtered output. Its dynamics are described as follows The initial conditions are set as follows: ; Calculate the throttle valve core displacement With the output of the third filter The difference is defined as the valve core displacement tracking error. Design the fourth virtual control law based on the error. in A positive control gain; Virtual control law With a time constant A first-order low-pass filter is used to obtain the filtered output. The dynamics are described as follows The initial conditions are set as follows: .

7. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S107 specifically includes the following steps: Calculate the throttle valve spool speed With the output of the fourth filter The difference is defined as the valve core speed tracking error. ; Prepare all the elements needed to calculate the final control law, including: system dynamics function. The derivative of the fourth filter output Valve core speed tracking error Valve core displacement tracking error Control gain and positive control gain ; Based on the current system state variables The value, according to the formula Real-time calculation of system dynamics functions The value; Obtain the differential signal of the filter output. ; According to the dynamic equation of the fourth first-order low-pass filter and currently known and The value, obtained through algebraic operations Real-time calculation to obtain differential signal ; All elements prepared in steps S1072 to S1074 are combined according to the control law structure to calculate the final throttle valve power. ; Power the calculated throttle valve Amplitude limiting is applied to ensure that the values ​​are constrained within a preset reasonable range.

8. The dynamic surface control method for an active vibration isolation system according to claim 1, characterized in that, S108 specifically includes the following steps: The calculated throttle valve power digital signal is converted into an analog drive signal; The analog drive signal is amplified and then output to the throttle valve actuator. Update the system state variables acquired in the current control cycle to the storage unit; Update the current smoothed load location data to the prediction model input buffer; Start a timer after the current control cycle ends, and trigger the next control cycle after a fixed time interval; The system monitors system signals during the execution of the control loop and activates a safety protection mechanism in case of anomalies.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the dynamic surface control method for the active vibration isolation system as described in any one of claims 1 to 8.

Images