Robot adaptive force control and admittance control method and system for wall panel assembly

Abstract

The present invention discloses a robot adaptive force control and admittance control method and system for wall panel assembly. First, a high-precision photogrammetry system is combined to measure the position and posture of the robot end in real time, and the trajectory is corrected according to the measurement result and the desired position and posture to achieve the positioning of the robot in the position control direction. In the force control direction, a mapping relationship between the robot end force and the contact surface contact force is established through a heteroscedastic neural network mapping model, and an inverse model is jointly trained to map the desired contact force back to the desired end force, so that the contact force on the contact surface is controlled by adjusting the end force, and an adaptive force control and admittance control model is constructed; by interconnecting a virtual energy tank and the adaptive force control and admittance control model, the update of the control model parameters is guided to ensure the passivity of the system; at the same time, the parameters of the virtual energy tank are dynamically adjusted according to the system state, which can maintain the passivity and maximize the assembly accuracy and performance of the robot.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing technology, and in particular relates to a robot adaptive force control and admittance control method and system for wall panel assembly. Background Art

[0002] Panel assemblies are the primary load-bearing component of an aircraft's fuselage structure and are typically assembled from multiple parts, including skins, bulkheads, and stringers. As a crucial component of an aircraft's aerodynamic shape and fuselage structure, the quality of their assembly directly impacts its aerodynamic performance and service life. Traditionally, the assembly and connection of panel assemblies relies primarily on workers manually positioning parts and connecting them using pneumatic tools. The quality of this method is highly dependent on the workers' experience and proficiency. However, traditional methods suffer from long assembly cycles and low efficiency, making them difficult to meet the demands of efficient, high-quality mass production.

[0003] With the development of robotics and robotic equipment technology, robots offer advantages such as flexible movement, large workspaces, and strong parallel and coordinated operation capabilities. They can also integrate multiple sensors and adapt to complex machining environments. Therefore, intelligent manufacturing technology centered around robots is becoming a new trend in the high-quality manufacturing of aircraft panel assemblies. During the assembly of panel assemblies, bulkheads, as key components of the fuselage's load-bearing framework, possess a complex, multi-interface structure. Their assembly with the fuselage skin requires controlling the contact force of each mating interface to ensure adequate contact and avoid excessive contact force that could degrade structural strength. Therefore, real-time monitoring of the contact force at each interface is necessary. However, industrial robots typically only have force sensors installed at the end of the robot. If the sensor data from the end of the robot could be pre-collected and mapped to the contact force of each interface, the contact force at each interface could be controlled by adjusting the robot's end-of-line force, thereby ensuring assembly quality.

[0004] Industrial robots are typically driven using position control. Therefore, if the tracking error of the robot's end-force could be used to dynamically update its desired trajectory, the assembly force could be indirectly adjusted by adjusting the robot's position. Environmental factors often present uncertainties during the assembly process. Admittance control is an effective method for safe assembly, but its performance depends on the interaction between the environment and the admittance parameters. Adaptive admittance control can adjust the admittance parameters online according to task requirements, adapting to unknown environmental changes. Furthermore, the force controller plays a crucial role in achieving force error convergence. Combining force control with adaptive admittance control to design an adaptive force and admittance control system can effectively ensure assembly safety and force convergence. However, adjusting the stiffness of the force controller or the adaptive admittance system may inject energy into the system, causing it to violate passivity and even cause it to diverge.

[0005] Based on this, a robot adaptive force control and admittance control method and system for wall panel assembly is proposed. Summary of the Invention

[0006] In response to the above technical problems, the present invention provides a robot adaptive force control and admittance control method and system for wall panel assembly.

[0007] The technical solution adopted by the present invention to solve the technical problem is:

[0008] A robot adaptive force control and admittance control method for wall panel assembly, the method comprising the following steps:

[0009] S100: Measure the robot's end-point posture in real time using external digital measurement equipment, synchronously collect end-point force and contact force data, and construct a data set;

[0010] S200: Establishing a heteroscedastic neural network mapping model, after training with the data set and the preset first loss function, combining the end force of the robot in the force control direction to predict the mean and sampling variance of the contact force on each contact surface;

[0011] S300: Establishing a neural network inverse mapping model, and training the neural network inverse mapping model using a data set, a preset second loss function, and the predicted contact force mean and sampling variance on each contact surface to achieve mapping from the desired contact force to the desired end force;

[0012] S400: Define the expected contact force range constraints and uniformity indicators for each contact surface, and optimize the expected end force through gradient calculation;

[0013] S500: Define the basic adaptive force control and admittance control models, update the admittance parameters based on the control objective of end force convergence, and perform stability analysis.

[0014] S600: Design a virtual energy tank with adaptive power limiting and energy injection and freezing capabilities, and interconnect it with the basic adaptive force and admittance control model. Through this interconnected structure, modify the adaptive stiffness regulation law and force controller of the admittance parameters to obtain the modified adaptive force and admittance control model.

[0015] S700: Obtaining a corrected expected trajectory value in the force control direction based on the corrected adaptive force control and admittance control model and its parameter values, and obtaining a trajectory correction value in the force control direction based on the actual position of the robot end in the force control direction;

[0016] S800: Measure the actual position of the robot end in the position control direction in real time, and subtract it from the desired position of the robot in the position control direction to obtain the trajectory correction value in the position control direction. The trajectory correction value in the position control direction and the trajectory correction value in the force control direction are sent to the industrial robot to achieve high-precision flexible assembly.

[0017] Preferably, S100 includes:

[0018] S110: Real-time measurement of the robot's end position with the help of external digital measuring equipment , according to the process requirements, it is divided into the position control direction The pose and force control direction , where the goal of position control direction is to achieve precise positioning, and the goal of force control direction is to achieve end force error convergence; the dimension of force control direction is Dimensions of position control direction Depends on the process requirements, when only considering position control, ; When force control is considered in all directions, ;

[0019] S120: Data collection, robot posture , Next, proceed Measure the secondary end force and contact force to obtain the end force sample of the robot in the force control direction And the contact force samples of each contact surface , calculate and record the corresponding sample statistics:

[0020] ;

[0021] in, Indicates the Second position, is the total number of poses obtained, Indicates the Measurement of secondary end force and contact force, represents the total number of measurements of the end force and the contact force, The robot posture in the force control direction Next, through The average value of the robot end force calculated by sampling, is the mean of the corresponding pressure sensor, and its sampling variance is:

[0022] .

[0023] Preferably, S200 includes:

[0024] S210: Construct a heteroscedastic neural network mapping model, divide the data set into training set, test set and validation set, and use the robot end force mean sample in the training set during the training process. After standardization, the heteroscedastic neural network is trained as input to establish a mapping relationship between the mean contact force of the robot end force and each contact surface of the bulkhead in the force control direction and the sample sampling variance. The mapping relationship is expressed as:

[0025] ;

[0026] in, is the number of samples in the training set, and are the neural network functions used to predict the logarithm of the mean and variance of the contact force for each contact surface; The output of is transformed exponentially to obtain the variance to ensure the non-negativity of the variance; and are the parameter sets of the corresponding networks; and The neural network input is the mean end force When , the mean value and sample sampling variance of the contact force of each contact surface of the bulkhead predicted by the neural network;

[0027] S220: Designing a loss function for neural network parameters, combining maximum likelihood estimation with neural network parameter sets , the specific loss function is designed as follows:

[0028] ;

[0029] Use the optimization algorithm to update the parameters of the neural network:

[0030] ;

[0031] in, is the learning rate;

[0032] S230: Use the heteroscedastic neural network mapping model to infer the samples in the validation set and obtain the predicted sample sampling variance ; Using the actual sample sampling variance corresponding to the contact force in the validation set , calibrate the prediction variance; for this purpose, define the following calibration loss function:

[0033] ;

[0034] in, is the sample variance calibration factor, is the number of samples in the validation set;

[0035] The process of updating the sample variance calibration factor using the optimization algorithm is as follows:

[0036] ;

[0037] in, is the learning rate, and the calibrated prediction sampling variance is ;

[0038] S240: Based on the trained neural network model and sample variance calibration factor, the robot's end force in the force control direction is combined , predict the mean contact force of each contact surface and the calibrated predicted sampling variance .

[0039] Preferably, S300 includes:

[0040] S310: Construct a neural network inverse mapping model. Use the dataset combined with the predicted contact force mean and sampling variance on each contact surface to train the neural network inverse mapping model. Define the joint training objective as minimizing the prediction error of the neural network inverse mapping model and the consistency error of the heteroscedastic neural network mapping model:

[0041] ;

[0042] in, The first The mean contact force of the samples The end force predicted by the inverse mapping model is ;and and They are The contact force mean and sampling variance predicted by the heteroscedastic neural network mapping model;

[0043] Use the optimization algorithm to update the parameters of the neural network inverse mapping model:

[0044] ;

[0045] in, is the learning rate;

[0046] S320: Using the trained neural network inverse mapping model to achieve the desired contact force mean To the desired end force Real-time reasoning, namely:

[0047] ;

[0048] in, is the parameter set of the neural network inverse mapping model, It is a neural network inverse mapping model.

[0049] Preferably, S400 includes:

[0050] S410: Define the target function for the expected contact force range of each contact surface. The target function must ensure sufficient contact between each contact surface while avoiding excessive contact force that may cause a decrease in structural strength. Therefore, the contact force value of each contact surface should meet the upper and lower limit requirements of the process when the assembly is completed.

[0051] Assumptions For the predicted The contact force of the contact surface is: , is the total number of contact surfaces; the actual contact force on each contact surface should meet the following conditions:

[0052] ;

[0053] in, and are the lower and upper bounds of the desired contact force, respectively;

[0054] Taking into account the uncertainty of the sensor sampling data, the conditions that the actual contact force of each contact surface should meet are modified to:

[0055] ;

[0056] in, is the first predicted by the neural network The sampling standard deviation of the contact force of each contact surface;

[0057] Taking full account of uncertainty and contact force uniformity, the allowable value of the upper limit of the contact force is set to , the allowable value of the lower limit of the contact force is set to ,in for The maximum standard deviation of the contact force on each contact surface is is the maximum value function; in addition, the upper limit of the desired contact force and lower bound Through the neural network inverse mapping model, it is converted into the upper limit of the end force expected in the force control direction. and lower bound ;

[0058] Since the contact force is indirectly adjusted by adjusting the end force, the objective function of the contact force range constraint is designed as:

[0059] ;

[0060] in, is a linear rectification function, and its specific expression is , , and They are 、 and No. Quantity

[0061] S420: Define the contact uniformity objective function to minimize the difference between contact forces in order to make the contact force distribution more uniform, that is, to minimize the contact force With their average The difference between the two is used to map the contact uniformity to the end using the neural network inverse model. The objective function is defined as:

[0062] ;

[0063] in, The expected average contact force of each contact surface is When , the expected end force on contact uniformity in the corresponding force control direction is obtained by inference of the neural network inverse model; therefore, it can be obtained about Gradient:

[0064] ;

[0065] S430: Calculate the desired contact force range gradient. According to the objective function of S410, let and ,available about Gradient:

[0066] ;

[0067] in, is the indicator function, and its specific expression is:

[0068] ;

[0069] S440: Design of the desired end force. Combined with the gradients calculated in S420 and S430, the ideal end force variation law is designed as follows:

[0070] ;

[0071] in, is a weight coefficient, is a constant, is a projection operator, used to ensure Once within the predetermined range Within the predetermined range, it is always maintained, and its specific expression is:

[0072] ;

[0073] The end force obtained according to the ideal end force variation law Desired end force ,Right now ,In addition, since industrial robots can usually only receive position ,commands, it is necessary to adjust the position of the robot according to the ,end force error to achieve the control of the end force.

[0074] Preferably, S500 includes:

[0075] S510: Robot end force tracking error that defines the force control direction , specifically:

[0076] ;

[0077] S520: Design a force controller and use the following PI force controller to adjust the robot's end force tracking error:

[0078] ;

[0079] in, and is a positive definite diagonal matrix, It is the adaptive factor of the force controller coefficient. When the force tracking error is large, the integral coefficient is increased to accelerate the convergence of the end force error; when the force tracking error is small, the integral coefficient is reduced to avoid overcorrection. The specific form is as follows:

[0080] ;

[0081] in, is a preset threshold. is a constant;

[0082] S530: Design the activation function of the PI force controller. When the bulkhead and the skin come into contact, the force controller is turned on. In addition, to ensure the safety of the bulkhead and the robot, the force controller is turned off when the contact force exceeds the set value. Therefore, the activation function matrix of the force controller is defined as: , where the activation function matrix of the force controller is diagonal elements The specific form is:

[0083] ;

[0084] in, , is a constant, is a positive constant and satisfies , represents a diagonal matrix, when Greater than threshold The force controller is activated only when When the force controller is closed, it starts to close smoothly until hour The force controller is fully closed to avoid excessive contact forces;

[0085] S540: Design the activation function of the adaptive admittance controller, defined as: ,in The specific form is:

[0086] ;

[0087] in, ; is the width of the transition section of the adaptive admittance control activation function;

[0088] S550: Design-based adaptive force and admittance control models, specifically:

[0089] ;

[0090] in, 、 、 are positive definite diagonal matrices, representing the inertia matrix, damping matrix, and stiffness matrix, respectively. , 、 and are the adjustments of the desired position, velocity and acceleration in the force control direction, is the expected assembly point planned in the force control direction, is the updated desired assembly point in the force control direction;

[0091] S560: Update the parameters of the adaptive admittance control and design the following adaptive law to update the stiffness matrix , its specific expression is:

[0092] ;

[0093] in, is the basic stiffness update gradient, and its specific expression is ; and is the learning rate, and They are vectors and No. elements; and They are the stiffness lower bound matrices and the stiffness upper bound matrix No. diagonal elements, let , ,function Used to punish within a predetermined range External stiffness value;

[0094] S570: Perform stability analysis on the basic adaptive force control and admittance control strategies, considering the following energy functions:

[0095] ;

[0096] right Differentiating it yields:

[0097] ;

[0098] Substitute the basic adaptive force control and admittance dynamic equations into In the equation, we can get:

[0099] ;

[0100] because is a positive definite matrix, so is always non-positive, but since it cannot be guaranteed in advance and is non-positive, so the basic adaptive force control and admittance control system cannot be guaranteed to be Passivity.

[0101] Preferably, S600 includes:

[0102] S610: Design the dynamics of an energy tank with online energy flow injection and freezing mechanism, and define the energy stored in the energy tank as ,in, The status of the energy tank. for time; The energy difference is regarded as the initial energy margin pre-existing in the energy tank that can be used for energy exchange, where Indicates the initial time, is the initial energy value stored in the energy tank, is the lower bound of energy in the energy tank, and the relationship between them satisfies ;

[0103] To ensure both safety and system performance, and in conjunction with the goal of force control, namely, convergence of the robot's end-of-line force error, the following dynamics model for an energy tank with online energy flow injection and freezing mechanisms is proposed. This dynamically adjusts the energy in the energy tank to ensure the system maintains efficient and safe operation during actual operation:

[0104] ;

[0105] in, is a constant used to store part of the energy consumed by damping; is the corrected speed update amount, is the corrected planned velocity vector No. elements, at this time, and Can be respectively Integration and differentiation give the force control direction The sum of the power of the force controller and the stiffness update The expression is ,in, and is the adaptive power scaling factor; and They are vectors and No. elements; is the stiffness update gradient of the foundation in S560; It is The function of online energy injection and freezing mechanism in the force control direction, coefficient The role of energy tank is to ensure the energy value stored in the energy tank Does not exceed the predetermined energy limit ,therefore is defined as follows:

[0106] ;

[0107] Among them, the function Used when the energy reaches the preset lower limit of the energy tank When , the energy exchange between the adaptive force control and admittance control system and the virtual energy tank is smoothly interrupted to ensure that the value of the energy tank is not lower than the lower limit value, so is defined as follows:

[0108] ;

[0109] in, is the width of the smooth transition section of energy value; It is The function of the online energy injection and freezing mechanism in the force control direction is to keep the end force of the robot within the safe range. If the energy stored in the energy tank is too low, resulting in system performance being limited, energy is injected online to ensure system performance; in addition, when the contact force is too large, the system When the contact force is large, the energy in the energy tank is frozen and exchanged with the energy in the energy tank to avoid the adaptive force control and admittance system from extracting energy from the energy tank, thereby ensuring the safety of the system; in order to achieve a smooth conversion from energy injection and energy freezing, a transition section from energy injection to energy freezing is set within a safe range when the contact force is large. , its function is when Start the smooth transition from energy injection function to energy freezing function, when When , the energy exchange is completely frozen, and its specific expression is:

[0110] ;

[0111] in, is a constant, function ; Constant; the first Adaptive scaling coefficient for energy injection and freezing The specific expression is ;

[0112] S620: Design adaptive power flow to adjust and update the maximum power allowed to be injected into the actual system in real time according to task requirements and system status; for this purpose, define the following adaptive power scaling coefficients ,in The specific form is as follows:

[0113] ;

[0114] Among them, the power of the force controller is , the power of the energy tank modified stiffness update gradient is , according to the goal of force control direction, the following design is made about The update law of:

[0115] ;

[0116] in, is the learning rate; It is the first end force components; is the forgetting factor, which is used to prevent the maximum power value allowed to be injected in the actual system from being too large;

[0117] S630: Design a revised force control and adaptive admittance control model, which is designed as follows:

[0118] ;

[0119] in, for dimensional identity matrix, is the function matrix of the online energy injection and freezing mechanism in the force-controlled direction, is the adaptive power scaling coefficient matrix, and the update law of the modified stiffness matrix is Designed to:

[0120] ;

[0121] in, is the first The update law for the diagonal elements, and They are the stiffness lower bound matrices and the stiffness upper bound matrix No. diagonal elements;

[0122] In order to ensure the stability and performance of the system, the damping value of the system is determined according to the critical damping condition:

[0123] ;

[0124] S640: Stability and performance analysis, defining the total energy function of the interconnected system consisting of the robot system and the energy tank as:

[0125] ;

[0126] in, is the energy of the robot system, the total power of the force controller and stiffness update in the force control direction and The expression is rewritten as ,in The gradient matrix is updated for the original stiffness with power limit, which is expressed as ;right Taking the derivative we get:

[0127] ;

[0128] In order to analyze the stability of the interconnected system consisting of the robot system and the energy tank, we define For the set of all force control directions, define To control the direction A collection of and Two situations are used to analyze the stability and performance of the system, where Indicates an empty set. In addition, From the expression of , we can see that the following inequality always holds:

[0129] ;

[0130] Scenario 1: Under this condition, for , both have , so there is and ,at this time Can be simplified to:

[0131] ;

[0132] Depend on ,It can be seen that in this case, the interconnected system is stable;

[0133] Scenario 2: ;for have and for , , at this time there is and ,in, And when hour , Simplified to:

[0134] ;

[0135] Since the matrix is a diagonal positive semidefinite matrix, so when Sometimes, there are , so when When , the interconnected system is stable; when hour, Decompose into ,in yes The corresponding Lyapunov function is, yes The corresponding Lyapunov function is and , therefore, when When the system The direction of is passive, for It is stable. Since passivity can ensure the stable interaction between the robot system and any passive environment, the system is stable in this case. In summary, both system cases 1 and 2 are stable, so the system is always stable.

[0136] S650: Force tracking error convergence analysis, the total system variable is defined as ,in, The environmental energy is The state of the environment at this time, the total energy function of the system is defined as Consider the optimal state and So that:

[0137] ;

[0138] in, yes The space of all possible values; since the environment always dissipates energy, there is and Sometimes, there are , so at this time there is ,so The whole system is asymptotically stable; consider the equilibrium point Under this condition, the modified adaptive force control and admittance control model becomes:

[0139] ;

[0140] Easy to prove is the only steady-state solution of the system, so the end force of the robot converges to the desired force.

[0141] Preferably, S700 includes:

[0142] According to the admittance parameter value obtained in S500, the modified adaptive force control and admittance control model is substituted to obtain the posture adjustment amount:

[0143] ;

[0144] in, and are the adjustments to the expected position and expected velocity at the previous sampling moment, is the sampling period;

[0145] Update the desired position of the force control direction:

[0146] ;

[0147] Thus, the trajectory correction in the force control direction is obtained: ,in The actual position of the end of the force-controlled robot measured by an external digital measuring device.

[0148] Preferably, S800 includes:

[0149] S810: The actual position of the end of the robot is controlled according to the position measured by the external digital measuring device , and the desired position in the position control direction Compare the two values and calculate the trajectory correction in the position control direction based on the difference between them: ;

[0150] S820: Write the robot control program. Since the industrial robot receives the position control instruction, the updated trajectory correction value Send to the robot, thereby achieving precise control of the robot.

[0151] A robot adaptive force control and admittance control system for panel assembly, comprising a data set acquisition module, a heteroscedastic neural network mapping model establishment module, a neural network inverse mapping model establishment module, an expected end force optimization module, an admittance parameter update module, an adaptive regulation law for admittance parameters and a force controller correction module, a trajectory expectation value correction module, and trajectory correction;

[0152] The data set acquisition module is used to measure the position and posture of the robot end in real time based on external digital measurement equipment, synchronously collect the end force and contact force data, and construct a data set;

[0153] A heteroscedastic neural network mapping model building module is used to build a heteroscedastic neural network mapping model. After training with a data set and a preset first loss function, the model predicts the mean and sampling variance of the contact force on each contact surface by combining the robot's end force in the force control direction.

[0154] A neural network inverse mapping model building module is used to build a neural network inverse mapping model. After training the neural network inverse mapping model using a data set, a preset second loss function, and the predicted contact force mean and sampling variance on each contact surface, the model can be mapped from the desired contact force to the desired end force.

[0155] The expected end force optimization module is used to define the expected contact force range constraints and uniformity indicators for each contact surface, and optimize the expected end force through gradient calculation;

[0156] The admittance parameter update module is used to define the basic adaptive force control and admittance control model, update the admittance parameters according to the control objective of end force convergence, and perform stability analysis;

[0157] The adaptive regulation law for admittance parameters and the force controller correction module are used to design a virtual energy tank with adaptive power limitation and energy injection and freezing functions. This is interconnected with the basic adaptive force control and admittance control model. Through this interconnected structure, the adaptive stiffness regulation law for admittance parameters and the force controller are modified to obtain the modified adaptive force control and admittance control model.

[0158] The trajectory expectation value correction module is used to obtain the corrected trajectory expectation value in the force control direction based on the corrected adaptive force control and admittance control model and its parameter values, and to obtain the trajectory correction value in the force control direction in combination with the actual position of the robot end in the force control direction;

[0159] The trajectory correction module is used to measure the actual posture of the robot end in the position control direction in real time, and subtract it from the expected posture of the robot in the position control direction to obtain the trajectory correction value in the position control direction. The trajectory correction value in the position control direction and the trajectory correction value in the force control direction are sent to the industrial robot to achieve high-precision flexible assembly.

[0160] The aforementioned adaptive force and admittance control method and system for robot panel assembly utilizes a high-precision photogrammetry system to measure the robot's end-point pose in real time. Trajectory correction is performed based on the measured results and the desired pose, achieving high-precision positioning of the robot in the direction of position control. In the force control direction, a heteroscedastic neural network mapping model is used to establish a mapping relationship between the robot's end-point force and the contact force on the contact surface. An inverse model is then trained to map the desired contact force back to the desired end-point force. This allows the contact force on the contact surface to be controlled by adjusting the end-point force, further constructing an adaptive force and admittance control model. Furthermore, by interconnecting a virtual energy tank with the adaptive force and admittance control model, the control model parameters are updated to ensure system passivity. Dynamic adjustment of the virtual energy tank parameters based on system status maintains system passivity while maximizing the robot's assembly accuracy and performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0161] Figure 1 Flowchart of a robot adaptive force control and admittance control method for wall panel assembly according to one embodiment of the present invention;

[0162] Figure 2 FIG. 4 is a control block diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0163] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below with reference to the accompanying drawings.

[0164] In one embodiment, Figure 1 As shown, a robot adaptive force control and admittance control method for wall panel assembly includes the following steps:

[0165] S100: Using an external digital measurement device to measure the robot's end-point posture in real time, synchronously collect end-point force and contact force data, and construct a data set; that is, measure the preset end-point force and contact force under the posture as a data set;

[0166] S200: Establishing a heteroscedastic neural network mapping model, using the data set and a preset first loss function to train the heteroscedastic neural network mapping model, and using the trained neural network model to combine the end force of the robot in the force control direction to predict the contact force mean and sampling variance on each contact surface;

[0167] S300: Establishing a neural network inverse mapping model, using the data set, a preset second loss function, and the predicted contact force mean and sampling variance on each contact surface to train the neural network inverse mapping model, and using the trained neural network inverse mapping model to achieve mapping from the desired contact force to the desired end force;

[0168] S400: defining the expected contact force range constraint and uniformity index for each contact surface, and optimizing the expected end force through gradient calculation; that is, defining the expected contact force range objective function and contact uniformity objective function for each contact surface, calculating the expected contact force range gradient based on the expected contact force range objective function for each contact surface, and optimizing the expected end force by combining the contact uniformity objective function and the expected contact force range gradient;

[0169] S500: Define the basic adaptive force control and admittance control models, update the admittance parameters based on the control objective of end force convergence, and perform stability analysis.

[0170] S600: Design a virtual energy tank with adaptive power limiting and energy injection and freezing capabilities, and interconnect it with the basic adaptive force and admittance control model. Through this interconnected structure, modify the adaptive stiffness regulation law and force controller of the admittance parameters to obtain the modified adaptive force and admittance control model.

[0171] S700: Obtaining a corrected expected trajectory value in the force control direction based on the corrected adaptive force control and admittance control model and its parameter values, and obtaining a trajectory correction value in the force control direction based on the real-time measurement of the actual position of the robot end in the force control direction;

[0172] S800: Use high-precision instruments to measure the actual position of the robot end in the position control direction in real time, and subtract it from the desired position of the robot in the position control direction to obtain the trajectory correction value in the position control direction. The trajectory correction value in the position control direction and the trajectory correction value in the force control direction are sent to the industrial robot to achieve high-precision flexible assembly.

[0173] Specifically, the present invention proposes a robot adaptive force control and admittance control method for panel assembly, but the following problems may occur during the assembly of robot aircraft panel components: (1) The robot has low absolute positioning accuracy, which makes it difficult to meet the high-precision positioning requirements of panel component assembly; in addition, in mass production, the contact surfaces of the bulkhead lack sensors, resulting in the inability to accurately measure the actual contact force; (2) The controller is an effective means to achieve contact force convergence, but it does not meet the passivity condition, which easily leads to instability of the robot system and is difficult to dynamically adjust according to environmental changes and lacks compliance; (3) Admittance control has good compliance. Combining admittance control with a force controller to form an adaptive force control and admittance control system helps to improve the robot's interactive ability in complex environments; however, the performance of the admittance system is affected by environmental factors and admittance parameters. When the environment is unknown, it is difficult to select appropriate admittance parameters to ensure the performance of the system. Although variable admittance control can achieve adaptive adjustment of admittance parameters, changing the stiffness of the admittance system may cause the system to violate passivity, thereby causing instability; (4) Although the energy tank can ensure the passivity of the adaptive force control and admittance control system, the safety and performance of the system still depend on the energy storage value in the energy tank and the maximum power allowed to be injected into the system, and these two parameters are difficult to determine in advance in practical applications. Based on the above problems, the following improvements are proposed: (1) The end position of the robot is measured in real time by a digital photogrammetry system, and a closed-loop control algorithm is designed based on the measurement results to correct the robot trajectory, thereby achieving high-precision positioning; at the same time, the mapping relationship between the end force and the contact force is established using a heteroscedastic neural network through the robot end force and its corresponding contact force data collected in the early stage, so that the contact force can be predicted by the robot end force in the actual batch production process. In addition, the inverse mapping model of the neural network is jointly trained to convert the desired contact force into the desired end force; (2) According to indicators such as the ideal contact force range of the contact surface, the loss function is defined and the desired end force is updated, and the activation function of the force controller and the adaptive admittance control is designed to ensure the tracking performance of the robot system in free space. The force controller is combined with the adaptive admittance control to construct a unified force control and adaptive admittance control system; (3) the adaptive force control and admittance system is interconnected with the virtual energy tank to ensure the passivity of the system; (4) to avoid rapid parameter changes, the power limit of the energy tank is set to ensure the safety of the system. At the same time, the power limit of the energy tank is dynamically adjusted according to the contact force error, and an energy injection and freezing mechanism is designed to further improve the system performance while ensuring the safety of the system. Through the above steps, this method effectively solves the positioning accuracy and force control problems of the robot during the wall panel assembly process, and ensures that the system operates with high precision and high stability by dynamically adjusting and optimizing the control parameters.

[0174] In one embodiment, S100 includes:

[0175] S110: Since industrial robots usually have low absolute positioning accuracy, in order to improve the robot's motion accuracy, it is necessary to use external digital measurement equipment (such as digital photogrammetry systems, laser trackers, etc.) to measure the position and posture of the robot end in real time. , according to the process requirements, it is divided into the position control direction The pose and force control direction , where the goal of position control direction is to achieve precise positioning, and the goal of force control direction is to achieve end force error convergence; the dimension of force control direction is Dimensions of position control direction Depends on the process requirements, when only considering position control, ; When force control is considered in all directions, ;

[0176] S120: Data collection, robot posture , Next, proceed Measure the secondary end force and contact force to obtain the end force sample of the robot in the force control direction And the contact force samples of each contact surface , calculate and record the corresponding sample statistics:

[0177] ;

[0178] in, Indicates the Second position, is the total number of poses obtained, Indicates the Measurement of secondary end force and contact force, represents the total number of measurements of the end force and the contact force, The robot posture in the force control direction Next, through The average value of the robot end force calculated by sampling, is the mean of the corresponding pressure sensor, and its sampling variance is:

[0179] .

[0180] In one embodiment, S200 includes:

[0181] S210: Construct a heteroscedastic neural network mapping model, divide the data set into training set, test set and validation set, and use the robot end force mean sample in the training set during the training process. After standardization, the heteroscedastic neural network is trained as input to establish a mapping relationship between the mean contact force of the robot end force and each contact surface of the bulkhead in the force control direction and the sample sampling variance. The mapping relationship is expressed as:

[0182] ;

[0183] in, is the number of samples in the training set, and are the neural network functions used to predict the logarithm of the mean and variance of the contact force for each contact surface; The output of is transformed exponentially to obtain the variance to ensure the non-negativity of the variance; and are the parameter sets of the corresponding networks; and The neural network input is the mean end force When , the mean value and sample sampling variance of the contact force of each contact surface of the bulkhead predicted by the neural network;

[0184] S220: Designing a loss function for neural network parameters, combining maximum likelihood estimation with neural network parameter sets , the specific loss function is designed as follows:

[0185] ;

[0186] Use the optimization algorithm to update the parameters of the neural network:

[0187] ;

[0188] in, is the learning rate;

[0189] S230: To ensure that the prediction variance of the neural network model can accurately reflect the uncertainty in the real data, the prediction variance needs to be calibrated; the heteroscedastic neural network mapping model is used to infer the samples in the validation set to obtain the predicted sample sampling variance ; Using the actual sample sampling variance corresponding to the contact force in the validation set , calibrate the prediction variance; for this purpose, define the following calibration loss function:

[0190] ;

[0191] in, is the sample variance calibration factor, is the number of samples in the validation set;

[0192] The process of updating the sample variance calibration factor using the optimization algorithm is as follows:

[0193] ;

[0194] in, is the learning rate, and the calibrated prediction sampling variance is ;

[0195] S240: Based on the trained neural network model and sample variance calibration factor, the robot's end force in the force control direction is combined , predict the mean contact force of each contact surface and the calibrated predicted sampling variance .

[0196] In one embodiment, S300 includes:

[0197] S310: Construct a neural network inverse mapping model. Use the dataset combined with the predicted contact force mean and sampling variance on each contact surface to train the neural network inverse mapping model. Define the joint training objective as minimizing the prediction error of the neural network inverse mapping model and the consistency error of the heteroscedastic neural network mapping model:

[0198] ;

[0199] in, The first The mean contact force of the samples The end force predicted by the inverse mapping model is ;and and They are The contact force mean and sampling variance predicted by the heteroscedastic neural network mapping model;

[0200] Use the optimization algorithm to update the parameters of the neural network inverse mapping model:

[0201] ;

[0202] in, is the learning rate;

[0203] S320: Using the trained neural network inverse mapping model to achieve the desired contact force mean To the desired end force Real-time reasoning, namely:

[0204] ;

[0205] in, is the parameter set of the neural network inverse mapping model, It is a neural network inverse mapping model.

[0206] In one embodiment, S400 includes:

[0207] S410: Define the target function for the expected contact force range of each contact surface. The target function must ensure sufficient contact between each contact surface while avoiding excessive contact force that may cause a decrease in structural strength. Therefore, the contact force value of each contact surface should meet the upper and lower limit requirements of the process when the assembly is completed.

[0208] Assumptions For the predicted The contact force of the contact surface is: , is the total number of contact surfaces; the actual contact force on each contact surface should meet the following conditions:

[0209] ;

[0210] in, and are the lower and upper bounds of the desired contact force, respectively;

[0211] Taking into account the uncertainty of the sensor sampling data, the conditions that the actual contact force of each contact surface should meet are modified to:

[0212] ;

[0213] in, is the first predicted by the neural network The sampling standard deviation of the contact force of each contact surface;

[0214] Taking full account of uncertainty and contact force uniformity, the allowable value of the upper limit of the contact force is set to , the allowable value of the lower limit of the contact force is set to ,in for The maximum standard deviation of the contact force on each contact surface is is the maximum value function; in addition, the upper limit of the desired contact force and lower bound Through the neural network inverse mapping model, it is converted into the upper limit of the end force expected in the force control direction. and lower bound ;

[0215] Since the contact force is indirectly adjusted by adjusting the end force, the objective function of the contact force range constraint is designed as:

[0216] ;

[0217] in, is a linear rectification function, and its specific expression is , , and They are 、 and No. Quantity

[0218] S420: Define the contact uniformity objective function to minimize the difference between contact forces in order to make the contact force distribution more uniform, that is, to minimize the contact force With their average The difference between the two is used to map the contact uniformity to the end using the neural network inverse model. The objective function is defined as:

[0219] ;

[0220] in, The expected average contact force of each contact surface is When , the expected end force on contact uniformity in the corresponding force control direction is obtained by inference of the neural network inverse model; therefore, it can be obtained about Gradient:

[0221] ;

[0222] S430: Calculate the desired contact force range gradient. According to the objective function of S410, let and ,available about Gradient:

[0223] ;

[0224] in, is the indicator function, and its specific expression is:

[0225] ;

[0226] S440: Design of the desired end force. Combined with the gradients calculated in S420 and S430, the ideal end force variation law is designed as follows:

[0227] ;

[0228] in, is a weight coefficient, is a constant, is a projection operator, used to ensure Once within the predetermined range Within the predetermined range, it is always maintained, and its specific expression is:

[0229] ;

[0230] It is worth noting that industrial robots usually only have position control and can only change the end force indirectly by adjusting the position. Therefore, the end force obtained according to the ideal end force change law Desired end force ,Right now Then, an adaptive force control and admittance model is designed to adjust the desired position of the force control direction so that the actual end force of the robot in the force control direction is Finally converges to the desired end force , so that the contact force of the corner piece converges to the desired contact force.

[0231] In one embodiment, S500 includes:

[0232] S510: Robot end force tracking error that defines the force control direction , specifically:

[0233] ;

[0234] S520: Design a force controller and use the following PI force controller to adjust the robot's end force tracking error:

[0235] ;

[0236] in, and is a positive definite diagonal matrix, It is the adaptive factor of the force controller coefficient. When the force tracking error is large, the integral coefficient is increased to accelerate the convergence of the end force error; when the force tracking error is small, the integral coefficient is reduced to avoid overcorrection. The specific form is as follows:

[0237] ;

[0238] in, is a preset threshold. is a constant;

[0239] S530: Design the activation function of the PI force controller. When the bulkhead and the skin come into contact, the force controller is turned on. In addition, to ensure the safety of the bulkhead and the robot, the force controller is turned off when the contact force exceeds the set value. Therefore, the activation function matrix of the force controller is defined as: , where the activation function matrix of the force controller is diagonal elements The specific form is:

[0240] ;

[0241] in, , is a constant, is a positive constant and satisfies , represents a diagonal matrix, when Greater than threshold This design helps to avoid false triggering due to sensor measurement noise, zero drift and other factors, thereby ensuring that the robot's tracking performance in free space is not affected; when When the force controller is closed, it starts to close smoothly until hour The force controller is fully closed to avoid excessive contact forces;

[0242] S540: Design the activation function of the adaptive admittance controller. The activation function of the adaptive admittance controller is designed to ensure the robot's motion performance in free space, while accurately executing adaptive admittance control when the bulkhead contacts the skin, achieving compliance and ensuring the safety of the bulkhead and skin. The activation function of the adaptive admittance controller is defined as: ,in The specific form is:

[0243] ;

[0244] in, ; is the width of the transition section of the adaptive admittance control activation function;

[0245] S550: Design-based adaptive force and admittance control models, specifically:

[0246] ;

[0247] in, 、 、 are positive definite diagonal matrices, representing the inertia matrix, damping matrix, and stiffness matrix, respectively. , 、 and are the adjustments of the desired position, velocity and acceleration in the force control direction, is the expected assembly point planned in the force control direction, is the updated desired assembly point in the force control direction;

[0248] S560: Update the parameters of the adaptive admittance control and design the following adaptive law to update the stiffness matrix , its specific expression is:

[0249] ;

[0250] in, is the basic stiffness update gradient, and its specific expression is ; and is the learning rate, and They are vectors and No. elements; and They are the stiffness lower bound matrices and the stiffness upper bound matrix No. diagonal elements, let , ,function Used to punish within a predetermined range External stiffness value;

[0251] S570: Perform stability analysis on the basic adaptive force control and admittance control strategies, considering the following energy functions:

[0252] ;

[0253] right Differentiating it yields:

[0254] ;

[0255] Substitute the basic adaptive force control and admittance dynamic equations into In the equation, we can get:

[0256] ;

[0257] because is a positive definite matrix, so is always non-positive, but since it cannot be guaranteed in advance and is non-positive, so the basic adaptive force control and admittance control system cannot be guaranteed to be Passivity.

[0258] In one embodiment, S600 includes:

[0259] S610: Design the dynamics of an energy tank with online energy flow injection and freezing mechanism, and define the energy stored in the energy tank as ,in, The status of the energy tank. for time; The energy difference is regarded as the initial energy margin pre-existing in the energy tank that can be used for energy exchange, where Indicates the initial time, is the initial energy value stored in the energy tank, is the lower bound of energy in the energy tank, and the relationship between them satisfies Since energy exchange occurs when the actual system is connected to the energy tank, the system performance will be affected by the energy storage value in the energy tank: if the initial energy margin is too small, the system performance requirements may not be met; if the initial energy margin is too large, the system may become unsafe.

[0260] To ensure both safety and system performance, and in conjunction with the goal of force control, namely, convergence of the robot's end-of-line force error, the following dynamics model for an energy tank with online energy flow injection and freezing mechanisms is proposed. This dynamically adjusts the energy in the energy tank to ensure the system maintains efficient and safe operation during actual operation:

[0261] ;

[0262] in, is a constant used to store part of the energy consumed by damping; is the corrected speed update amount, is the corrected planned velocity vector No. elements, at this time, and Can be respectively Integration and differentiation give the force control direction The sum of the power of the force controller and the stiffness update The expression is ,in, and is the adaptive power scaling factor; and They are vectors and No. elements; is the stiffness update gradient of the foundation in S560; It is The function of online energy injection and freezing mechanism in the force control direction, coefficient The role of energy tank is to ensure the energy value stored in the energy tank Does not exceed the predetermined energy limit ,therefore is defined as follows:

[0263] ;

[0264] Among them, the function Used when the energy reaches the preset lower limit of the energy tank When , the energy exchange between the adaptive force control and admittance control system and the virtual energy tank is smoothly interrupted to ensure that the value of the energy tank is not lower than the lower limit value, so is defined as follows:

[0265] ;

[0266] in, is the width of the smooth transition section of energy value; It is The function of the online energy injection and freezing mechanism in the force control direction is to keep the end force of the robot within the safe range. If the energy stored in the energy tank is too low, resulting in system performance being limited, energy is injected online to ensure system performance; in addition, when the contact force is too large, the system When the contact force is large, the energy in the energy tank is frozen and exchanged with the energy in the energy tank to avoid the adaptive force control and admittance system from extracting energy from the energy tank, thereby ensuring the safety of the system; in order to achieve a smooth conversion from energy injection and energy freezing, a transition section from energy injection to energy freezing is set within a safe range when the contact force is large. , its function is when Start the smooth transition from energy injection function to energy freezing function, when When , the energy exchange is completely frozen, and its specific expression is:

[0267] ;

[0268] in, is a constant, function ; Constant; the first Adaptive scaling coefficient for energy injection and freezing The specific expression is ;

[0269] S620: Design adaptive power flow. Although energy tanks can ensure the passivity of interconnected systems, if the power injected into the system is too large, it may cause system insecurity. Conversely, if the injected power is too small, it may limit the actual performance and response speed of the system. Therefore, according to the mission requirements and system status, the maximum power allowed to be injected into the actual system is adjusted and updated in real time. To this end, the following adaptive power scaling factor is defined: ,in The specific form is as follows:

[0270] ;

[0271] Among them, the power of the force controller is , the power of the energy tank modified stiffness update gradient is , according to the goal of force control direction, the following design is made about The update law of:

[0272] ;

[0273] in, is the learning rate; It is the first end force components; is the forgetting factor, which is used to prevent the maximum power value allowed to be injected in the actual system from being too large;

[0274] S630: Design a revised force control and adaptive admittance control model, which is designed as follows:

[0275] ;

[0276] in, for dimensional identity matrix, is the function matrix of the online energy injection and freezing mechanism in the force-controlled direction, is the adaptive power scaling coefficient matrix, and the update law of the modified stiffness matrix is Designed to:

[0277] ;

[0278] in, is the first The update law for the diagonal elements, and They are the stiffness lower bound matrices and the stiffness upper bound matrix No. diagonal elements;

[0279] In order to ensure the stability and performance of the system, the damping value of the system is determined according to the critical damping condition:

[0280] ;

[0281] S640: Stability and performance analysis, defining the total energy function of the interconnected system consisting of the robot system and the energy tank as:

[0282] ;

[0283] in, is the energy of the robot system, the total power of the force controller and stiffness update in the force control direction and The expression is rewritten as ,in The gradient matrix is updated for the original stiffness with power limit, which is expressed as ;right Taking the derivative we get:

[0284] ;

[0285] In order to analyze the stability of the interconnected system consisting of the robot system and the energy tank, we define For the set of all force control directions, define To control the direction A collection of and Two situations are used to analyze the stability and performance of the system, where Indicates an empty set. In addition, From the expression of , we can see that the following inequality always holds:

[0286] ;

[0287] Scenario 1: Under this condition, for , both have , so there is and ,at this time Can be simplified to:

[0288] ;

[0289] Depend on ,It can be seen that in this case, the interconnected system is stable;

[0290] Scenario 2: ;for have and for , , at this time there is and ,in, And when hour , Simplified to:

[0291] ;

[0292] Since the matrix is a diagonal positive semidefinite matrix, so when Sometimes, there are , so when When , the interconnected system is stable; when hour, Decompose into ,in yes The corresponding Lyapunov function is, yes The corresponding Lyapunov function is and , therefore, when When the system The direction of is passive, for It is stable. Since passivity can ensure the stable interaction between the robot system and any passive environment, the system is stable in this case. In summary, both system cases 1 and 2 are stable, so the system is always stable.

[0293] S650: Force tracking error convergence analysis, the total system variable is defined as ,in, The environmental energy is The state of the environment at this time, the total energy function of the system is defined as Consider the optimal state and So that:

[0294] ;

[0295] in, yes The space of all possible values; since the environment always dissipates energy, there is and Sometimes, there are , so at this time there is ,so The whole system is asymptotically stable; consider the equilibrium point Under this condition, the modified adaptive force control and admittance control model becomes:

[0296] ;

[0297] Easy to prove is the only steady-state solution of the system, so the end force of the robot converges to the desired force.

[0298] In one embodiment, S700 includes:

[0299] According to the admittance parameter value obtained in S500, the modified adaptive force control and admittance control model is substituted to obtain the posture adjustment amount:

[0300] ;

[0301] in, and are the adjustments to the expected position and expected velocity at the previous sampling moment, is the sampling period;

[0302] Update the desired position of the force control direction:

[0303] ;

[0304] Thus, the trajectory correction in the force control direction is obtained: ,in The actual position of the end of the force-controlled robot measured by an external digital measuring device.

[0305] In one embodiment, S800 includes:

[0306] S810: The actual position of the end of the robot is controlled according to the position measured by the external digital measuring device , and the desired position in the position control direction Compare the two values and calculate the trajectory correction in the position control direction based on the difference between them: ;

[0307] S820: Write the robot control program. Since the industrial robot receives the position control instruction, the updated trajectory correction value Send to the robot, thereby achieving precise control of the robot.

[0308] Compared with the existing technology, the beneficial effects of the present invention are as follows: (1) An innovative adaptive force control and admittance control model is proposed to optimize the desired force at the robot end based on performance indicators and realize adaptive adjustment of the admittance parameters according to the force control target. The model is interconnected with a virtual energy tank with adaptive power limitation and online energy injection and freezing mechanism to ensure the safety and performance stability of the system. At the same time, by ensuring that the robot end force accurately tracks the desired force, the precise and flexible assembly of the wall panel components is achieved. (2) A contact force prediction model for each contact surface of the bulkhead based on a heteroscedastic neural network is proposed. The model can predict the contact force and sampling variance of each contact surface of the bulkhead in real time based on the robot end force, thereby controlling the contact force by adjusting the end force. This method effectively ensures the assembly quality of the bulkhead and avoids the problems of insufficient contact or excessive contact force. (3) By designing the force control activation function and the adaptive admittance activation function, the proposed scheme can be applied to both free motion space and contact-rich scenarios. (4) A real-time high-precision control method for robot motion based on digital measurement is designed to improve the robot's motion accuracy.

[0309] A robot adaptive force control and admittance control system for panel assembly, comprising a data set acquisition module, a heteroscedastic neural network mapping model establishment module, a neural network inverse mapping model establishment module, an expected end force optimization module, an admittance parameter update module, an adaptive regulation law for admittance parameters and a force controller correction module, a trajectory expectation value correction module, and trajectory correction;

[0310] The data set acquisition module is used to measure the position and posture of the robot end in real time based on external digital measurement equipment, synchronously collect the end force and contact force data, and construct a data set;

[0311] A heteroscedastic neural network mapping model building module is used to build a heteroscedastic neural network mapping model. After training with a data set and a preset first loss function, the model predicts the mean and sampling variance of the contact force on each contact surface by combining the robot's end force in the force control direction.

[0312] A neural network inverse mapping model building module is used to build a neural network inverse mapping model. After training the neural network inverse mapping model using a data set, a preset second loss function, and the predicted contact force mean and sampling variance on each contact surface, the model can be mapped from the desired contact force to the desired end force.

[0313] The expected end force optimization module is used to define the expected contact force range constraints and uniformity indicators for each contact surface, and optimize the expected end force through gradient calculation;

[0314] The admittance parameter update module is used to define the basic adaptive force control and admittance control model, update the admittance parameters according to the control objective of end force convergence, and perform stability analysis;

[0315] The adaptive regulation law for admittance parameters and the force controller correction module are used to design a virtual energy tank with adaptive power limitation and energy injection and freezing functions. This is interconnected with the basic adaptive force control and admittance control model. Through this interconnected structure, the adaptive stiffness regulation law for admittance parameters and the force controller are modified to obtain the modified adaptive force control and admittance control model.

[0316] The trajectory expectation value correction module is used to obtain the corrected trajectory expectation value in the force control direction based on the corrected adaptive force control and admittance control model and its parameter values, and to obtain the trajectory correction value in the force control direction in combination with the actual position of the robot end in the force control direction;

[0317] The trajectory correction module is used to measure the actual posture of the robot end in the position control direction in real time, and subtract it from the expected posture of the robot in the position control direction to obtain the trajectory correction value in the position control direction. The trajectory correction value in the position control direction and the trajectory correction value in the force control direction are sent to the industrial robot to achieve high-precision flexible assembly.

[0318] The specific definition of the robotic adaptive force control and admittance control system for wall panel assembly can be found in the definition of the robotic adaptive force control and admittance control method for wall panel assembly described above, and will not be repeated here. The various modules in the above-mentioned robotic adaptive force control and admittance control system for wall panel assembly can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above-mentioned modules.

[0319] The above is a detailed introduction to the robot adaptive force control and admittance control method and system for wall panel assembly provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A robot adaptive force control and admittance control method for wall panel assembly, characterized in that: The method comprises the following steps: S100: Measure the robot's end-point posture in real time using external digital measurement equipment, synchronously collect end-point force and contact force data, and construct a data set; S200: Establishing a heteroscedastic neural network mapping model, after training with the data set and the preset first loss function, combining the end force of the robot in the force control direction to predict the mean and sampling variance of the contact force on each contact surface; S300: Establishing a neural network inverse mapping model, and training the neural network inverse mapping model using a data set, a preset second loss function, and the predicted contact force mean and sampling variance on each contact surface to achieve mapping from the desired contact force to the desired end force; S400: Define the expected contact force range constraints and uniformity indicators for each contact surface, and optimize the expected end force through gradient calculation; S500: Define the basic adaptive force control and admittance control models, update the admittance parameters based on the control objective of end force convergence, and perform stability analysis. S600: Design a virtual energy tank with adaptive power limiting and energy injection and freezing capabilities, and interconnect it with the basic adaptive force and admittance control model. Through this interconnected structure, modify the adaptive stiffness regulation law and force controller of the admittance parameters to obtain the modified adaptive force and admittance control model. S700: Obtaining a corrected expected trajectory value in the force control direction based on the corrected adaptive force control and admittance control model and its parameter values, and obtaining a trajectory correction value in the force control direction based on the actual position of the robot end in the force control direction; S800: Measure the actual position of the robot end in the position control direction in real time, and subtract it from the desired position of the robot in the position control direction to obtain the trajectory correction value in the position control direction. The trajectory correction value in the position control direction and the trajectory correction value in the force control direction are sent to the industrial robot to achieve high-precision flexible assembly.

2. The method according to claim 1, characterized in that S100 includes: S110: Real-time measurement of the robot's end position with the help of external digital measuring equipment , according to the process requirements, it is divided into the position control direction The pose and force control direction , where the goal of position control direction is to achieve precise positioning, and the goal of force control direction is to achieve end force error convergence; the dimension of force control direction is Dimensions of position control direction Depends on the process requirements, when only considering position control, ; When force control is considered in all directions, ; S120: Data collection, robot posture , Next, proceed Measure the secondary end force and contact force to obtain the end force sample of the robot in the force control direction And the contact force samples of each contact surface , calculate and record the corresponding sample statistics: ； in, Indicates the Second position, is the total number of poses obtained, Indicates the Measurement of secondary end force and contact force, represents the total number of measurements of the end force and the contact force, The robot posture in the force control direction Next, through The average value of the robot end force calculated by sampling, is the mean of the corresponding pressure sensor, and its sampling variance is: 。 3. The method according to claim 2, characterized in that S200 includes: S210: Construct a heteroscedastic neural network mapping model, divide the data set into training set, test set and validation set, and use the robot end force mean sample in the training set during the training process. After standardization, the heteroscedastic neural network is trained as input to establish a mapping relationship between the mean contact force of the robot end force and each contact surface of the bulkhead in the force control direction and the sample sampling variance. The mapping relationship is expressed as: ； in, is the number of samples in the training set, and are the neural network functions used to predict the logarithm of the mean and variance of the contact force for each contact surface; The output of is transformed exponentially to obtain the variance to ensure the non-negativity of the variance; and are the parameter sets of the corresponding networks; and The neural network input is the mean end force When , the mean value and sample sampling variance of the contact force of each contact surface of the bulkhead predicted by the neural network; S220: Designing a loss function for neural network parameters, combining maximum likelihood estimation with neural network parameter sets , the specific loss function is designed as follows: ； Use the optimization algorithm to update the parameters of the neural network: ； in, is the learning rate; S230: Use the heteroscedastic neural network mapping model to infer the samples in the validation set and obtain the predicted sample sampling variance ; Using the actual sample sampling variance corresponding to the contact force in the validation set , calibrate the prediction variance; for this purpose, define the following calibration loss function: ； in, is the sample variance calibration factor, is the number of samples in the validation set; The process of updating the sample variance calibration factor using the optimization algorithm is as follows: ； in, is the learning rate, and the calibrated prediction sampling variance is ; S240: Based on the trained neural network model and sample variance calibration factor, the robot's end force in the force control direction is combined , predict the mean contact force of each contact surface and the calibrated predicted sampling variance .

4. The method according to claim 3, characterized in that S300 includes: S310: Construct a neural network inverse mapping model. Use the dataset combined with the predicted contact force mean and sampling variance on each contact surface to train the neural network inverse mapping model. Define the joint training objective as minimizing the prediction error of the neural network inverse mapping model and the consistency error of the heteroscedastic neural network mapping model: ； in, The first The mean contact force of the samples The end force predicted by the inverse mapping model is ;and and They are The contact force mean and sampling variance predicted by the heteroscedastic neural network mapping model; Use the optimization algorithm to update the parameters of the neural network inverse mapping model: ； in, is the learning rate; S320: Using the trained neural network inverse mapping model to achieve the desired contact force mean To the desired end force Real-time reasoning, namely: ； in, is the parameter set of the neural network inverse mapping model, It is a neural network inverse mapping model.

5. The method according to claim 4, characterized in that S400 includes: S410: Define the target function for the expected contact force range of each contact surface. The target function must ensure sufficient contact between each contact surface while avoiding excessive contact force that may cause a decrease in structural strength. Therefore, the contact force value of each contact surface should meet the upper and lower limit requirements of the process when the assembly is completed. Assumptions For the predicted The contact force of the contact surface is: , is the total number of contact surfaces; the actual contact force on each contact surface should meet the following conditions: ； in, and are the lower and upper bounds of the desired contact force, respectively; Taking into account the uncertainty of the sensor sampling data, the conditions that the actual contact force of each contact surface should meet are modified to: ； in, is the first predicted by the neural network The sampling standard deviation of the contact force of each contact surface; Taking full account of uncertainty and contact force uniformity, the allowable value of the upper limit of the contact force is set to , the allowable value of the lower limit of the contact force is set to ,in for The maximum standard deviation of the contact force on each contact surface is is the maximum value function; in addition, the upper limit of the desired contact force and lower bound Through the neural network inverse mapping model, it is converted into the upper limit of the end force expected in the force control direction. and lower bound ; Since the contact force is indirectly adjusted by adjusting the end force, the objective function of the contact force range constraint is designed as: ； in, is a linear rectification function, and its specific expression is , , and They are 、 and No. Quantity S420: Define the contact uniformity objective function to minimize the difference between contact forces in order to make the contact force distribution more uniform, that is, to minimize the contact force With their average The difference between the two is used to map the contact uniformity to the end using the neural network inverse model. The objective function is defined as: ； in, The expected average contact force of each contact surface is When , the expected end force on contact uniformity in the corresponding force control direction is obtained by inference of the neural network inverse model; therefore, it can be obtained about Gradient: ； S430: Calculate the desired contact force range gradient. According to the objective function of S410, let and ,available about Gradient: ； in, is the indicator function, and its specific expression is: ； S440: Design of the desired end force. Combined with the gradients calculated in S420 and S430, the ideal end force variation law is designed as follows: ； in, is a weight coefficient, is a constant, is a projection operator, used to ensure Once within the predetermined range Within the predetermined range, it is always maintained, and its specific expression is: ； The end force obtained according to the ideal end force variation law Desired end force ,Right now ,In addition, since industrial robots can usually only receive position ,commands, it is necessary to adjust the position of the robot according to the ,end force error to achieve the control of the end force.

6. The method according to claim 5, characterized in that S500 includes: S510: Robot end force tracking error that defines the force control direction , specifically: ； S520: Design a force controller and use the following PI force controller to adjust the robot's end force tracking error: ； in, and is a positive definite diagonal matrix, It is the adaptive factor of the force controller coefficient. When the force tracking error is large, the integral coefficient is increased to accelerate the convergence of the end force error; when the force tracking error is small, the integral coefficient is reduced to avoid overcorrection. The specific form is as follows: ； in, is a preset threshold. is a constant; S530: Design the activation function of the PI force controller. When the bulkhead and the skin come into contact, the force controller is turned on. In addition, to ensure the safety of the bulkhead and the robot, the force controller is turned off when the contact force exceeds the set value. Therefore, the activation function matrix of the force controller is defined as: , where the activation function matrix of the force controller is diagonal elements The specific form is: ； in, , is a constant, is a positive constant and satisfies , represents a diagonal matrix, when Greater than threshold The force controller is activated only when When the force controller is closed, it starts to close smoothly until hour The force controller is fully closed to avoid excessive contact forces; S540: Design the activation function of the adaptive admittance controller, defined as: ,in The specific form is: ; in, ; is the width of the transition section of the adaptive admittance control activation function; S550: Design-based adaptive force and admittance control models, specifically: ; in, 、 、 are positive definite diagonal matrices, representing the inertia matrix, damping matrix, and stiffness matrix, respectively. , 、 and are the adjustments of the desired position, velocity and acceleration in the force control direction, is the expected assembly point planned in the force control direction, is the updated desired assembly point in the force control direction; S560: Update the parameters of the adaptive admittance control and design the following adaptive law to update the stiffness matrix , its specific expression is: ； in, is the basic stiffness update gradient, and its specific expression is ; and is the learning rate, and They are vectors and No. elements; and They are the stiffness lower bound matrices and the stiffness upper bound matrix No. diagonal elements, let , ,function Used to punish within a predetermined range External stiffness value; S570: Perform stability analysis on the basic adaptive force control and admittance control strategies, considering the following energy functions: ; right Differentiating it yields: ; Substitute the basic adaptive force control and admittance dynamic equations into In the equation, we can get: ; because is a positive definite matrix, so is always non-positive, but since it cannot be guaranteed in advance and is non-positive, so the basic adaptive force control and admittance control system cannot be guaranteed to be Passivity.

7. The method according to claim 6, characterized in that S600 includes: S610: Design the dynamics of an energy tank with online energy flow injection and freezing mechanism, and define the energy stored in the energy tank as ,in, The status of the energy tank. for time; The energy difference is regarded as the initial energy margin pre-existing in the energy tank that can be used for energy exchange, where Indicates the initial time, is the initial energy value stored in the energy tank, is the lower bound of energy in the energy tank, and the relationship between them satisfies ; To ensure both safety and system performance, and in conjunction with the goal of force control, namely, convergence of the robot's end-of-line force error, the following dynamics model for an energy tank with online energy flow injection and freezing mechanisms is proposed. This dynamically adjusts the energy in the energy tank to ensure the system maintains efficient and safe operation during actual operation: ； in, is a constant used to store part of the energy consumed by damping; is the corrected speed update amount, is the corrected planned velocity vector No. elements, at this time, and Can be respectively Integration and differentiation give the force control direction The sum of the power of the force controller and the stiffness update The expression is ,in, and is the adaptive power scaling factor; and They are vectors and No. elements; is the stiffness update gradient of the foundation in S560; It is The function of online energy injection and freezing mechanism in the force control direction, coefficient The role of energy tank is to ensure the energy value stored in the energy tank Does not exceed the predetermined energy limit ,therefore is defined as follows: ； Among them, the function Used when the energy reaches the preset lower limit of the energy tank When , the energy exchange between the adaptive force control and admittance control system and the virtual energy tank is smoothly interrupted to ensure that the value of the energy tank is not lower than the lower limit value, so is defined as follows: ； in, is the width of the smooth transition section of energy value; It is The function of the online energy injection and freezing mechanism in the force control direction is to keep the end force of the robot within the safe range. If the energy stored in the energy tank is too low, resulting in system performance being limited, energy is injected online to ensure system performance; in addition, when the contact force is too large, the system When the contact force is large, the energy in the energy tank is frozen and exchanged with the energy in the energy tank to avoid the adaptive force control and admittance system from extracting energy from the energy tank, thereby ensuring the safety of the system; in order to achieve a smooth conversion from energy injection and energy freezing, a transition section from energy injection to energy freezing is set within a safe range when the contact force is large. , its function is when Start the smooth transition from energy injection function to energy freezing function, when When , the energy exchange is completely frozen, and its specific expression is: ； in, is a constant, function ; Constant; the first Adaptive scaling coefficient for energy injection and freezing The specific expression is ; S620: Design adaptive power flow to adjust and update the maximum power allowed to be injected into the actual system in real time according to task requirements and system status; for this purpose, define the following adaptive power scaling coefficients ,in The specific form is as follows: ； Among them, the power of the force controller is , the power of the energy tank modified stiffness update gradient is , according to the goal of force control direction, the following design is made about The update law of: ； in, is the learning rate; It is the first end force components; is the forgetting factor, which is used to prevent the maximum power value allowed to be injected in the actual system from being too large; S630: Design a revised force control and adaptive admittance control model, which is designed as follows: ； in, for dimensional identity matrix, is the function matrix of the online energy injection and freezing mechanism in the force-controlled direction, is the adaptive power scaling coefficient matrix, and the update law of the modified stiffness matrix is Designed to: ； in, is the first The update law for the diagonal elements, and They are the stiffness lower bound matrices and the stiffness upper bound matrix No. diagonal elements; In order to ensure the stability and performance of the system, the damping value of the system is determined according to the critical damping condition: ； S640: Stability and performance analysis, defining the total energy function of the interconnected system consisting of the robot system and the energy tank as: ； in, is the energy of the robot system, the total power of the force controller and stiffness update in the force control direction and The expression is rewritten as ,in The gradient matrix is updated for the original stiffness with power limit, which is expressed as ;right Taking the derivative we get: ； In order to analyze the stability of the interconnected system consisting of the robot system and the energy tank, we define For the set of all force control directions, define To control the direction A collection of and Two situations are used to analyze the stability and performance of the system, where Indicates an empty set. In addition, From the expression of , we can see that the following inequality always holds: ； Scenario 1: Under this condition, for , both have , so there is and ,at this time Can be simplified to: ； Depend on ,It can be seen that in this case, the interconnected system is stable; Scenario 2: ;for have and for , , at this time there is and ,in, And when hour , Simplified to: ； Since the matrix is a diagonal positive semidefinite matrix, so when Sometimes, there are , so when When , the interconnected system is stable; when hour, Decompose into ,in yes The corresponding Lyapunov function is, yes The corresponding Lyapunov function is and , therefore, when When the system The direction of is passive, for It is stable. Since passivity can ensure the stable interaction between the robot system and any passive environment, the system is stable in this case. In summary, both system cases 1 and 2 are stable, so the system is always stable. S650: Force tracking error convergence analysis, the total system variable is defined as ,in, The environmental energy is The state of the environment at this time, the total energy function of the system is defined as Consider the optimal state and So that: ； in, yes The space of all possible values; since the environment always dissipates energy, there is and Sometimes, there are , so at this time there is ,so The whole system is asymptotically stable; consider the equilibrium point Under this condition, the modified adaptive force control and admittance control model becomes: ； Easy to prove is the only steady-state solution of the system, so the end force of the robot converges to the desired force.

8. The method according to claim 7, characterized in that The S700 includes: According to the admittance parameter value obtained in S500, the modified adaptive force control and admittance control model is substituted to obtain the posture adjustment amount: ； in, and are the adjustments to the expected position and expected velocity at the previous sampling moment, is the sampling period; Update the desired position of the force control direction: ； Thus, the trajectory correction in the force control direction is obtained: ,in The actual position of the end of the force-controlled robot measured by an external digital measuring device.

9. The method according to claim 8, characterized in that The S800 includes: S810: The actual position of the end of the robot is controlled according to the position measured by the external digital measuring device , and the desired position in the position control direction Compare the two values and calculate the trajectory correction in the position control direction based on the difference between them: ; S820: Write the robot control program. Since the industrial robot receives the position control instruction, the updated trajectory correction value Send to the robot, thereby achieving precise control of the robot.

10. A robot adaptive force control and admittance control system for wall panel assembly, characterized in that: It includes a data set acquisition module, a heteroscedastic neural network mapping model establishment module, a neural network inverse mapping model establishment module, an expected terminal force optimization module, an admittance parameter update module, an adaptive adjustment law of the admittance parameter and a force controller correction module, a trajectory expectation value correction module, and a trajectory correction module; The data set acquisition module is used to measure the position and posture of the robot end in real time based on external digital measurement equipment, synchronously collect the end force and contact force data, and construct a data set; A heteroscedastic neural network mapping model building module is used to build a heteroscedastic neural network mapping model. After training with a data set and a preset first loss function, the model predicts the mean and sampling variance of the contact force on each contact surface by combining the robot's end force in the force control direction. A neural network inverse mapping model building module is used to build a neural network inverse mapping model. After training the neural network inverse mapping model using a data set, a preset second loss function, and the predicted contact force mean and sampling variance on each contact surface, the model can be mapped from the desired contact force to the desired end force. The expected end force optimization module is used to define the expected contact force range constraints and uniformity indicators for each contact surface, and optimize the expected end force through gradient calculation; The admittance parameter update module is used to define the basic adaptive force control and admittance control model, update the admittance parameters according to the control objective of end force convergence, and perform stability analysis; The adaptive regulation law for admittance parameters and the force controller correction module are used to design a virtual energy tank with adaptive power limitation and energy injection and freezing functions. This is interconnected with the basic adaptive force control and admittance control model. Through this interconnected structure, the adaptive stiffness regulation law for admittance parameters and the force controller are modified to obtain the modified adaptive force control and admittance control model. The trajectory expectation value correction module is used to obtain the corrected trajectory expectation value in the force control direction based on the corrected adaptive force control and admittance control model and its parameter values, and to obtain the trajectory correction value in the force control direction in combination with the actual position of the robot end in the force control direction; The trajectory correction module is used to measure the actual posture of the robot end in the position control direction in real time, and subtract it from the expected posture of the robot in the position control direction to obtain the trajectory correction value in the position control direction. The trajectory correction value in the position control direction and the trajectory correction value in the force control direction are sent to the industrial robot to achieve high-precision flexible assembly.

Images