A hole-finding control method and device for positioning and assembling a charging gun

Abstract

The present invention discloses a hole-finding control method and device for positioning and assembling a charging gun, and relates to the field of automation technology for charging equipment of new energy vehicles. The method obtains the real-time position and environmental perception data of the charging gun, and uses a Bayesian network to construct a contact state probability model to dynamically identify plane contact, chamfer contact, and in-hole centering state; combines the estimated values ​​of physical parameters to adaptively adjust the mass, damping, and stiffness matrices of the admittance control to generate target impedance characteristics; constructs a multi-physics field coupling model prediction framework to optimize the three-dimensional motion trajectory, and integrates dynamic admittance parameters to generate high-precision motion instructions. The present invention innovatively combines probabilistic state reasoning with multi-physics field coupling optimization, solving the technical problems of inaccurate contact state identification, rigid control parameters, and interference from multi-field coupling effects under complex working conditions, significantly improving the robustness and efficiency of charging gun assembly, while reducing dependence on high-cost hardware, and has important industrial application value.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy charging equipment, and in particular to a hole-finding control method and device for positioning and assembling a charging gun. Background Art

[0002] The current automated assembly of new energy vehicle charging guns relies primarily on high-precision visual positioning and rigid admittance control. This presents a bottleneck: insufficient accuracy in contact state recognition under complex contact conditions. It is difficult to accurately distinguish between planar contact, chamfered contact, and in-hole alignment in real time, leading to a mismatch in control strategies. Furthermore, traditional admittance control uses fixed impedance parameters, making it unable to adapt to sudden changes in contact stiffness and time-varying friction characteristics caused by aging charging sockets and surface contamination. Existing methods often employ a hybrid force / position control architecture, requiring the configuration of sophisticated hardware such as six-dimensional force sensors and lidar. This results in high system costs and weak anti-interference capabilities, making it difficult to meet assembly requirements under complex outdoor conditions. Summary of the Invention

[0003] (1) Technical issues to be resolved

[0004] To solve the above problems, the present invention proposes a hole-finding control method and device for positioning and assembling a charging gun, aiming to solve the problem in the prior art of insufficient accuracy in contact state recognition under complex contact conditions, difficulty in accurately distinguishing between plane contact, chamfered contact and in-hole centering states in real time, resulting in control strategy mismatch.

[0005] (2) Technical solution

[0006] A hole-finding control method for positioning and assembling a charging gun according to the present invention comprises:

[0007] Acquire real-time posture data and environmental perception data of the charging gun, including contact force / torque signals, three-dimensional point cloud information, and local contact pressure distribution;

[0008] Based on the posture data and environmental perception data, a Bayesian network is used to generate a probability distribution of contact states and physical parameter estimates between the charging gun and the charging socket, where the contact states include planar contact, chamfered contact, and in-hole centering contact.

[0009] Dynamically adjusting admittance control parameters according to the probability distribution and the estimated values ​​of the physical parameters to generate target impedance characteristics and dynamic admittance parameters, the admittance control parameters including a mass matrix, a damping matrix, and a stiffness matrix;

[0010] Constructing a multi-physics coupling model, optimizing the three-dimensional motion trajectory of the charging gun through the multi-physics coupling model predictive control framework, and generating optimized motion instructions by combining the dynamic admittance parameters and the target impedance characteristics;

[0011] The optimized motion instructions are sent to the execution end to complete the precise assembly of the charging gun and the charging socket.

[0012] Another hole-finding control device for positioning and assembling a charging gun according to the present invention comprises:

[0013] A perception unit is configured to obtain real-time posture data and environmental perception data of the charging gun end effector;

[0014] a state inference unit configured to generate a probability distribution of the contact state between the charging gun and the charging socket and an estimated value of a physical parameter through a probabilistic state inference model based on the posture data and the environmental perception data;

[0015] an admittance control unit configured to dynamically adjust admittance control parameters according to the probability distribution and the estimated value of the physical parameter to generate a target impedance characteristic;

[0016] A trajectory optimization unit is configured to optimize the three-dimensional motion trajectory of the charging gun through a model predictive control framework, and generate optimized motion instructions by combining dynamic admittance parameters and target impedance characteristics;

[0017] The execution unit is configured to execute the optimized motion instructions to complete the precise assembly of the charging gun and the charging socket.

[0018] Another computing device of the present invention comprises:

[0019] at least one processor; and

[0020] The memory stores instructions, which, when executed by the at least one processor, enable the at least one processor to execute the hole-finding control method for positioning and assembling a charging gun as described in any one of the above technical solutions.

[0021] Another non-transitory machine-readable storage medium of the present invention stores executable instructions, which, when executed, enable a machine to perform the hole-finding control method for positioning and assembling a charging gun as described in any one of the above technical solutions.

[0022] (3) Beneficial effects

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] In the present invention, a causal reasoning model of contact status is constructed by fusing multi-source perception data through a Bayesian network, thereby realizing probabilistic identification of plane contact, chamfer contact and in-hole alignment status. The accuracy of contact status classification is greatly improved, and the risk of control instability caused by misjudgment is significantly reduced.

[0025] The present invention adopts the collaborative optimization of dynamic admittance control and model predictive control framework to achieve adaptive adjustment of impedance parameters and maintain a small contact force tracking error under the working condition of fluctuation of the socket surface friction coefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 Schematic diagram of the process structure of the control method;

[0028] Figure 2 This is a thermal diagram of the pressure distribution on the contact surface between the charging gun and the socket (ideal state);

[0029] Figure 3 This is a thermal diagram of the pressure distribution on the contact surface between the charging gun and the socket (abnormal state);

[0030] Figure 4 Schematic diagram of the curve of the Bayesian network state probability distribution;

[0031] Figure 5 This is a schematic diagram of the unit module structure of the control device in Example 2;

[0032] Figure 6 Schematic diagram of the framework structure of the execution equipment.

[0033] 1. Processor, 2. Memory, 3. Communication interface, 4. Communication bus. DETAILED DESCRIPTION

[0034] This technical solution addresses the challenge of accurate hole-finding when a robot carrying a charging gun performs a charging task. It proposes an innovative approach that combines probabilistic state reasoning with dynamic admittance control. This approach significantly reduces hardware dependency and improves system robustness through software optimization. The core of this disclosure lies in using Bayesian networks to explicitly model the causal chain and uncertainty propagation paths of contact physics, and inferring contact states in real time based on multi-sensor data.

[0035] Example 1

[0036] like Figure 1-Figure 3 The hole-finding control method for positioning and assembling a charging gun shown includes the following steps:

[0037] S100: Acquire real-time posture data and environmental perception data of the charging gun, where the environmental perception data includes contact force / torque signals, three-dimensional point cloud information, and local contact pressure distribution.

[0038] To achieve the acquisition of real-time posture data and environmental perception data of the charging gun, it is necessary to replace hardware dependence through multimodal data fusion and algorithm modeling. The real-time posture data of the charging gun is jointly calculated by the visual SLAM module and the inverse kinematics of the robotic arm. In particular, the visual SLAM module can be set on the charging gun or the robotic arm. In this disclosure, the visual SLAM module obtains continuous frame RGB images from the monocular camera on the charging gun, uses ORB feature point extraction and optical flow tracking algorithm, and combines the extended Kalman filter (EKF) to estimate the position of the charging gun in the charging pile coordinate system. With posture The robot arm's inverse kinematics model inputs joint encoder data and calculates the charging gun's end-effector pose using the DH parameter matrix. This is then weightedly fused with the visual SLAM results to improve positioning accuracy and ensure robustness against camera occlusion or sudden changes in illumination.

[0039] Specifically, the posture transformation matrix of the charging gun end effector calculated by the DH parameter matrix is: , where each link transformation matrix is Defined by DH parameters: , is the connecting rod length, indicating the distance from the current connecting rod to the arrive axis The distance between the axes, is the connecting rod angle, indicating the rotation around Axis Rotate to Angle, is the connecting rod offset, indicating that Axis Rotate to For example, a joint parameter of the robot arm is , , , ,but Describes the position and attitude of the link relative to the previous link.

[0040] The weighted fusion expression of the charging gun end effector pose and visual SLAM result is: Visual SLAM output pose , DH model calculates pose , the position fusion is , is the position output by the visual SLAM, which is calculated by tracking the feature points in the RGB image captured by the camera. The position calculated by the DH model is solved by inverse kinematics based on the robot arm joint encoder data. The posture fusion is ,in They are the quaternion poses output by the visual SLAM and DH models respectively. Represents the weight coefficient, which represents the trust of the visual SLAM and DH models respectively, satisfying , in this embodiment , Represents the spherical linear interpolation function to avoid the singularity problem of Euler angle fusion. Indicates the fused quaternion attitude, which can be converted into Euler angles The final fusion pose is expressed as .

[0041] The software estimation of contact force / torque signals is based on virtual force sensor technology. The contact mechanics relationship between the charging gun head and the charging socket is modeled using Hertz contact theory, and the input is the posture deviation time series data. , establish the nonlinear equation , where the stiffness matrix The calibration error is ≤0.3N through offline calibration experiment fitting. The lightweight LSTM network further optimizes force estimation, inputs the pose deviation sequence, and outputs the six-dimensional force / torque. , the mean square error of the test set is 1.5N², meeting the real-time requirements.

[0042] 3D point cloud information is achieved through monocular depth estimation and geometric registration. The MobileDepth network generates a 640×480 resolution depth map from a single RGB image frame and is trained jointly using a multi-scale photometric loss (SSIM) with a gradient smoothness constraint. The point cloud data is aligned with the charging socket CAD model using an iterative closest point (ICP) algorithm. Initial alignment uses SVD decomposition to solve the rotation and translation matrices. The iteration termination criteria are a maximum of 50 iterations or a residual change rate of <0.1%. For low-light scenarios, the CycleGAN image enhancement module is integrated. The generator uses a U-Net structure and the discriminator is PatchGAN. The training dataset contains 5,000 normal-low-light image pairs. The PSNR of the output images is improved to 28dB, ensuring the reliability of 3D reconstruction in complex lighting conditions.

[0043] The local contact pressure distribution is inferred by the tactile simulation model. Based on the simplified contact mechanics model of finite element analysis (FEA), the contact area between the charging gun head and the socket is discretized into 1mm×1mm grid cells, and the contact force estimate is input. Calculate the pressure value of each grid node in real time based on the posture data in, is the local stiffness coefficient, The pressure distribution heat map is rendered by OpenGL, with color mapping from blue to red, indicating that the pressure increases from small to large, which can assist in diagnosing unilateral overload or uneven contact. Figure 2-Figure 4 As shown, it represents the local pressure distribution heat map of the charging gun head and the socket. The black circle represents the contact area between the pin and the socket, with a radius of 80 pixels. Figure 2 The pressure in the local center area is the highest. If the center of this area is the ideal contact area, it means that the contact is in an ideal state. Figure 3 The middle red area is concentrated in the lower right corner, and the secondary pressure point appears in the upper left corner. Deviating from the ideal area indicates unilateral overload or uneven contact, requiring adjustment of the robotic arm to adjust the position of the charging gun or cleaning the contact surface.

[0044] For complex contact conditions, an elastic-plastic deformation correction term is introduced based on the Hertz contact theory. When the contact pressure exceeds the material yield strength, for example, the aluminum alloy socket is set to 250MPa, the contact area calculation is corrected to: in, is the yield load, is the plastic deformation coefficient, in this disclosure is 0.12.

[0045] S200. Based on the posture data and environmental perception data, generate a contact state probability distribution and physical parameter estimation values ​​between the charging gun and the charging socket through a Bayesian network, where the contact state includes plane contact, chamfer contact, and in-hole centering contact.

[0046] This paper uses Bayesian network to dynamically model the causal relationship between the contact state of the charging gun and the charging socket, and combines multi-source sensor data to infer the contact probability distribution and physical parameters in real time, providing a decision basis for core control. The original contact force / torque signal is obtained from the six-dimensional force sensor, and the Kalman filter noise reduction process is used to eliminate high-frequency electromagnetic interference and mechanical vibration noise. After normalization, the standardized force vector is obtained. and torque vector At the same time, based on the visual point cloud data, the real-time point cloud is aligned with the charging socket CAD model through the ICP registration algorithm to output the contact point coordinates The terminal position data is converted to the charging gun coordinate system through the kinematic model to calculate the relative offset. , which is input into the Bayesian network as a priori estimate of the contact point location.

[0047] Specifically, the core nodes of the Bayesian network include three types of contact states: plane contact, chamfer contact, and in-hole centering contact, as well as Bayesian network nodes for key physical parameters such as contact stiffness, friction coefficient, and contact point coordinates. The contact states are discrete nodes, and the physical parameters are continuous nodes. The process of constructing Bayesian network nodes begins with a physical analysis of the contact scenario between the charging gun and the charging socket. Combined with the actual needs of multi-sensor data, the node types are defined hierarchically and causal relationships are established. First, three types of discrete contact states are identified as core parent nodes. Their classification is based on the physical laws of contact force distribution and geometric characteristics. For example, the normal force-dominant characteristic of plane contact is supported by Hertz contact theory, the lateral force and torque relationship of chamfer contact is determined by the geometric constraints of the chamfer inclination angle, and the in-hole centering state is defined by the characteristics of stiffness mutation and lateral force tending to zero.

[0048] On this basis, continuous sub-nodes are introduced to describe key physical parameters. Contact stiffness reflects the relationship between material elasticity and contact area, the friction coefficient characterizes the dynamic friction behavior between contact surfaces, and the contact point coordinates are located by fusion of visual and force data. The definitions of these physical parameter nodes do not exist in isolation, but form bidirectional constraints through physical equations and observation data to ensure that the conditional dependency between nodes conforms to the theoretical model and can cooperate with actual sensor input. In this disclosure, the physical equations include the contact stiffness formula: , The unit of elastic modulus of a material is , such as ABS plastic , the ability of the reaction material to resist deformation, Represents the Poisson's ratio of the material, dimensionless, such as plastic , used to describe the ratio of the transverse strain to the longitudinal strain of the material, Represents the contact area in units of , calculated by Hertz contact theory, and the indentation depth and contact radius In particular, this formula is used to quantify the rigidity characteristics of the contact area under the in-hole centering state, which directly affects the adjustment of the rigidity parameters of the admittance control; the torque relationship formula of the chamfer contact is: , Represents the moment around the y-axis in units of , caused by the lateral contact force, Indicates the lateral offset of the contact point relative to the center of the socket, in units , Indicates the chamfer inclination angle, in radians, which is geometrically defined as the angle between the chamfered surface and the axis of the socket. It represents the tangent value of the chamfer inclination angle, reflecting the ratio of the lateral force to the normal force. In particular, this formula is used to associate the chamfer inclination angle with the torque through geometric constraints, that is, it is used to calibrate the conditional probability of the chamfer contact state; the normal force formula for planar contact is: , Represents the normal contact force in units of , the force component perpendicular to the contact plane, Represents the equivalent radius of the contact area in units of , determined by the geometric matching between the charging gun head and the socket surface, Indicates the indentation depth in units of , reflecting the contact deformation. This expression is used to explain the physical mechanism in which the normal force dominates under the plane contact state and provides a theoretical basis for conditional probability. The dynamic update formula of the friction coefficient is , represents the dynamic friction coefficient, dimensionless, initial value , Indicates the sliding friction coefficient in units of , used to characterize friction energy loss, Represents the normal force in units of ,Right now , Represents sliding displacement in units of , Represents the learning rate, dimensionless, in this disclosure , which controls the parameter update step size. Specifically, this formula dynamically corrects the friction coefficient through online learning to account for changes in friction characteristics caused by aging or contamination on the socket surface. It should be noted that in addition to the four aforementioned physical equations, constraints can also be added to the lateral force ratio and a visual positioning model for contact point coordinates. In other words, the physical equations are not limited to the four aforementioned ones and can be further constrained based on actual needs, thereby improving overall fit accuracy.

[0049] The calibration of conditional probability is the core of network construction and requires the integration of prior physical knowledge and experimental data. For contact state nodes, their conditional probabilities are determined by combining theoretical derivation with statistical learning. For example, the probability threshold of the normal force exceeding 80% in the plane contact state is derived from the quantitative relationship between indentation depth and contact area in Hertz contact theory. At the same time, the statistical significance of this threshold is verified based on multiple sets of offline experimental data. The lateral force ratio range of chamfer contact is determined by the tangent function of the chamfer inclination angle. Defined by the linear coefficient of torque and offset Further constraints, in particular, the linear coefficients of the moment and offset Related to the elastic modulus of the material, in this disclosure, the lateral force ratio range is 0.3-0.6. The determination of the in-hole centering state depends on the stiffness threshold and the lateral force zeroing characteristic. Its probability distribution is calibrated by the material mechanical parameters and the measured contact reaction force. In this disclosure, the stiffness threshold is .like This indicates that the charging gun tip has fully contacted the socket hole wall, and the contact area is significantly increased. For example, the contact area is larger when the hole is centered. The geometric projection area is close to the socket aperture. At this time, the elastic modulus of the material is The interaction with the contact geometry causes a jump in stiffness, leading the system to identify an "in-hole alignment" state. The conditional distribution of the physical parameter nodes also adheres to dual constraints: the contact stiffness follows a lognormal distribution to reflect the uncertainty of material properties, the friction coefficient adopts a truncated Gaussian distribution to constrain its reasonable range, and the prior distribution of the contact point coordinates is constructed using a Gaussian model based on visual positioning residuals to ensure the statistical reliability of spatial positioning.

[0050] At the initial stage of Bayesian network inference, at least 1000 particles are generated, each particle contains randomly initialized contact state, contact stiffness, friction coefficient and contact point coordinates, etc. The next moment position is predicted based on the robot kinematic model. For example, if the current state is chamfer contact, according to the chamfer inclination angle Update contact point coordinate offset ,in is the insertion speed, The actual observation data is compared with the particle prediction value to calculate the likelihood probability of each particle. For example, for the force ratio, assuming the actual observation value is 0.5, the particle prediction value is The weights are Gaussian distributed Calculated as . Visual localization residual The same weight calculation is involved. The smaller the residual, the higher the weight. A system resampling strategy is used to retain high-likelihood particles and eliminate low-weight particles according to the weight ratio. Finally, the proportion of each contact state in the particle set is counted as the posterior probability, such as 65% for plane contact, and the mean and confidence interval of the physical parameters are calculated. Figure 4 The figure shows a schematic curve diagram of the Bayesian network state probability distribution. The yellow highlighted area around 1.2 seconds shows the friction coefficient update event. After the update, the chamfer contact probability decreases by 8%, and the in-hole centering probability increases by 20%. The probability curve shows a non-smooth transition at the event point, reflecting the online learning process.

[0051] Friction coefficient Dynamic correction is made through the sliding window online learning mechanism. After each insertion task is completed, the force / displacement data of the contact phase is extracted and the friction energy loss model is used to calculate the force / displacement data. Calculate current Gradient , according to the learning rate Update. Contact stiffness According to Hertz contact theory, real-time calibration is performed. If the relationship between normal force and indentation depth deviates from the theoretical value, If the value exceeds 10%, the stiffness recalibration is triggered and the new value is fitted by the least square method. Values ​​are estimates.

[0052] S300 , dynamically adjusting admittance control parameters according to the probability distribution and the estimated values ​​of the physical parameters to generate target impedance characteristics and dynamic admittance parameters, wherein the admittance control parameters include a mass matrix, a damping matrix, and a stiffness matrix.

[0053] The probability distribution of the contact state between the charging gun and the charging pile and the estimated values ​​of the physical parameters are obtained in real time through the probabilistic state inference model. Based on the probability distribution output by the Bayesian network, the dynamic adjustment of the admittance control parameters is divided into two layers of logic: the upper rule base maps the target impedance characteristics according to the contact state confidence. For example, when the chamfer contact probability exceeds 60%, the target stiffness matrix Adjust the default value from 250 N / m to 278 N / m to reduce contact reaction force fluctuations. If the confidence interval span of the friction coefficient is detected to be too large, the online identification of the friction model is triggered, and the actual friction force curve is fitted using the least squares method to update the dynamic parameters of the LuGre friction model. When the confidence interval span of the friction coefficient is greater than 0.1, the LuGre friction model parameter update is triggered: ,in is the pre-sliding dynamics formula, is the friction formula, For mane deformation, is the sliding speed, is a dynamic parameter, which is updated by fitting the actual friction curve using the least squares method. The LuGre model is a classic method for friction modeling and will not be elaborated on here.

[0054] The underlying model predictive control (MPC) adjusts the quality matrix in real time using a rolling optimization method With the damping matrix , the objective function integrates force tracking error, trajectory smoothness and energy consumption weight, and the specific form is: in, is the expected contact force, is the expected trajectory, the weight matrix Dynamic adjustment based on contact state confidence, for example, when When, increase The force tracking weight is used to suppress high-frequency oscillations, while the constraints are embedded in the robot dynamics equation and the estimated stiffness of the environment. The optimization problem is implemented using the OSQP solver, which combines CUDA acceleration to compress the single-step solution time to less than 6ms, ensuring real-time synchronization between parameter adjustment and robot movement. , then reduce the trajectory smoothing weight to prioritize contact force tracking, where It is the chamfer contact probability threshold obtained through Bayesian network particle filtering statistics, that is, the proportion of particles marked as "chamfer contact" in the statistical particle set.

[0055] Weight Matrix Each element of corresponds to the penalty weight for different degrees of freedom to control input changes. The value will increase significantly The value of the term forces the optimization algorithm to select a control input sequence with a smaller change range. This can effectively reduce the instantaneous load of the robot arm joint, reduce motor energy consumption and mechanical wear, for example, when the charging gun approaches the socket, it can avoid contact force overshoot caused by sudden speed changes. The rate of change of the weight matrix Indirectly affects the smoothness of the trajectory. The value will make the trajectory transition smoother and reduce the sudden change of acceleration, thereby reducing the vibration of the end effector, ensuring the stable contact force between the charging gun and the socket during the hole-finding process, and avoiding positioning errors caused by jitter. The value of the force tracking error weight and trajectory deviation weight For example, if the system has extremely high requirements for force control accuracy, If it is larger, it needs to be appropriately reduced The weight of the control input is increased to allow for greater changes in control input and ensure rapid response to fluctuations in contact force. On the contrary, if energy consumption and mechanical life are prioritized, If the value is larger, a slightly higher force tracking error or trajectory deviation must be accepted.

[0056] At the physical level, the dynamic adjustment of the admittance parameters is performed through Lyapunov stability analysis to ensure global asymptotic stability. Define the Lyapunov function ,in is the trajectory tracking error, is the stiffness parameter deviation, and the adaptive law is designed as , by solving the Lyapunov equation Ensure that the system is stable when stiffness changes suddenly.

[0057] To achieve the above process, the system relies on the efficient fusion of perception data. The virtual force sensor module combines the posture deviation time series data with the pre-calibrated stiffness matrix. The lightweight TinyLSTM network estimates the six-dimensional contact force within a 20ms time window. Mapped to The inference latency is kept under 1ms. The monocular vision module generates a dense depth map through the MobileDepth network and aligns the point cloud with the charging pile CAD model using the ICP registration algorithm, providing accurate environmental geometric feature input for the Bayesian network.

[0058] The dynamic admittance parameters are sent to the robot controller in real time through the edge computing platform. The high-frequency force control thread is based on the updated 、 、 The matrix generates compensation speed instructions. For example, in the chamfer contact phase, if the lateral force ratio is detected to 30 Ns / m to suppress the sliding tendency, the mass matrix is ​​reduced. to 0.15 kg·m² to improve response speed.

[0059] S400 , constructing a multi-physics field coupling model, optimizing the three-dimensional motion trajectory of the charging gun through the multi-physics field coupling model predictive control framework, and generating optimized motion instructions by combining the dynamic admittance parameters and the target impedance characteristics.

[0060] By constructing a multi-physics coupling model and integrating the mechanical, electromagnetic, and thermal interactions between the charging gun and the charging station during contact, the system achieves precise prediction and control of the three-dimensional motion trajectory. In the mechanical field, a nonlinear relationship between contact force and displacement is established based on Hertz contact theory. Finite element analysis is used to calculate the stress distribution in the contact area. Combined with real-time feedback from the six-dimensional force sensor on the charging gun head, the system dynamically corrects the contact stiffness and friction coefficient.

[0061] The electromagnetic field model considers the current conduction characteristics of the charging interface, solving for contact resistance and electromagnetic interference using Maxwell's equations to ensure current stability during charging. The thermal model uses the heat conduction equation to predict the temperature rise in the contact area, avoiding thermal expansion and deformation of the material caused by prolonged contact or high current. Multi-physics coupling is synchronized and updated through an iterative solver. For example, during the insertion of the charging gun, changes in contact force can cause instantaneous fluctuations in contact resistance, which in turn affects the local temperature distribution. The model adjusts the prediction parameters through real-time data feedback to ensure accurate modeling of the synergistic effects between the various physical fields.

[0062] The predictive control framework, centered on a multiphysics coupling model, employs a rolling-horizon optimization strategy to generate three-dimensional motion trajectories. Within each control cycle, the model predicts the contact state several steps into the future and constructs an objective function to minimize trajectory deviation, force tracking error, and energy consumption. The contact state includes contact force, displacement deviation, and temperature variation.

[0063] For example, the objective function can be defined as: The optimization problem is embedded in the robot's dynamic constraints and the estimated stiffness of the environment. The calculation is completed by an efficient QP solver, and the optimized pose increment instructions are output.

[0064] The combination of dynamic admittance parameters and target impedance characteristics enables refined control of the contact process. The admittance parameters are adjusted in real time based on the prediction results of the multi-physics model, where the admittance parameters include mass , damping , stiffness For example, when a sudden increase in contact resistance is detected, indicating poor contact, the admittance stiffness is reduced to minimize reaction force fluctuations. When the predicted temperature approaches a threshold, the damping coefficient is increased to slow the insertion speed and avoid sticking caused by thermal expansion. The target impedance characteristic is designed through Lyapunov stability analysis to ensure that the system remains asymptotically stable under parameter changes or external disturbances. Specifically, the admittance control equation is expressed as: ,in is the pose error, dynamic parameter 、 、 Updates are performed via online recognition algorithms, such as recursive least squares, and are collaboratively optimized with model predictions.

[0065] The charging gun integrates a miniature force / torque sensor, temperature sensor, and current monitoring module to collect contact force, temperature, and current data in real time, transmitting it to the edge computing unit via a high-speed bus. The charging pile has a preset database of geometric features, such as chamfer angles and inner wall curvature, and communicates with the charging gun via radio frequency identification (RFID) or visual tags to provide prior information on socket parameters. For example, when the charging gun detects that the socket is marked as an "aging model," it automatically calls the friction coefficient model from historical data to optimize the admittance control parameters. The robotic arm serves only as an actuator, receiving motion commands from the charging gun control unit to ensure a high-bandwidth position and force control closed loop.

[0066] S500: Send the optimized motion instruction to the execution end to complete the precise assembly of the charging gun and the charging socket.

[0067] The optimized motion instructions achieve precise assembly of the charging gun and the charging socket through the deep integration of multimodal perception and dynamic control parameters. During the contact phase, the virtual force sensor estimates the six-dimensional contact force in real time based on the end-position deviation time series data and the Hertz contact theory model. Specifically, the end-position deviation is obtained by inverse kinematics calculation of the manipulator, combined with the offline calibration stiffness matrix and friction coefficient , using nonlinear least squares optimization to solve the contact force The lightweight TinyLSTM network further extracts features from the pose deviation time series window and outputs contact force estimates, effectively suppressing noise caused by electromagnetic interference.

[0068] The monocular 3D reconstruction module uses the MobileDepth network to generate a dense depth map from a single RGB image frame. It employs a multi-scale SSIM photometric loss and a gradient smoothing loss to optimize depth estimation accuracy. The reconstructed point cloud is registered with the socket CAD model using the ICP algorithm. Iterations terminate when the residual change rate is less than 0.1% or after a maximum of 50 iterations, providing precise initial position corrections for motion commands. In low-light scenarios, the CycleGAN generator enhances the input image, increasing the PSNR to 28.6dB, ensuring robust 3D reconstruction.

[0069] The Bayesian network infers the probability distribution of contact states through sequential Monte Carlo particle filtering, and the node conditional probability is calibrated by experimental data and static equations. For example, the lateral force ratio in the "chamfer contact" state The probability is 85%, the friction coefficient The confidence interval is updated online through a sliding window, and the estimation error is reduced from 0.12 to 0.04. The network outputs the posterior probability and the environmental stiffness , providing a basis for dynamic parameter adjustment for admittance control, such as the inner wall contact probability of 72% and .

[0070] The admittance control module generates target impedance parameters based on state probability. For example, when the inner wall contact probability is >50%, the stiffness , damping The model predictive control (MPC) framework uses the objective function to optimize the trajectory, the weight matrix 、 、 Smooth trajectories are generated using the OSQP solver. The Lyapunov function ensures global stability during the parameter adaptation process, and the standard deviation of the force tracking error is reduced from 3.1N to 0.8N under sudden stiffness changes.

[0071] The execution end receives optimized motion commands and implements high-frequency control via the edge computing platform. As the robotic arm moves along the planned trajectory, the contact stress distribution is calculated in real time using the PyBullet finite element model. A thermal map dynamically displays the peak pressure areas, triggering fine-tuning of the trajectory to avoid mechanical damage.

[0072] Example 2

[0073] like Figure 5The hole-finding control device for positioning and assembling a charging gun is shown. The hole-finding control device for positioning and assembling a charging gun achieves precise operation through the collaboration of multiple modules. The perception unit uses the force sensor module and the visual reconstruction module to obtain real-time data, wherein the force sensor module generates contact force / torque estimates through the time series data of the posture deviation of the charging gun end and the contact mechanics model. Specifically, the posture deviation data includes the tiny displacement and angular offset of the charging gun end relative to the charging socket. Combined with the stiffness matrix and friction model based on Hertz contact theory, a nonlinear equation is constructed. , real-time estimation of six-dimensional contact force / torque. For example, when the charging gun head contacts the chamfer of the socket, the lateral force With normal force The ratio will change with the contact point offset Significant changes have been made. A lightweight LSTM network extracts features from temporal pose deviations and ultimately outputs force estimates. The mean squared error on the test set is kept within 1.5 N², meeting both real-time and precision requirements. The visual reconstruction module captures RGB images using a monocular camera, generates a dense depth map using the MobileDepth network, and then uses the Iterative Closest Point (ICP) algorithm to align the real-time point cloud with the CAD model of the charging socket to correct positioning errors. For example, in low-light scenarios, the CycleGAN module enhances the input image, increasing the PSNR to 28 dB and ensuring the reliability of 3D reconstruction. The fusion of visual and force data not only compensates for the limitations of a single sensor but also verifies the accuracy of contact point coordinates through registration residuals, providing high-confidence input for state reasoning.

[0074] The state reasoning unit constructs a Bayesian network through the state modeling module, defines the contact state and physical parameters as nodes, and establishes conditional dependencies based on physical equations and sensor data. The contact state includes plane contact, chamfer contact, and hole alignment, and the physical parameters include contact stiffness and friction coefficient. For example, the determination of the chamfer contact state depends on the lateral force ratio. and torque The linear relationship between the two is established, while the in-hole centering state is defined by the stiffness threshold and the lateral force zeroing characteristic. The probability reasoning module uses a particle filter algorithm to initialize 1000 particles and dynamically update the weights based on the observed data, and finally calculates the posterior probability distribution. When , it indicates that the charging gun is in the chamfer guidance stage, at this time the system prioritizes adjusting the admittance parameters to suppress lateral sliding. The confidence interval of the physical parameters is dynamically modified through the sliding window online learning mechanism, for example, using the friction energy loss model Gradients are calculated and parameters updated to ensure adaptability to aging or contamination of the socket surface.

[0075] The admittance control unit dynamically adjusts the mass, damping and stiffness matrices according to the state inference results. For example, in the chamfer contact stage, the stiffness matrix Increased from 250N / m to 278N / m to enhance lateral restraint, while optimizing trajectory smoothness and force tracking error through model predictive control (MPC) rolling. In the weight matrix Dynamic adjustment with contact state, increases when centered in the hole To suppress oscillation, reduce the chamfer contact probability when it is high To prioritize force control accuracy. The optimization problem is solved in 6ms in a single step by combining OSQP solver with CUDA acceleration, ensuring real-time synchronization of parameters and instructions at 100Hz control frequency. Lyapunov stability analysis further ensures the global asymptotic stability of the adjustment process, defining the function , through the adaptive law Ensure that the system is stable when the stiffness parameters change suddenly.

[0076] The trajectory optimization unit and the execution unit finally transform the theoretical calculation into physical action. The optimized motion instructions are sent to the robot controller via the EtherCAT bus with a cycle of 1ms. The high-frequency force control thread generates the compensation speed instruction based on the updated admittance parameters. For example, in the chamfer contact stage, the damping matrix Increased from 30Ns / m to 35N / m to suppress the sliding tendency, while the mass matrix The force is reduced to 0.15 kg / m² to improve response speed. The actuator uses joint servo motors to drive the charging gun along an optimized trajectory, with positioning accuracy guaranteed by a closed-loop vision-force fusion. The core innovation of the entire system lies in the deep integration of environmental perception, state reasoning, and admittance control into the interaction model between the charging gun and the charging station. The robotic arm serves only as a high-precision actuator, ultimately achieving robust assembly in complex scenarios.

[0077] Example 3

[0078] An embodiment of the present invention provides a computer-readable storage medium.

[0079] The computer-readable storage medium provided in the embodiment of the present invention stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above-mentioned hole-finding control methods for positioning and assembling a charging gun can be implemented.

[0080] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0081] For an introduction to the computer-readable storage medium provided in an embodiment of the present invention, please refer to the above method embodiment, and the present invention will not elaborate on it here.

[0082] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0083] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0084] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0085] Example 4

[0086] An embodiment of the present invention provides an execution device.

[0087] Please refer to Figure 6 , Figure 6 This is a schematic diagram of the structure of an execution device provided by the present invention, which may include:

[0088] memory for storing computer programs;

[0089] The processor, when used to execute the computer program, can implement the steps of any one of the above-mentioned hole-finding control methods for positioning and assembling a charging gun.

[0090] like Figure 6 FIG2 is a schematic diagram of the structure of the execution device, which may include a processor 5, a memory 6, a communication interface 7, and a communication bus 8. The processor 5, the memory 6, and the communication interface 7 communicate with each other via the communication bus 8.

[0091] In the embodiment of the present invention, the processor 5 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array, or other programmable logic devices.

[0092] The processor 5 may call the program stored in the memory 6. Specifically, the processor 5 may execute the operations in the embodiment of the push button switch fault detection method.

[0093] The memory 6 is used to store one or more programs. The programs may include program codes, and the program codes include computer operating instructions. In the embodiment of the present invention, the memory 6 stores at least a program for implementing the following functions:

[0094] Acquire real-time posture data and environmental perception data of the charging gun, including contact force / torque signals, three-dimensional point cloud information, and local contact pressure distribution;

[0095] Based on the posture data and environmental perception data, a Bayesian network is used to generate a probability distribution of contact states and physical parameter estimates between the charging gun and the charging socket, where the contact states include planar contact, chamfered contact, and in-hole centering contact.

[0096] Dynamically adjusting admittance control parameters according to the probability distribution and the estimated values ​​of the physical parameters to generate target impedance characteristics and dynamic admittance parameters, the admittance control parameters including a mass matrix, a damping matrix, and a stiffness matrix;

[0097] Constructing a multi-physics coupling model, optimizing the three-dimensional motion trajectory of the charging gun through the multi-physics coupling model predictive control framework, and generating optimized motion instructions by combining the dynamic admittance parameters and the target impedance characteristics;

[0098] The optimized motion instructions are sent to the execution end to complete the precise assembly of the charging gun and the charging socket.

[0099] In one possible implementation, the memory 6 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function, etc.; the data storage area may store data created during use.

[0100] In addition, the memory 6 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0101] The communication interface 7 may be an interface of a communication module, used for connecting to other devices or systems.

[0102] Of course, it needs to be explained that Figure 6 The structure shown does not constitute a limitation on the execution device in the embodiment of the present invention. In actual applications, the execution device may include Figure 6 More or fewer components than shown, or combinations of certain components.

[0103] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Any modifications and improvements made to the technical solution of the present invention by a person of ordinary skill in the art without departing from the design concept of the present invention shall fall within the scope of protection of the present invention. The technical content for which protection is sought in the present invention is fully set forth in the claims.

Claims

1. A hole-finding control method for positioning and assembling a charging gun, characterized in that: include: Acquire real-time posture data and environmental perception data of the charging gun, including contact force / torque signals, three-dimensional point cloud information, and local contact pressure distribution; Based on the posture data and environmental perception data, a Bayesian network is used to generate a probability distribution of contact states and physical parameter estimates between the charging gun and the charging socket, where the contact states include planar contact, chamfered contact, and in-hole centering contact. Dynamically adjusting admittance control parameters according to the probability distribution and the estimated values ​​of the physical parameters to generate target impedance characteristics and dynamic admittance parameters, the admittance control parameters including a mass matrix, a damping matrix, and a stiffness matrix; Constructing a multi-physics coupling model, optimizing the three-dimensional motion trajectory of the charging gun through the multi-physics coupling model predictive control framework, and generating optimized motion instructions by combining the dynamic admittance parameters and the target impedance characteristics; The optimized motion instructions are sent to the execution end to complete the precise assembly of the charging gun and the charging socket.

2. The hole-finding control method for positioning and assembling a charging gun according to claim 1, characterized in that: The step of generating the contact state probability distribution includes: Collect sensor data when the charging gun contacts the charging socket. The sensor data includes six-dimensional force / torque timing signals, three-dimensional point cloud data acquired by an RGB-D camera, and local pressure distribution acquired by a tactile array. Construct a Bayesian network that includes contact state type, contact point coordinates, environmental stiffness, and friction coefficient. Perform state inference using a particle filter algorithm to output the posterior probability distribution of the contact state and parameter confidence intervals. The contact state types include at least plane contact, single-point chamfer contact, double-point chamfer contact and inner wall contact.

3. The hole-finding control method for positioning and assembling a charging gun according to claim 2, characterized in that: The construction and reasoning of the Bayesian network include: Define a node network including contact state type, contact point spatial relationship and physical parameters; Calibrate the conditional probability relationship between nodes based on physical constraints and historical data; The posterior distribution of contact states and parameter estimation results are output through a probabilistic inference algorithm, and the network weights are dynamically updated to adapt to environmental changes. The conditional probability of the Bayesian network is calibrated as follows: Establishing a correlation between contact state and observed variables based on static equations, including the lateral force / torque ratio, contact point geometry, and friction coefficient range; Node weights are dynamically updated through a sliding window online learning mechanism, and model correction is triggered when KL divergence detects parameter distribution drift.

4. The hole-finding control method for positioning and assembling a charging gun according to claim 1 or 3, characterized in that: The dynamic adjustment of admittance control parameters includes: Generate target stiffness and damping parameters based on contact state probability distribution; The parameter adaptation mechanism is designed through Lyapunov stability analysis to ensure the asymptotic stability of the control loop under sudden stiffness changes; A dynamic friction compensation model is introduced to correct the nonlinear interference component in the control instructions.

5. The hole-finding control method for positioning and assembling a charging gun according to claim 4, characterized in that: The multi-physics coupling model integrates contact mechanics, friction effects, and trajectory dynamics models. The optimization objective function of the multi-physics coupling model predictive control framework is: in, is the expected contact force, is the expected trajectory, the weight matrix Dynamic adjustment is made based on the confidence level of the contact state, while the constraints are embedded in the manipulator dynamics equations and the estimated stiffness of the environment; A real-time solver is used to generate smooth trajectories that satisfy the dynamic constraints of the robot arm.

6. The hole-finding control method for positioning and assembling a charging gun according to claim 5, characterized in that: The generating of optimized motion instructions by combining the dynamic admittance parameter and the target impedance characteristic includes: Based on the mass matrix, damping matrix and stiffness matrix in the dynamic admittance parameters, the admittance control equation is constructed to convert the real-time contact force / torque signal into position compensation; The force-displacement response relationship is defined based on the target impedance characteristics, and an optimization algorithm is used to balance the force tracking error and trajectory smoothness to generate a motion trajectory that meets the dynamic constraints of the manipulator. Combining the admittance control equation with the optimized motion trajectory, the motion instructions are dynamically adjusted through a closed-loop feedback mechanism to ensure the flexibility and positioning accuracy of the charging gun during the contact process.

7. A hole-finding control device for positioning and assembling a charging gun, characterized in that: include: A perception unit is configured to obtain real-time posture data and environmental perception data of the charging gun end effector; a state inference unit configured to generate a probability distribution of the contact state between the charging gun and the charging socket and an estimated value of a physical parameter through a probabilistic state inference model based on the posture data and the environmental perception data; an admittance control unit configured to dynamically adjust admittance control parameters according to the probability distribution and the estimated value of the physical parameter to generate a target impedance characteristic; A trajectory optimization unit is configured to optimize the three-dimensional motion trajectory of the charging gun through a model predictive control framework, and generate optimized motion instructions by combining dynamic admittance parameters and target impedance characteristics; The execution unit is configured to execute the optimized motion instructions to complete the precise assembly of the charging gun and the charging socket.

8. The hole-finding control device for positioning and assembling a charging gun according to claim 7, characterized in that: The sensing unit includes: The sensing unit includes: A force sensor module is configured to generate contact force / torque estimates based on the charging gun terminal posture deviation time series data and the contact mechanics model; a visual reconstruction module configured to generate a three-dimensional environment model from monocular vision data and correct positioning errors through a registration algorithm; The state reasoning unit includes: a state modeling module configured to construct a node network including contact state types, spatial relationships of contact points, and physical parameters; The probabilistic reasoning module is configured to output the posterior probability distribution of the contact state and the parameter confidence interval through a dynamic parameter updating mechanism.

9. A computing device, characterized in that include: at least one processor; as well as A memory storing instructions, which, when executed by the at least one processor, enable the at least one processor to execute the hole-finding control method for positioning and assembling a charging gun according to any one of claims 1 to 6.

10. A non-transitory machine-readable storage medium, characterized in that The device stores executable instructions, which, when executed, enable the machine to perform the hole-finding control method for positioning and assembling a charging gun according to any one of claims 1 to 6.

Images