A smart electric screwdriver compliance compensation and driving pose collaborative topology optimization method

Abstract

The application discloses a kind of intelligent electric screwdriver compliance compensation and driving pose collaborative topology optimization method, constructs intelligent electric screwdriver rigid-flexible coupling dynamics model, the coupling relationship between rigid driving movement, compliance deformation and external contact reaction force is characterized;Further establish the multi-objective collaborative topology optimization model with topological optimization material density variable and driving pose variable as joint optimization variable;Again based on adaptive element optimization framework with element controller is solved by adaptive collaborative optimization, obtains optimal structure topological configuration, optimal driving pose reference, optimal time-varying stiffness matrix and optimal time-varying damping matrix;Finally, the optimal solution above is injected into the collaborative control law containing compliance deformation direct feedback item, executes closed-loop compliance compensation and pose collaborative control.The application realizes the synchronous optimal matching and adaptive regulation and control of intelligent electric screwdriver structure topology, driving pose and compliance characteristics, significantly improves pose accuracy, torque control stability and compliance compensation response performance.

Description

Technical Field

[0001] This invention relates to a method for intelligent electric screwdriver compliance compensation and drive pose collaborative topology optimization, belonging to the field of digital data processing and precision manufacturing technology. Background Technology

[0002] In high-precision manufacturing fields such as consumer electronics, medical devices, and aerospace, the precise tightening of threaded fasteners is a critical process that determines the assembly quality and reliability of products. As a core execution device, the performance of intelligent electric screwdrivers directly impacts product yield and production efficiency. Current mainstream electric screwdriver technologies fall into two main categories. The first category employs a high-rigidity structural design combined with high-precision torque closed-loop control, enabling high-precision positional repeatability. However, due to the lack of an effective compliance compensation mechanism, it is prone to generating excessive contact force when dealing with workpiece positioning deviations, material size fluctuations, or assembly stress, leading to thread damage, abnormal torque, or even workpiece scrap. The second category introduces a passive compliance mechanism at the end of the electric screwdriver to mitigate rigid impact. However, this mechanism is often designed independently from the drive system, and the additional deformation it introduces severely degrades the positional accuracy of the drive end. Furthermore, it cannot dynamically adjust compliance characteristics according to real-time operating conditions, resulting in poor tightening torque stability. The root cause of these problems lies in the disconnect between the electric screwdriver's mechanical structure topology design, the drive system's motion positional planning, and the compliance control strategy, preventing the system from achieving global optimization. Summary of the Invention

[0003] This invention provides a method for intelligent electric screwdriver compliance compensation and drive pose collaborative topology optimization to solve the problems existing in the prior art.

[0004] The technical solutions adopted in this invention are as follows:

[0005] A method for topology optimization in collaboration between intelligent electric screwdriver compliance compensation and drive pose includes the following steps:

[0006] A rigid-flexible coupling dynamic model of the entire intelligent electric screwdriver is constructed. The rigid-flexible coupling dynamic model represents the coupling relationship between rigid driving motion, compliant deformation and external contact reaction force.

[0007] A multi-objective collaborative topology optimization model is established with the topology optimization material density variable of the driving pose variable and the compliant compensation mechanism as the joint optimization variable;

[0008] The multi-objective cooperative topology optimization model is adaptively and cooperatively optimized based on an adaptive meta-optimization framework with a meta-controller to obtain the optimal structural topology configuration, optimal driving pose reference, optimal time-varying stiffness matrix, and optimal time-varying damping matrix; the meta-controller adaptively calls the optimization algorithm according to the optimization task state.

[0009] The optimal structural topology, the optimal driving pose reference, the optimal time-varying stiffness matrix, and the optimal time-varying damping matrix are used as the optimal solution to inject into the cooperative control law, and closed-loop compliance compensation and pose cooperative control are performed. The cooperative control law includes a compliance deformation direct feedback term.

[0010] Furthermore, the specific steps for constructing the rigid-flexible coupling dynamic model of the intelligent electric screwdriver are as follows:

[0011] The intelligent electric screwdriver is disassembled into a rigid drive mechanism and a compliant compensation mechanism. The rigid drive mechanism includes a servo drive motor, a transmission component, and a bit end actuator, while the compliant compensation mechanism includes a flexible compensation actuator and a flexible hinge component.

[0012] Based on the second type of Lagrange equations, a fundamental dynamic equation is established to characterize the coupling relationship between rigid-driven motion, compliant deformation, and external contact reaction force.

[0013] Furthermore, the steps for constructing a rigid-flexible coupling dynamic model of the intelligent electric screwdriver also include:

[0014] A six-dimensional pose error vector is introduced, and the mapping relationship between pose deviation and compliant deformation of rigid drive mechanism is established through Jacobian matrix;

[0015] By introducing time-varying stiffness and time-varying damping matrices, the basic dynamic equations are extended to obtain a complete rigid-flexible coupling dynamic model.

[0016] Furthermore, the optimization objectives of the multi-objective collaborative topology optimization model include structural lightweighting, driving pose positioning accuracy, structural mechanical strength, tightening torque stability, and compliance compensation accuracy.

[0017] The objective function of the multi-objective cooperative topology optimization model is:

[0018] ,

[0019] in, For topology optimization of material density variables, To drive pose variables, This is a six-dimensional pose error vector. To optimize the volume of the structure, The initial structural volume, Let be the second norm of the six-dimensional pose error vector. Let be the second norm of the maximum allowable deformation of the structure. The maximum stress of the structure, The stress yield strength of the material. For servo drive torque, The rated tightening torque, This represents the maximum value of the compliance deformation. This is the allowable value for compliance deformation. , , , as well as Weighting coefficients to satisfy the normalization condition.

[0020] Furthermore, the optimized task status of the meta-controller includes task characteristics, resource status characteristics, historical optimized performance characteristics, and user preferences.

[0021] Furthermore, the optimization algorithms for adaptive invocation of the meta-controller include deep offline optimization algorithms based on NSGA-III, online adaptive optimization algorithms based on deep reinforcement learning, or hybrid collaborative optimization algorithms.

[0022] Furthermore, the deep offline optimization algorithm uses the parameterized level set method to characterize the macroscopic boundary of the compliant compensation mechanism, uses the solid isotropic microstructure penalty model to characterize the material density distribution inside the compliant compensation mechanism, and solves the dynamic response through the extended finite element method and iteratively evolves to convergence.

[0023] Furthermore, an online adaptive optimization algorithm is used to construct a deep reinforcement learning agent. The agent is trained using the proximal policy optimization (PPO) algorithm and initialized with behavior cloning using offline optimization experience pool data. The state space of the agent includes real-time perceived operating parameters, and the action space of the agent includes topology density adjustment, driving pose adjustment, time-varying stiffness matrix adjustment, and time-varying damping matrix adjustment.

[0024] Furthermore, the hybrid collaborative optimization algorithm first generates expert solutions through offline optimization using finite algebras, and then preheats and locally refines the online reinforcement learning strategy based on the expert solutions.

[0025] Furthermore, the expression for the cooperative control law is:

[0026] ,

[0027] in, For servo drive torque, As the optimal driving pose reference, The speed that is the optimal driving pose reference. To drive the generalized coordinates of the joints, To drive the generalized velocity of the joint, For position scaling gain, For velocity differential gain, Gain for front-end pose error compensation. External contact force feedback compensates for gain. For compliant deformation feedback gain, This is a six-dimensional pose error vector. For external contact reaction force, This represents the amount of compliant deformation.

[0028] This invention constructs a rigid-flexible coupled dynamic model of an intelligent electric screwdriver, unifying the representation of rigid driving motion, compliant deformation, and external contact reaction force. Furthermore, it establishes a multi-objective collaborative topology optimization model with topology optimization material density variables and driving pose variables as joint optimization variables. This fundamentally solves the problem of the separation between structural topology design, driving pose planning, and compliant control strategies in existing technologies. The resulting beneficial effects are:

[0029] (1) By using rigid-flexible coupling dynamic modeling and a collaborative control law that includes a direct feedback term for compliant deformation, the pose compensation of the drive system, the interaction of external forces and the compliant deformation of the structure itself are unified in a control framework, realizing the synchronous and collaborative regulation of compliant compensation and pose control, effectively reducing the external contact reaction force during the assembly process, and avoiding thread damage and workpiece scrap.

[0030] (2) By using the topology optimization material density variable and the driving pose variable as joint optimization variables for multi-objective collaborative topology optimization, the optimal matching of the structural topology configuration and the driving pose reference is realized, avoiding the degradation of the driving end pose accuracy by the additional deformation of the compliant compensation mechanism, and improving the stability of the tightening torque.

[0031] (3) By adaptively calling multi-objective genetic algorithms, deep reinforcement learning algorithms or hybrid collaborative modes according to the state of the optimization task through the adaptive meta-optimization framework, the system realizes the organic integration of offline global exploration and online real-time response, enabling the system to dynamically select the optimal solution strategy according to task characteristics, computing resources and real-time requirements, thereby improving optimization efficiency and decision adaptability.

[0032] (4) By injecting the optimal structural topology, optimal driving pose reference, optimal time-varying stiffness matrix and optimal time-varying damping matrix as the optimal solution into the cooperative control law, the optimization result is directly driven by the closed-loop control, so that the electric screwdriver structural topology, driving pose and compliance characteristics are continuously kept in the optimal matching state during operation, and the global optimality of the overall system performance is improved. Attached Figure Description

[0033] Figure 1 This is a schematic diagram illustrating the principle of the present invention.

[0034] Figure 2 This is a schematic diagram of the internal architecture and operation mechanism of the meta controller.

[0035] Figure 3 This is a flowchart of the meta controller's self-learning process.

[0036] Figure 4 This is a flowchart illustrating the execution process of the method of the present invention. Detailed Implementation

[0037] The invention will now be further described with reference to the accompanying drawings.

[0038] This invention discloses a method for topology optimization of intelligent electric screwdriver with compliance compensation and drive pose coordination. The intelligent electric screwdriver is an existing conventional intelligent equipment. This application only briefly introduces its structural composition to help understand the technical solution of the method of this invention, without going into detail about its specific mechanical principles.

[0039] The intelligent electric screwdriver includes a servo drive mechanism, a compliance compensation mechanism, a multi-source pose detection mechanism, and a tightening actuator, among which:

[0040] The servo drive mechanism includes a servo drive motor and transmission components;

[0041] The compliant compensation mechanism includes a flexible compensation actuator and a flexible hinge assembly;

[0042] The multi-source pose detection mechanism includes a six-dimensional force sensor, a non-contact pose detection unit, a torque encoder, and a servo motor encoder;

[0043] The tightening actuator includes the end actuator of the screwdriver bit.

[0044] The method is based on the collaborative working characteristics of the servo drive mechanism, compliant compensation mechanism, multi-source pose detection mechanism, and tightening actuator. It constructs a rigid-flexible coupling dynamic model of the system and, based on this, establishes a multi-objective collaborative topology optimization model with structural topology configuration and drive pose parameters as joint optimization variables. Finally, through a closed-loop control law incorporating real-time feedback of compliant deformation, it achieves synchronous optimal matching and adaptive control of the electric screwdriver's structural topology and drive pose. The basic structure of this method is as follows: Figure 1 As shown.

[0045] Step 1: Construct a rigid-flexible coupling dynamic model of the entire intelligent electric screwdriver.

[0046] This invention distinguishes between rigid drive mechanisms and compliant compensation mechanisms by mechanically disassembling the servo drive motor, transmission components, flexible compensation execution components, and bit end actuator of an intelligent electric screwdriver. The servo drive motor, transmission components, and bit end actuator are rigid drive mechanisms, while the flexible compensation execution components and flexible hinge components, which have significant elastic deformation capabilities, are compliant compensation mechanisms.

[0047] Based on the second type of Lagrange equations, a rigid-flexible coupling dynamic equation for the whole machine is established, which can accurately characterize the coupling relationship between rigid-driven motion, compliant deformation, and external tightening reaction force, as shown in the following equation:

[0048] ,

[0049] in, For the overall machine inertia matrix, The matrix of Coriolis force and centrifugal force. For gravity, This represents the vector of elastic damping force generated by the compliant compensation mechanism. For servo drive torque, For external contact reaction force, For the generalized coordinates of the driven joints, representing the position / angle of the driven joints of the intelligent electric screwdriver, The generalized velocity vector of the driving joint represents the motion velocity of the driving joint. The first derivative with respect to time, Let be the generalized acceleration vector driving the joint, representing the motion acceleration driving the joint. The second derivative with respect to time.

[0050] To establish the correlation between pose deviation and compliant deformation of a rigid drive mechanism, a six-dimensional pose error vector is introduced. And through the Jacobian matrix Establish mapping relationship:

[0051] ,

[0052] in, The amount of compliant deformation can be measured by a displacement sensor mounted on the electric screwdriver.

[0053] Further introduce time-varying stiffness matrix With damping matrix This yields the complete dynamic model:

[0054] ,

[0055] in, This refers to the compliant deformation speed.

[0056] To unify the topology optimization and control objectives of the intelligent electric screwdriver system, this invention constructs a multi-objective collaborative topology optimization function with five optimization objectives: lightweight structure, drive pose positioning accuracy, structural mechanical strength, tightening torque stability, and compliance compensation accuracy. Simultaneously, multiple constraints are set, including mechanical constraints, pose constraints, process constraints, and drive capability constraints. The resulting multi-objective optimization function is as follows:

[0057] ,

[0058] in, For topology optimization, the material density variable takes values ​​in the range [0,1]. To drive pose variables, This is a six-dimensional pose error vector. To optimize the volume of the structure, The initial structural volume, Let be the second norm of the six-dimensional pose error vector. Let be the second norm of the maximum allowable deformation of the structure. The maximum stress of the structure, The stress yield strength of the material. For servo drive torque, The rated tightening torque, This represents the maximum value of the compliance deformation. This is the allowable value for compliance deformation. to Weighting coefficients If the normalization conditions are met, the initial calibration can be performed using the analytic hierarchy process (AHP) for different precision tightening scenarios, and the subsequent calibration can be dynamically corrected by an adaptive adjustment mechanism during the optimization process.

[0059] Step 2: Adaptive collaborative optimization solution based on the meta-optimization framework.

[0060] To optimize the rigid-flexible coupling dynamic model of an intelligent electric screwdriver system, this invention proposes an innovative adaptive meta-optimization framework. This framework deeply integrates the NSGA-Ⅲ (Non-dominated Sorting Genetic Algorithm III) multi-objective genetic algorithm with a deep reinforcement learning algorithm, forming an integrated solver that can autonomously adjust its optimization behavior based on task characteristics, computational resources, real-time requirements, and historical optimization performance. The core of this framework is a meta-controller, whose internal architecture and operating mechanism are as follows: Figure 2 As shown.

[0061] The meta-controller monitors a set of feature variables that reflect the current state of the optimization task in real time, which constitute its decision state space. This state space includes, but is not limited to: the dimension of the optimization problem, the tightness of the constraints, the available offline computation time window, the system's requirement for real-time response, the convergence curve shape of the objective function value during historical optimization processes, and the performance deviation reported by the verification iteration module, as shown in the following equation:

[0062] ,

[0063] in, This indicates task characteristics, including the dimensions of design variables, constraint tightness, and relative volume of the feasible region; Indicates resource status characteristics, including available computing power resources and real-time requirement levels; It represents historical optimization performance characteristics, including the shape of historical convergence curves; Indicates user preferences.

[0064] Based on the aforementioned state information, the meta-controller embeds a lightweight classifier to initially categorize the current optimization task into one of three modes: exploration-first, real-time response, or hybrid collaborative. Simultaneously, the meta-controller is equipped with a periodically trainable meta-policy network to continuously optimize the decision-making logic.

[0065] In the initial stage of operation, the classifier makes independent decisions, and the meta-policy network does not participate. After accumulating enough task samples, the two enter a parallel fusion stage, where the output of the classifier and the output of the meta-policy network are weighted by confidence gating and jointly determine the mode selection. As the performance of the meta-policy network matures, the classifier is gradually downgraded to a pre-feature extractor, and its hidden features become part of the input of the meta-policy network, forming a cascaded structure.

[0066] To implement this evolution mechanism, the meta-controller continuously records the features of each task, the selected optimization mode, the parameter settings in the hybrid collaborative mode, and the final performance achievement, forming a meta-learning dataset. Based on this dataset, the meta-controller periodically trains the meta-policy network to optimize its decision logic and simultaneously adjusts the gating fusion parameters. The meta-controller's self-learning execution process is as follows: Figure 3 As shown.

[0067] Explore-first mode: Deep offline optimization based on NSGA-Ⅲ.

[0068] When the meta-controller determines that the optimization task has the following characteristics, it enters the exploration-first mode: high design space dimension, no prior knowledge, multiple local optima and unclear global optimum, no real-time response required by the system and sufficient offline computing resources. In this mode, an offline optimization algorithm based on NSGA-III is used for optimization.

[0069] structural topological density variables With driving pose variables As joint optimization variables, the parameterized level set method is used to explicitly characterize the macroscopic boundary of the compliant compensation mechanism, while the material density distribution inside the mechanism is implicitly characterized based on a solid isotropic microstructure penalty model. In each iteration, the extended finite element method is used to solve the rigid-flexible coupling dynamic response under the current topology and driving pose, accurately obtaining key performance indicators such as structural stress field, deformation field, end pose error, driving torque fluctuation, and compliant deformation.

[0070] The NSGA-III algorithm uses the aforementioned five sub-objective functions as independent optimization guides, iteratively evolving the population through selection, crossover, and mutation operations, with a convergence threshold set to... The optimization process incorporates topological filtering operators throughout to eliminate numerical singularities and grid dependencies.

[0071] The NSGA-III method ultimately outputs a set of optimal solutions and the corresponding optimal compliant compensation mechanism topology, optimal drive pose reference parameters, and optimal stiffness-damping parameter matrix. , Meanwhile, all intermediate iteration data generated under this model will be structured and stored to form an offline optimization experience pool, which can be used for warm-up or transfer learning of subsequent models.

[0072] Real-time response mode: Online adaptive optimization based on deep reinforcement learning.

[0073] When the meta-controller determines that the optimization task possesses the following characteristics, it enters real-time response mode: the operating environment exhibits known dynamic changes, the system has high real-time requirements for optimization decisions, or the system has accumulated a certain amount of historical operation data. In this mode, the system invokes an online optimization engine based on deep reinforcement learning to reconstruct the collaborative topology optimization problem into a sequential decision problem within a deep reinforcement learning framework.

[0074] First, a deep reinforcement learning agent is constructed, whose state space... Includes the current topological density field Driving pose error and its rate of change End contact force Real-time deformation of the compliant compensation mechanism Current torque deviation The target value corresponding to the nearest optimal solution to the current state, extracted from the offline optimization experience pool, is used as a reference signal. .

[0075] To achieve effective fusion of high-dimensional topology optimization material density variables and low-dimensional physical state vectors in a unified state space, this embodiment employs principal component analysis (PCA) to perform offline dimensionality reduction on the topology optimization material density variables. It should be noted that the current topology density field... This refers to the aforementioned topology optimization material density variable. At the instantaneous distribution state at time t, both represent the same physical quantity.

[0076] In the exploration-first mode, a training set is constructed using high-quality topology-optimized material density samples generated during the NSGA-III iteration. The first 64 principal components are extracted through matrix singular value decomposition to obtain the projection matrix. With mean vector .

[0077] When the reinforcement learning agent interacts with the environment in real time, the current topological density field will be... Flattened into a one-dimensional vector The projection matrix obtained in the offline stage With mean vector Mapped to a low-dimensional latent vector :

[0078] .

[0079] Low-dimensional topological latent vectors With driving pose error and its rate of change End contact force Real-time deformation of the compliant compensation mechanism Current torque deviation and reference signal By concatenating the vectors, a fixed-dimensional state vector is obtained. :

[0080] ,

[0081] The policy network employs a multilayer perceptron structure, with the number of input layer neurons and the state vector... The dimensions are consistent, and the low-dimensional action vector is output by the output layer after being mapped layer by layer through two hidden layers; each hidden layer contains 256 neurons and uses the ReLU activation function.

[0082] The action space contains low-dimensional potential adjustments to the material density variables for topology optimization. , drive pose adjustment amount Adjustment amount of time-varying stiffness matrix and the adjustment amount of the time-varying damping matrix The low-dimensional topology adjustment amount output by the policy network. via projection matrix Inverse transformation to full-dimensional topological density adjustment Then, through tensor reconstruction, it is restored to have the same dimension as the topological density field. Driven pose adjustment amount Adjustment amount of time-varying stiffness matrix With time-varying damping matrix adjustment The outputs of the independent branches of the policy network are directly mapped to the driving pose adjustment amounts. Adjustment amount of time-varying stiffness matrix With time-varying damping matrix adjustment .

[0083] Action space The adjustment amount is defined as the optimization variable: , , , .

[0084] reward function ,in To constrain the penalty coefficient for violations.

[0085] The agent is trained using the Proximal Policy Optimization (PPO) algorithm and initializes its behavior using data from an offline optimization experience pool generated by the exploration-first mode. This supervised learning allows the policy network to pre-mimic the optimal or near-optimal solutions generated by NSGA-III optimization, significantly shortening the cold-start adaptation time after online deployment. This mode continuously collects real-world data during ongoing operations, fine-tuning the policy network online and enabling continuous evolution of the optimization policy. After sufficient training, the agent can output optimal topology adjustment commands, driving pose correction commands, and compliance characteristic adjustment commands within milliseconds based on real-time perceived states.

[0086] Hybrid Collaborative Mode: Joint optimization of offline guidance and online refinement.

[0087] When the meta-controller determines that the optimization task has high precision requirements, and there is prior knowledge that can be used for guidance, but the environment still experiences significant dynamic changes, it enters the hybrid collaborative mode. This mode is a deep fusion of the first two modes, and its core idea is: to use offline optimization results as high-quality initial priors, and online reinforcement learning as a fine-tuning mechanism.

[0088] The specific operation process of this mode is as follows:

[0089] First, based on the constraints and target weights of the current operating conditions, an exploration-first mode is invoked to perform a finite number of NSGA-III optimizations, quickly producing a relatively good, but not necessarily globally optimal, solution set. Then, several leading solutions are selected from this solution set, and their corresponding state-action pairs are used as expert teaching data. Behavioral cloning is used to warm up the deep reinforcement learning policy network. After warm-up, the system switches to real-time response mode, but the action space of the reinforcement learning agent is now constrained to a finite range around the expert solution (e.g., topology density adjustment does not exceed ±0.1, and driving pose adjustment does not exceed ±5%). An expert bias penalty is added to its reward function to encourage the agent to explore locally near the expert solution rather than search globally. As the system accumulates data in actual operations, the agent gradually learns better strategies that surpass the initial expert solution. At this point, the constraints can be dynamically relaxed, achieving a smooth transition from "offline guidance" to "online surpassing".

[0090] Step 3: Closed-loop compliant compensation and pose coordination control based on optimization results.

[0091] The core of compliant compensation and pose coordination control is to inject the optimal solution output from step two into a coordination control law that integrates impedance control and pose closed-loop control. The expression of this control law is:

[0092] ,

[0093] in, For servo drive torque, As the optimal driving pose reference, The speed that is the optimal driving pose reference. To drive the generalized coordinates of the joints, To drive the generalized velocity of the joint, For position scaling gain, For velocity differential gain, Gain for front-end pose error compensation. External contact force feedback compensates for gain. For compliant deformation feedback gain, This is a six-dimensional pose error vector. For external contact reaction force, This represents the amount of compliant deformation.

[0094] This is a key feature of the present invention, which utilizes the deformation amount obtained in step two through optimization. As a direct feedback quantity, it enables proactive regulation of compliant behavior.

[0095] The essence of a control law lies in the fact that it is not simply a superposition of multiple independent control terms, but rather through... , and The coupling operation of the three components unifies the pose compensation of the driving system, the interaction of external forces, and the compliant deformation of the structure itself within a single control framework, thereby achieving synchronous and coordinated control of the three components: force, pose, and compliance.

[0096] To implement the aforementioned control law, the system acquires key data in real time through a multi-source sensing and acquisition module: a six-dimensional force sensor is used to collect the contact force between the bit and the workpiece. A non-contact pose detection unit (such as binocular vision combined with subpixel edge detection algorithm) is used to measure the end-effector pose error in real time. The drive torque is obtained by using a torque encoder and a servo motor encoder respectively. With joint motion data All data is input into the control law at a sampling frequency of no less than 1kHz. After real-time calculation, the servo drive output and compliance compensation are dynamically corrected to achieve rapid and accurate elimination of posture deviation and force deformation deviation.

[0097] When step two adopts a hybrid collaborative mode, the control law in... The item used Not only do the measurements originate from real-time sensors, but the target value or desired range can also be dynamically set by the deep reinforcement learning agent based on the current state, thereby achieving adaptive adjustment of the control target. The overall execution flowchart of the system is as follows: Figure 4 As shown.

[0098] To verify the effectiveness of this invention, we systematically verified its advantages over existing technologies by comparing it with traditional electric screwdrivers through physical testing. The experimental results are shown in Tables 1 and 2. Table 1 compares the overall performance indicators of the system, and Table 2 compares the adaptability of workpiece position deviation. The two experimental results show that this invention is significantly superior to traditional electric screwdrivers in terms of position accuracy, torque control, and compliance compensation, and has significant generational technological advantages and engineering application value.

[0099] Table 1 Comparison of Overall System Performance Indicators

[0100]

[0101] Table 2 Comparison of workpiece position deviation adaptability

[0102]

[0103] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for topology optimization in collaboration between intelligent electric screwdriver compliance compensation and drive pose, characterized in that: Includes the following steps: A rigid-flexible coupling dynamic model of the entire intelligent electric screwdriver is constructed. The rigid-flexible coupling dynamic model represents the coupling relationship between rigid driving motion, compliant deformation and external contact reaction force. A multi-objective collaborative topology optimization model is established with the topology optimization material density variable of the driving pose variable and the compliant compensation mechanism as the joint optimization variable; The multi-objective cooperative topology optimization model is solved by adaptive cooperative optimization based on the adaptive meta-optimization framework with meta-controller, and the optimal structural topology configuration, optimal driving pose reference, optimal time-varying stiffness matrix and optimal time-varying damping matrix are obtained. The meta-controller adaptively invokes the optimization algorithm based on the optimization task status; The optimal structural topology, the optimal driving pose reference, the optimal time-varying stiffness matrix, and the optimal time-varying damping matrix are used as the optimal solution to inject into the cooperative control law, and closed-loop compliance compensation and pose cooperative control are performed. The cooperative control law includes a compliance deformation direct feedback term. The optimization objectives of the multi-objective collaborative topology optimization model include structural lightweighting, driving pose positioning accuracy, structural mechanical strength, tightening torque stability, and compliance compensation accuracy. The objective function of the multi-objective cooperative topology optimization model is: ， in, For topology optimization of material density variables, To drive pose variables, This is a six-dimensional pose error vector. To optimize the volume of the structure, The initial structural volume, Let be the second norm of the six-dimensional pose error vector. Let be the second norm of the maximum allowable deformation of the structure. The maximum stress of the structure, The stress yield limit of the material. For servo drive torque, The rated tightening torque, This represents the maximum value of the compliance deformation. This is the allowable value for compliance deformation. , , , as well as Weighting coefficients to satisfy the normalization condition; The expression for the cooperative control law is: ， in, For servo drive torque, As the optimal driving pose reference, The speed that is the optimal driving pose reference. To drive the generalized coordinates of the joints, To drive the generalized velocity of the joint, For position scaling gain, For velocity differential gain, Gain for front-end pose error compensation. External contact force feedback compensates for gain. For compliant deformation feedback gain, This is a six-dimensional pose error vector. For external contact reaction force, This is the amount of compliant deformation.

2. The intelligent electric screwdriver compliance compensation and drive pose collaborative topology optimization method as described in claim 1, characterized in that: The specific steps for constructing a rigid-flexible coupling dynamic model of an intelligent electric screwdriver are as follows: The intelligent electric screwdriver is disassembled into a rigid drive mechanism and a compliant compensation mechanism. The rigid drive mechanism includes a servo drive motor, a transmission component, and a bit end actuator, while the compliant compensation mechanism includes a flexible compensation actuator and a flexible hinge component. Based on the second type of Lagrange equations, a fundamental dynamic equation is established to characterize the coupling relationship between rigid-driven motion, compliant deformation, and external contact reaction force.

3. The intelligent electric screwdriver compliance compensation and drive pose collaborative topology optimization method as described in claim 2, characterized in that: The steps for constructing a rigid-flexible coupling dynamic model of an intelligent electric screwdriver also include: A six-dimensional pose error vector is introduced, and the mapping relationship between pose deviation and compliant deformation of rigid drive mechanism is established through Jacobian matrix; By introducing time-varying stiffness and time-varying damping matrices, the basic dynamic equations are extended to obtain a complete rigid-flexible coupling dynamic model.

4. The intelligent electric screwdriver compliance compensation and drive pose cooperative topology optimization method as described in claim 1, characterized in that: The optimized task status of the meta controller includes task characteristics, resource status characteristics, historical optimized performance characteristics, and user preferences.

5. The intelligent electric screwdriver compliance compensation and drive pose collaborative topology optimization method as described in claim 1, characterized in that: The optimization algorithms for adaptive invocation of the meta-controller include deep offline optimization algorithms based on NSGA-III, online adaptive optimization algorithms based on deep reinforcement learning, or hybrid collaborative optimization algorithms.

6. The intelligent electric screwdriver compliance compensation and drive pose cooperative topology optimization method as described in claim 5, characterized in that: The deep offline optimization algorithm uses the parameterized level set method to characterize the macroscopic boundary of the compliant compensation mechanism, and uses the solid isotropic microstructure penalty model to characterize the material density distribution inside the compliant compensation mechanism. It solves the dynamic response through the extended finite element method and iteratively evolves to convergence.

7. The intelligent electric screwdriver compliance compensation and drive pose cooperative topology optimization method as described in claim 5, characterized in that: An online adaptive optimization algorithm is used to construct a deep reinforcement learning agent. The agent is trained using the proximal policy optimization (PPO) algorithm and initialized with behavior cloning using offline optimization experience pool data. The state space of the agent includes real-time perceived operating parameters, and the action space of the agent includes topology density adjustment, driving pose adjustment, time-varying stiffness matrix adjustment, and time-varying damping matrix adjustment.

8. The intelligent electric screwdriver compliance compensation and drive pose cooperative topology optimization method as described in claim 5, characterized in that: The hybrid collaborative optimization algorithm first generates expert solutions through offline optimization using finite algebras, and then uses the expert solutions to preheat and refine the online reinforcement learning strategy locally.

Images