Self-adaptive suspension control system based on dynamic inertia prediction and layered feedforward compensation

Abstract

The invention discloses a self-adaptive suspension control system based on dynamic inertia prediction and layered feed-forward compensation. The self-adaptive suspension control system comprises a basic gravity compensation layer, a dynamic inertia compensation layer and an instruction execution layer, through organic combination of'dynamic inertia prediction 'and'layered feed-forward compensation', rapid, accurate and stable compensation of the suspension system in a high-dynamic and multi-mode motion scene is realized, and the technical bottlenecks of a traditional system in the aspects of response lag, bandwidth limitation, insufficient sensing dimension and the like are fundamentally solved; the method has remarkable innovativeness, practicability and popularization value. According to the method, the dynamic response speed and compensation precision of the system are remarkably improved, the control bandwidth and high-dynamic working condition adaptive capacity of the system are expanded, multi-source sensor fusion and intelligent motion state recognition are achieved, the overall stability and anti-interference capacity of the system are enhanced, the application field is widened, and the engineering practical value is improved.

Description

Technical Field

[0001] This invention relates to the field of motion simulation and gravity compensation technology, specifically to an adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation. Background Technology

[0002] Gravity compensation suspension systems are key equipment in fields such as aerospace simulation, sports science analysis, rehabilitation medical training, and virtual reality interaction. Their core function is to actively apply tension to offset all or part of the weight of the test subject (such as astronauts, athletes, or patients) in order to simulate a microgravity environment or provide motion support.

[0003] Currently, most mainstream suspension systems adopt a feedback control architecture based on tension sensors. They use classic proportional-integral-derivative (PID) controllers or their improved algorithms to measure rope tension in real time and compare it with the target compensation value. This drives the actuator (such as a servo motor) to adjust the tension in order to eliminate steady-state errors.

[0004] However, this type of traditional feedback control system has inherent drawbacks, mainly in the following aspects:

[0005] 1. Response Lag: Tension sensors detect changes in force that have already occurred. When the test object undergoes violent acceleration or deceleration, such as jumping or landing, it generates enormous inertial forces. From the detection of the force change, through the controller's calculation, to the final response of the actuator, there is an unavoidable electromechanical delay, resulting in a severe lag in the compensation force and causing significant instantaneous overweight or weightlessness.

[0006] 2. Insufficient system bandwidth: The effective bandwidth of actuators, sensors and control loops is limited, making it difficult to track and compensate for high-frequency, high-dynamic force impacts such as jumping and collisions, which can easily lead to insufficient compensation or system oscillation.

[0007] 3. Limited sensing dimension: Relying solely on a tension sensor cannot distinguish between static gravity and dynamic inertial force, nor can it predict the future motion trend of the test object, thus limiting the system's performance when dealing with complex dynamic conditions.

[0008] To alleviate the aforementioned problems, existing research attempts to introduce inertial measurement units (IMUs) for multi-sensor fusion or apply predictive control algorithms. However, these solutions are mostly in the theoretical research stage and still have shortcomings in real-time performance, reliability, and generalization ability to different motion modes. A mature, efficient, and engineering-applicable solution is lacking. Therefore, there is an urgent need for a control strategy that can actively predict inertial forces and compensate for them in advance using a feedforward approach, fundamentally improving the response speed and compensation accuracy of suspension systems in high-speed dynamic scenarios. Summary of the Invention

[0009] The purpose of this invention is to provide an adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation, comprising: a basic gravity compensation layer, a dynamic inertia compensation layer, and a command execution layer.

[0010] The basic gravity compensation layer obtains the actual tension in real time based on the tension sensor installed on the rope of the suspension device, calculates the deviation between the actual tension and the target compensation force, and generates a basic compensation force command.

[0011] The dynamic inertial compensation layer monitors the motion state of the object under test through an inertial measurement unit, and predicts the inertial force at the next moment based on the monitored motion state, generating a dynamic compensation force increment.

[0012] The instruction execution layer generates the final control command based on the basic compensation force command and the dynamic compensation force increment, and drives the suspension device to output the target tension based on the final control command to compensate for the weight of the object under test, so as to simulate a microgravity environment or provide motion support.

[0013] Furthermore, the target compensation force is as follows:

[0014] (1)

[0015] In the formula, It represents the target compensating force. It indicates the quality of the object being tested. It represents the acceleration due to gravity at the Earth's surface. This represents the simulated gravitational acceleration on the lunar surface.

[0016] Furthermore, the basic gravity compensation layer includes an adaptive PID controller.

[0017] The deviation between the actual tensile force and the target compensation force is input into the adaptive PID controller, which adjusts the PID parameters in real time based on different working conditions to obtain the basic compensation force command used to balance the static gravity of the object under test.

[0018] Furthermore, the step of monitoring the motion state of the object under test through the inertial measurement unit is as follows:

[0019] The data monitored by the inertial measurement unit is preprocessed, and the preprocessed data is analyzed using the machine learning model LSTM network to identify the current motion pattern of the object under test.

[0020] The LSTM network using the machine learning model includes an input layer, a first LSTM hidden layer, a second LSTM hidden layer, a global average pooling layer, a first fully connected layer, a second fully connected layer, and an output layer.

[0021] The input layer extracts preprocessed data through a sliding time window to obtain multi-dimensional input samples.

[0022] The first LSTM hidden layer is used to perform temporal correlation analysis on the features of multi-dimensional input samples and output the hidden state at the current time step.

[0023] The second LSTM hidden layer is used to perform a second abstraction on the hidden state of the current time step output by the first LSTM hidden layer, extract higher-level temporal features, and output the second abstracted hidden state.

[0024] The global average pooling layer is used to perform global average pooling on the hidden state after secondary abstraction of the output of the second LSTM hidden layer to obtain a static feature vector.

[0025] The first fully connected layer is used to perform nonlinear mapping on the static feature vector, enhance the feature representation capability through the ReLU activation function, and output the features.

[0026] The second fully connected layer is used to map the features output by the first fully connected layer to the original output values ​​of various motion modes.

[0027] The output layer uses the Softmax activation function to convert the original output value of the second fully connected layer into probability values ​​for various motion modes.

[0028] If the highest probability exists, it is determined to be the corresponding sports mode category; if the highest probability does not exist, the final sports mode category is determined by voting.

[0029] Furthermore, the movement modes include stationary, walking, jumping, in the air, and landing.

[0030] Furthermore, the data monitored by the inertial measurement unit includes the triaxial acceleration of the object under test, the triaxial angular velocity of the object under test, the length of the suspension device rope, and the angle of the suspension device rope.

[0031] Furthermore, the preprocessing includes filtering, synchronization, and coordinate unification.

[0032] Furthermore, the inertial force predicted for the next moment based on the monitored motion state is as follows:

[0033] (2)

[0034] In the formula, This represents the inertial force predicted for the next moment. This indicates the effective mass of the object under test in the suspension direction. This represents the acceleration of the object under test in the reverse direction of suspension.

[0035] Furthermore, the increment of the dynamic compensation force is equal in magnitude and opposite in direction to the predicted inertial force at the next moment.

[0036] Furthermore, the final control command is generated by vector superposition of the basic compensation force command and the dynamic compensation force increment, as shown below:

[0037] (3)

[0038] In the formula, For the final control command, , , These represent the sum of the components of the basic compensation force command and the dynamic compensation force increment in different directions, respectively.

[0039] The technical effects of this invention are undeniable. By organically combining "dynamic inertial prediction" and "layered feedforward compensation", this invention achieves rapid, accurate and stable compensation of the suspension system in high dynamic and multi-mode motion scenarios. It fundamentally solves the technical bottlenecks of traditional systems in terms of response lag, bandwidth limitation and insufficient sensing dimensions, and has significant innovation, practicality and promotion value.

[0040] The beneficial effects of this invention include:

[0041] 1. Significantly improves the system's dynamic response speed and compensation accuracy.

[0042] By introducing an IMU-based inertial force feedforward prediction mechanism, the system can generate corresponding dynamic compensation forces before the inertial force occurs, fundamentally overcoming the response lag problem caused by electromechanical delay in traditional feedback control systems. When the test object performs high acceleration and deceleration movements such as jumping, landing, and rapid turning, the system can respond to changes in its motion state in advance, significantly reducing the probability and magnitude of instantaneous weightlessness or overload, and greatly improving the realism, continuity, and human-machine safety of the simulated environment.

[0043] 2. Expand system control bandwidth and adaptability to high dynamic operating conditions

[0044] A hierarchical control architecture is adopted to decouple steady-state gravity compensation from dynamic inertial feedforward compensation, effectively reducing the signal processing burden of a single control loop. The middle dynamic compensation layer is dedicated to handling high-frequency, highly dynamic inertial force components, significantly improving the system's effective control bandwidth. This enables the system to better respond to transient force changes such as impacts and vibrations, avoids insufficient compensation or system oscillations, and enhances the system's tracking stability in complex motion scenarios.

[0045] 3. Achieve multi-source sensor fusion and intelligent motion state recognition

[0046] The system integrates a tension sensor and an IMU unit, combined with machine learning algorithms (such as LSTM networks) to fuse multi-source sensor information, enabling accurate differentiation between "static gravity" and "dynamic inertial force." Through online recognition and prediction of typical motion patterns (such as walking, running, jumping, landing, and mid-air posture adjustment), the system possesses adaptive compensation capabilities for various motion behaviors, demonstrating good generalization and robustness.

[0047] 4. Enhance the overall stability and anti-interference capability of the system.

[0048] The high-level system employs a PID control algorithm with adaptively tunable parameters, which can dynamically adjust control parameters according to real-time load changes and external disturbances, improving the system's adaptability to non-ideal operating conditions. Combined with a feedforward compensation mechanism, it effectively suppresses system chattering and overshoot caused by sudden changes in inertial force, improves the smoothness of output force and the stability of the control process, which is beneficial for extending the life of actuators and enhancing the user experience.

[0049] 5. Expanding application areas and enhancing the practical value of engineering

[0050] This invention can be applied not only to astronaut microgravity simulation training on the ground, but also widely extended to high-value scenarios such as sports biomechanical analysis, weight-loss gait training in rehabilitation medicine, and somatosensory simulation in virtual reality interaction, providing users with a more realistic, stable, and safe gravity compensation experience. The system possesses strong engineering feasibility and system integration capabilities, providing reliable technical support for high-end equipment and intelligent rehabilitation devices, and has broad market application prospects. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of an adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation.

[0052] Figure 2 A diagram illustrating the steps involved in generating instructions. Detailed Implementation

[0053] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0054] Example 1:

[0055] See Figures 1 to 2 An adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation includes: a basic gravity compensation layer, a dynamic inertial compensation layer, and a command execution layer.

[0056] The basic gravity compensation layer obtains the actual tension in real time based on the tension sensor installed on the rope of the suspension device, calculates the deviation between the actual tension and the target compensation force, and generates a basic compensation force command.

[0057] The dynamic inertial compensation layer monitors the motion state of the object under test through an inertial measurement unit, and predicts the inertial force at the next moment based on the monitored motion state, generating a dynamic compensation force increment.

[0058] The instruction execution layer generates the final control command based on the basic compensation force command and the dynamic compensation force increment, and drives the suspension device to output the target tension based on the final control command to compensate for the weight of the object under test, so as to simulate a microgravity environment or provide motion support.

[0059] Example 2:

[0060] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents are described in Example 1. Furthermore, the target compensation force is as follows:

[0061] (1)

[0062] In the formula, It represents the target compensating force. It indicates the quality of the object being tested. It represents the acceleration due to gravity at the Earth's surface. This represents the simulated gravitational acceleration on the lunar surface.

[0063] Example 3:

[0064] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Embodiments 1 and 2, further wherein the basic gravity compensation layer includes an adaptive PID controller.

[0065] The deviation between the actual tensile force and the target compensation force is input into the adaptive PID controller, which adjusts the PID parameters in real time based on different working conditions to obtain the basic compensation force command used to balance the static gravity of the object under test.

[0066] Example 4:

[0067] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Embodiments 1 to 3, further, the step of monitoring the motion state of the object under test through the inertial measurement unit is as follows:

[0068] The data monitored by the inertial measurement unit is preprocessed, and the preprocessed data is analyzed using the machine learning model LSTM network to identify the current motion pattern of the object under test.

[0069] The LSTM network using the machine learning model includes an input layer, a first LSTM hidden layer, a second LSTM hidden layer, a global average pooling layer, a first fully connected layer, a second fully connected layer, and an output layer.

[0070] The input layer extracts preprocessed data through a sliding time window to obtain multi-dimensional input samples.

[0071] The first LSTM hidden layer is used to perform temporal correlation analysis on the features of multi-dimensional input samples and output the hidden state at the current time step.

[0072] The second LSTM hidden layer is used to perform a second abstraction on the hidden state of the current time step output by the first LSTM hidden layer, extract higher-level temporal features, and output the second abstracted hidden state.

[0073] The global average pooling layer is used to perform global average pooling on the hidden state after secondary abstraction of the output of the second LSTM hidden layer to obtain a static feature vector.

[0074] The first fully connected layer is used to perform nonlinear mapping on the static feature vector, enhance the feature representation capability through the ReLU activation function, and output the features.

[0075] The second fully connected layer is used to map the features output by the first fully connected layer to the original output values ​​of various motion modes.

[0076] The output layer uses the Softmax activation function to convert the original output value of the second fully connected layer into probability values ​​for various motion modes.

[0077] If the highest probability exists, it is determined to be the corresponding sports mode category; if the highest probability does not exist, the final sports mode category is determined by voting.

[0078] Example 5:

[0079] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation has the main technical contents described in any one of Examples 1 to 4. Furthermore, the motion modes include stationary, walking, jumping, in the air, and landing.

[0080] Example 6:

[0081] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents are described in any one of embodiments 1 to 5. Furthermore, the data monitored by the inertial measurement unit includes the three-axis acceleration of the object under test, the three-axis angular velocity of the object under test, the length of the suspension device rope, and the angle of the suspension device rope.

[0082] Example 7:

[0083] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation has the main technical contents described in any one of Examples 1 to 6. Furthermore, the preprocessing includes filtering, synchronization, and coordinate unification.

[0084] Example 8:

[0085] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Embodiments 1 to 7, further, the inertial force predicted at the next moment based on the monitored motion state is as follows:

[0086] (2)

[0087] In the formula, This represents the inertial force predicted for the next moment. This indicates the effective mass of the object under test in the suspension direction. This represents the acceleration of the object under test in the reverse direction of suspension.

[0088] Example 9:

[0089] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation has the main technical contents described in any one of Examples 1 to 8. Furthermore, the increment of the dynamic compensation force is equal in magnitude and opposite in direction to the predicted inertial force at the next moment.

[0090] Example 10:

[0091] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Examples 1 to 9, further wherein the final control command is generated by vector superposition of the basic compensation force command and the dynamic compensation force increment, as shown below:

[0092] (3)

[0093] In the formula, For the final control command, , , These represent the sum of the components of the basic compensation force command and the dynamic compensation force increment in different directions, respectively.

[0094] Example 11:

[0095] See Figures 1 to 2 An adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation includes: a basic gravity compensation layer, a dynamic inertial compensation layer, and a command execution layer.

[0096] The basic gravity compensation layer measures the actual tension in real time based on tension sensors installed on the suspension device ropes, calculates the deviation between the actual tension and the target compensation force, and generates a basic compensation force command. This layer is the core steady-state compensation loop, mainly responsible for eliminating the steady-state error of the system.

[0097] The dynamic inertial compensation layer monitors the motion state of the object under test through an inertial measurement unit and predicts the inertial force at the next moment based on the monitored motion state, generating a dynamic compensation force increment. This layer is responsible for the prediction and forward compensation of inertial forces.

[0098] The command execution layer generates the final control command based on the basic compensation force command and the dynamic compensation force increment. Based on this final control command, it drives the suspension device to output the target tension to compensate for the weight of the object under test, thereby simulating a microgravity environment or providing motion support. This layer is the command aggregation and execution unit.

[0099] The control system mainly includes a three-layer control model, namely a high-level basic gravity compensation layer, a middle-level dynamic inertial compensation layer, and a low-level command execution layer, which can realize multi-level control fusion from steady-state compensation to dynamic feedforward.

[0100] Example 12:

[0101] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in Example 11, further specify the target compensation force as follows:

[0102] (1)

[0103] In the formula, It represents the target compensating force. It indicates the quality of the object being tested. It represents the acceleration due to gravity at the Earth's surface. This represents the simulated gravitational acceleration on the lunar surface.

[0104] Example 13:

[0105] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Embodiments 11 to 12, further wherein the basic gravity compensation layer includes an adaptive PID controller.

[0106] The deviation between the actual tensile force and the target compensation force is input into the adaptive PID controller, which adjusts the PID parameters in real time based on different working conditions to obtain the basic compensation force command used to balance the static gravity of the object under test.

[0107] Based on the system's background requirements in aerospace simulation, the operating conditions are divided into the following three categories, and the characteristics and control objectives of each category are clearly defined as follows:

[0108] Table 1. Various working conditions and their typical application scenarios

[0109]

[0110] Examples of PID parameters and adjustment criteria under different operating conditions are shown below:

[0111] Table 2 PID Parameter Adjustment under Static Steady-State Conditions

[0112]

[0113] Table 3 PID Parameter Adjustment for Low Dynamic Slow Motion Conditions

[0114]

[0115] Table 4 PID Parameter Adjustment under High Dynamic Disturbance Conditions

[0116]

[0117] Example 14:

[0118] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Embodiments 11 to 13, further wherein the step of monitoring the motion state of the object under test through the inertial measurement unit is as follows:

[0119] The data monitored by the inertial measurement unit is preprocessed, and the preprocessed data is analyzed using the machine learning model LSTM network to identify the current motion pattern of the object under test.

[0120] The LSTM network using the machine learning model includes an input layer, a first LSTM hidden layer, a second LSTM hidden layer, a global average pooling layer, a first fully connected layer, a second fully connected layer, and an output layer.

[0121] The input layer extracts preprocessed data through a sliding time window to obtain multi-dimensional input samples.

[0122] The first LSTM hidden layer is used to perform temporal correlation analysis on the features of multi-dimensional input samples and output the hidden state at the current time step.

[0123] The second LSTM hidden layer is used to perform a second abstraction on the hidden state of the current time step output by the first LSTM hidden layer, extract higher-level temporal features, and output the second abstracted hidden state.

[0124] The global average pooling layer is used to perform global average pooling on the hidden state after secondary abstraction of the output of the second LSTM hidden layer to obtain a static feature vector.

[0125] The first fully connected layer is used to perform nonlinear mapping on the static feature vector, enhance the feature representation capability through the ReLU activation function, and output the features.

[0126] The second fully connected layer is used to map the features output by the first fully connected layer to the original output values ​​of various motion modes.

[0127] The output layer uses the Softmax activation function to convert the original output value of the second fully connected layer into probability values ​​for various motion modes.

[0128] If the highest probability exists, it is determined to be the corresponding sports mode category; if the highest probability does not exist, the final sports mode category is determined by voting.

[0129] The LSTM network analyzes the preprocessed data, following a standardized process of "data input - feature extraction - pattern determination," with the specific steps as follows:

[0130] Step 1: Data Input and Timing Window Construction

[0131] The preprocessed data after "filtering, synchronization, and coordinate unification" is received and continuously collected at a sampling frequency of 20ms / time; a sliding time window is used to extract the data to form input samples with dimensions [5,8] to ensure temporal continuity.

[0132] Step 2: Temporal Feature Extraction from LSTM Hidden Layers

[0133] 1. The first layer of LSTM performs temporal correlation analysis on the 8-dimensional features of the input sample. It retains key historical features through the forget gate, captures the current instantaneous features through the input gate, and outputs the hidden state of the current time step through the output gate.

[0134] 2. The second LSTM layer performs a second abstraction on the 64-dimensional hidden state output by the first layer, extracting higher-level temporal features and outputting a 32-dimensional hidden state;

[0135] 3. Perform global average pooling on the 32-dimensional hidden state output of the second layer to obtain a 32-dimensional static feature vector, thus eliminating the difference in dimensionality at each time step.

[0136] Step 3: Feature Mapping of Fully Connected Layers

[0137] 1. The first fully connected layer performs nonlinear mapping on the 32-dimensional static feature vector, enhances the feature representation capability through the ReLU activation function, and outputs 64-dimensional features;

[0138] 2. The second fully connected layer maps the 64-dimensional features to 5-dimensional logits, corresponding to the original output values ​​of the 5 motion modes.

[0139] Step 4: Determine the motion pattern

[0140] 1. The output layer uses the Softmax activation function to convert the 5-dimensional logits into probability values ​​for 5 motion modes, corresponding to stationary, walking, jumping, airborne, and landing respectively;

[0141] 2. If the probability is the highest, directly determine the current motion pattern as the corresponding category; if it is less than (e.g., the transition phase from "jump to mid-air"), use the recognition results of three consecutive input samples to vote, and determine the final motion pattern by majority vote to ensure the stability of the determination.

[0142] Through the above network structure and data analysis steps, the LSTM network can achieve efficient processing of multi-dimensional time series data, with a motion pattern recognition accuracy of ≥98% and a real-time inference latency of ≤5ms, providing a reliable motion state basis for the dynamic inertial compensation layer to predict the inertial force at the next moment.

[0143] Example 15:

[0144] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation has the main technical contents described in any one of Examples 11 to 14. Furthermore, the motion modes include stationary, walking, jumping, in the air, and landing.

[0145] Example 16:

[0146] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of embodiments 11 to 15, further wherein the data monitored by the inertial measurement unit includes the three-axis acceleration of the object under test, the three-axis angular velocity of the object under test, the length of the suspension device rope, and the angle of the suspension device rope.

[0147] Example 17:

[0148] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation has the main technical contents described in any one of Examples 11 to 16. Furthermore, the preprocessing includes filtering, synchronization, and coordinate unification.

[0149] Example 18:

[0150] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Examples 11 to 17, further, the inertial force predicted at the next moment based on the monitored motion state is as follows:

[0151] (2)

[0152] In the formula, This represents the inertial force predicted for the next moment. This indicates the effective mass of the object under test in the suspension direction. This represents the acceleration of the object under test in the reverse direction of suspension.

[0153] Example 19:

[0154] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation has the main technical contents described in any one of Examples 11 to 18. Furthermore, the dynamic compensation force increment is equal in magnitude and opposite in direction to the predicted inertial force at the next moment.

[0155] Example 20:

[0156] The adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, the main technical contents of which are described in any one of Examples 11 to 19, further wherein the final control command is generated by vector superposition of the basic compensation force command and the dynamic compensation force increment, as shown below:

[0157] (3)

[0158] In the formula, For the final control command, , , These represent the sum of the components of the basic compensation force command and the dynamic compensation force increment in different directions, respectively.

[0159] Among them, the dynamic compensation force increment The directional components of the force are:

[0160] (4)

[0161] In the formula, if If positive, that is, horizontally to the right, then direction and Consistent, used to counteract inertial forces in that direction; similarly applicable to , .

[0162] The final target tensile force vector superposition calculation: due to the foundation compensation force command The force acts only along the suspension direction (Z-axis), with a horizontal component of zero. Therefore, the directional components of the final target tension are obtained by vector synthesis of the "basic compensation force component + dynamic compensation force component", as shown in the following formula:

[0163] (5)

[0164] Example 21:

[0165] See Figures 1 to 2 The main technical contents of the adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation include:

[0166] like Figure 1 As shown, the control system mainly comprises a three-layer control model: a high-level basic gravity compensation layer, a mid-level dynamic inertial compensation layer, and a low-level command execution layer. It can achieve multi-level control fusion from steady-state compensation to dynamic feedforward. The high-level control is primarily based on adaptive PID control using force gauge feedback, aiming to eliminate the steady-state error between the actual tension from the force sensor and the target compensation force. The mid-level control is mainly based on machine learning to predict inertial force and calculate the compensation force increment. Through a three-axis accelerometer and a three-axis gyroscope mounted on the test object, it collects the motion acceleration and angular velocity data in real time and outputs the dynamic compensation force increment. All sensor data undergoes filtering, synchronization, and coordinate unification processing to form a complete motion state input.

[0167] The system comprises a top-down three-layer structure:

[0168] 1. High-rise control: Foundation gravity compensation layer

[0169] This layer is the core steady-state compensation loop, primarily responsible for eliminating the system's steady-state error. Its inputs are the tension value measured by a tension sensor mounted on the suspension cable, and the simulated lunar surface gravitational acceleration based on the target. and the acceleration due to gravity on the Earth's surface The target compensation force, calculated based on the object's mass, is calculated using the following formula:

[0170]

[0171] This layer employs an adaptive PID controller, taking the deviation between the target compensation force and the measured tensile force as input. By adjusting the PID parameters in real time to adapt to different working conditions, it outputs a basic compensation force command to balance the static gravity of the test object. .

[0172] 2. Mid-level control: Dynamic inertia compensation layer

[0173] This layer is the core innovation, responsible for predicting and proactively compensating for inertial forces. Its inputs are triaxial acceleration and triaxial angular velocity data measured by an IMU fixed to the test subject, as well as rope length / angle information. Motion state recognition: First, the fused IMU and rope data are analyzed using a machine learning model, Long Short-Term Memory (LSTM), to identify the test subject's current motion pattern in real time, including: stationary, walking, jumping, in the air, and landing. Then, inertial force prediction: Based on the identified motion state and real-time acceleration data, the inertial force that may be generated at the next moment is predicted according to Newton's second law. The prediction formula is:

[0174] (2)

[0175] In the formula: The effective mass of the object in the suspension direction;

[0176] The predicted acceleration of the object in the opposite direction of suspension.

[0177] Final compensation force generation: This layer outputs a dynamic compensation force increment that is equal in magnitude and opposite in direction to the predicted inertial force. , The aim is to counteract inertial forces before they occur.

[0178] 3. Low-level control: Instruction execution layer

[0179] This layer is the instruction aggregation and execution unit. Its input is the basic compensation force instruction output from the higher layers. and the dynamic compensation force increment of the middle layer output The underlying controller combines the two to generate the final control command, driving the servo motor or other actuators to quickly and accurately output the target tension.

[0180] according to Figure 2 The steps are described as follows:

[0181] 201 Basic Compensation Force Command Generation:

[0182] The tension sensor measures the actual tension of the suspension rope in real time; simultaneously, it simulates gravitational acceleration based on the lunar surface. and the acceleration due to gravity on the Earth's surface Based on the mass of the test object, the target compensation force is calculated. The deviation between the actual tension and the target compensation force is input to the adaptive PID controller, which then calculates and outputs the basic compensation force command. It is used for steady-state compensation of static gravity.

[0183] 202 Dynamic Compensation Force Increment Prediction and Generation:

[0184] An IMU fixed to the test object collects triaxial acceleration and triaxial angular velocity data in real time. After filtering, synchronization, and coordinate unification preprocessing, the data is input into an LSTM network for motion state recognition to determine the current motion mode. Based on the recognized motion state and real-time acceleration information, the inertial force at the next moment is predicted according to Newton's second law. The calculation formula is:

[0185] = - Effective mass of the object in the suspension direction * Predicted acceleration of the object in the opposite direction of suspension; further generate dynamic compensation force increments that are equal in magnitude and opposite in direction to the predicted inertial force. This achieves feedforward compensation of 203 inertial forces.

[0186] 203 Instruction Fusion and Execution Control:

[0187] The basic compensation force command generated in step 201 Compared with the dynamic compensation force increment generated in step 202 The forces are superimposed to obtain the final total compensation force command; this command is then sent to the underlying actuator to drive it to output the target tension quickly and accurately, thereby achieving composite compensation for the gravity and inertial forces of the test object.

Claims

1. An adaptive suspension control system based on dynamic inertial prediction and hierarchical feedforward compensation, characterized in that, The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device.

2. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 1, wherein, The application relates to a gravity compensation method and device. （1） wherein represents the target compensation force; represents the mass of the object under test; represents the terrestrial surface gravity acceleration; represents the lunar surface simulated gravity acceleration.

3. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 1, wherein, The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device.

4. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 1, wherein, The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device.

5. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 4, wherein, The application relates to a gravity compensation method and device.

6. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 4, wherein, The application relates to a gravity compensation method and device.

7. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 4, wherein, The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application relates to a gravity compensation method and device. The application 8. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 1, wherein, The predicted inertia force of the next moment based on the monitored motion state is as follows: （2） wherein represents the predicted next moment inertial force; represents the effective mass of the object under test in the suspension direction; represents the acceleration of the object under test in the opposite direction of suspension.

9. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 1, wherein, The dynamic compensation force increment is equal in size and opposite in direction to the predicted inertia force of the next moment.

10. The adaptive suspension control system based on dynamic inertia prediction and hierarchical feedforward compensation of claim 1, wherein, The final control command is generated by vector superposition of the basic compensation force instruction and the dynamic compensation force increment, as follows: （3） In the formula, is the final control command, , , respectively represent the total force of the base compensation force command and the dynamic compensation force increment in different directions.

Images