Robot grinding and polishing process parameter self-adaptive decision-making method and system

Abstract

The invention relates to a robot grinding and polishing process parameter adaptive decision-making method and a robot grinding and polishing process parameter adaptive decision-making system. The method comprises the following steps: after a track of each grinding and polishing process is finished, judging whether the track enters an excellent experience pool or not according to a return value of the track; wherein the track is composed of a state, an action and a return value, and the state, the action and the return value represent the machining allowance, the grinding parameter change and the grinding quality of the workpiece in the grinding and polishing process respectively; training an optimization network for the tracks in the excellent experience pool; optimizing the action of the track in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool; and optimizing and updating the optimized network according to the optimized excellent experience pool. According to the method, the technological parameters can be rapidly and adaptively adjusted among different grinding and polishing tasks through a small number of training samples, and meanwhile, the high material removal precision can be achieved on the specific grinding and polishing tasks.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for adaptively deciding robot grinding and polishing process parameters. Background Art

[0002] The surface profile accuracy and surface roughness of aircraft engine blades are key factors affecting the aerodynamic performance of the engine, improving the stability of the engine operation, reducing the stress concentration of the blades themselves, and improving the corrosion resistance. Grinding and polishing are usually required to ensure the final surface quality and surface state of the parts. The material properties, profile characteristics and processing allowances of the repaired aircraft engine blades are different, and the grinding ability of the grinding and polishing tools is dynamically weakened during the grinding and polishing process. It is necessary to rely on the experience of the grinding and polishing operators to continuously adjust the grinding and polishing process parameters to meet different grinding and polishing requirements.

[0003] In order to get rid of the dependence of the grinding and polishing process of repaired blades on the experience of the grinders and polishers, and to improve the accuracy of material removal and the consistency of surface grinding and polishing quality, in the existing disclosed technologies, machine learning methods or methods of establishing grinding and polishing material removal models are used to achieve the optimization of grinding process parameters and the prediction of grinding and polishing quality.

[0004] In the existing public papers: "Prediction of Material Removal Rate in Chemical Mechanical Polishing Based on Residual Convolutional Neural Network", a data-driven material removal rate prediction method for mechanical chemical polishing was proposed. Convolutional neural networks and residual convolutional neural networks were developed to train material removal rate prediction models, respectively, and a high material removal rate prediction accuracy was obtained. "Combining physical models and dual deep neural networks to accurately predict and compensate for machining errors of large parts with time-varying characteristics" and other papers proposed a hybrid method of "physical model" + "artificial intelligence" (LSTM+CNN). The physical model is used to calculate the theoretical material removal amount, and the artificial intelligence model is used to predict the system stiffness at real time and adjust the machining force. A high machining error prediction and compensation accuracy is achieved in the grinding of large shaft parts with different contours.

[0005] In the existing public patent: A method and device for predicting the cutting depth of robot belt grinding and polishing for complex curved surfaces (patent number: CN110119566B), a method for predicting the cutting depth of robot belt grinding and polishing is disclosed. The method obtains the stress distribution model between the contact wheel and the workpiece based on the finite element analysis method and the Hertz contact model. Combined with the grinding time and feed speed of each path point, the cutting depth prediction model of each path point is obtained. And based on the cutting depth prediction model, the normal force and grinding speed in the grinding process are accurately controlled. The existing patent: A method for predicting the surface roughness of elastic workpiece robot grinding and polishing based on virtual samples (patent number: CN113255732B) discloses a method for predicting the surface roughness of robot grinding and polishing. This patent establishes a mapping relationship model between the process parameters such as grinding head speed, grinding pressure, particle size, edge distance, grinding inclination angle and their corresponding roughness values ​​in the existing robot grinding and polishing process.

[0006] These patents and papers that have been published combine machine learning algorithms with mathematical models for material removal, and have conducted certain research in the prediction of material removal and surface roughness during the grinding and polishing process, as well as the optimization of grinding and polishing process parameters. However, most traditional machine learning algorithms or mathematical models for material removal can only determine process parameters for specific parts, grinding and polishing requirements, and grinding and polishing conditions, and obtain accurate material removal. In addition, traditional machine learning algorithms require a large amount of data training to obtain a grinding and polishing process parameter decision model with better performance. In actual grinding and polishing operations, parts with different material properties and characteristics are often required to undergo a variety of grinding and polishing processes to achieve the final grinding and polishing technical requirements, and the grinding capacity of grinding consumables is dynamically reduced during the grinding and polishing process. Models trained using traditional machine learning algorithms cannot quickly adapt to different grinding tasks with a small number of training samples. When there are differences between the test set and the training set faced by the model, the model's ability to make decisions on process parameters is greatly reduced, and training often needs to be restarted. Summary of the invention

[0007] To this end, the present invention provides a method and system for adaptive decision-making of robot grinding and polishing process parameters, which can quickly and adaptively adjust process parameters between different grinding and polishing tasks through a small amount of training samples, while achieving higher material removal accuracy on specific grinding and polishing tasks.

[0008] In order to solve the above technical problems, the present invention provides a robot grinding and polishing process parameter adaptive decision method, comprising:

[0009] After each grinding and polishing process track is completed, it is determined whether it enters the excellent experience pool according to the reward value of the track; wherein the track is composed of state, action and reward value, and the state, action and reward value respectively represent the machining allowance, grinding parameter change and grinding quality of the workpiece in the grinding and polishing process;

[0010] Training and optimizing a network for trajectories in the excellent experience pool;

[0011] Optimizing the actions of the trajectories in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool;

[0012] The optimized network is optimized and updated according to the optimized excellent experience pool.

[0013] In one embodiment of the present invention, judging whether the trajectory enters the excellent experience pool according to the reward value of the trajectory includes:

[0014] When the reward value of the trajectory is greater than the average reward value of all trajectories in the current excellent experience pool, the trajectory with the smallest reward value in the current excellent experience pool is discarded, and the reward value of the trajectory is added to the current excellent experience pool.

[0015] In one embodiment of the present invention, the objective function L of the optimization network is clip (θ) is:

[0016]

[0017] In the formula, is the importance sampling weight, representing the ratio of the new strategy to the old strategy; π θ (a t s t ) means that at a given s t New strategy π in state θ Produce action a t The probability of θold (a t s t ) means that at a given s t The old policy π in the state θ Produce action a t The probability, p t The larger the (θ) is, the greater the update gradient of the new strategy relative to the old strategy; clip(p t (θ),1-ε,1+ε) is a clipping function used to control the policy update gradient, which limits the policy update gradient p t (θ) is in the interval [1-ε,1+ε], where ε is a hyperparameter; is the valuation of the advantage function; it represents the value of the current state and action With status s t The average value of The difference is used to measure the quality of the current state and action relative to the average level.

[0018] In one embodiment of the present invention, the loss function L of the optimization network isO(φ) As shown below:

[0019]

[0020] In the formula, O(φ) is the optimized network, a t For action.

[0021] In one embodiment of the present invention, optimizing the action of the trajectory in the excellent experience pool according to the optimization network includes:

[0022] Update the importance sampling weights of the optimized network to:

[0023]

[0024] In the formula, β is the optimization factor constrained between (0,1); Represents the distribution probability of the current network strategy.

[0025] In one embodiment of the present invention, optimizing and updating the optimized network includes:

[0026] The updated objective function is shown as follows:

[0027]

[0028] Where C1 and C2 are two constant hyperparameters, which are the deviation terms of the state value function and represent the entropy of the strategy; is the state value function after restriction; V target is the target state function; is the current state value function.

[0029] In one embodiment of the present invention, the restricted state value function as follows:

[0030]

[0031] Where V πθ (s t-1 ) is the state value function at the previous moment; the difference is limited to [-ε,ε].

[0032] The present invention also provides a robot grinding and polishing process parameter adaptive decision-making system, comprising:

[0033] A judgment module is used to judge whether each grinding and polishing process trajectory enters the excellent experience pool according to the reward value of the trajectory after the trajectory ends; wherein the trajectory is composed of a state, an action and a reward value, and the state, the action and the reward value respectively represent the machining allowance of the workpiece in the grinding and polishing process, the grinding parameter change and the grinding quality of the workpiece;

[0034] A training module, used for training and optimizing a network for trajectories in the excellent experience pool;

[0035] An action optimization module, used for optimizing the action of the trajectory in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool;

[0036] An updating module is used to optimize and update the optimized network according to the optimized excellent experience pool.

[0037] The above technical solution of the present invention has the following advantages compared with the prior art:

[0038] The method and system for adaptive decision-making of robot grinding and polishing process parameters described in the present invention is based on a meta-reinforcement learning algorithm that combines a model-agnostic meta-learning algorithm with a proximal strategy optimization algorithm that optimizes action selection based on its own excellent experience. This method can not only ensure the accuracy of material removal in specific blade repair tasks, but also quickly adapt to different repair blades and grinding and polishing tasks through a small number of training samples, and adaptively adjust the grinding and polishing process parameters to achieve accurate material removal. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.

[0040] Figure 1 It is a flow chart of the adaptive decision-making method of robot grinding and polishing process parameters of the present invention.

[0041] Figure 2 It is an adaptive decision-making system for grinding and polishing process parameters based on meta-reinforcement learning algorithm.

[0042] Figure 3 It is MAML-PPO BE Schematic diagram of algorithm training principle.

[0043] Figure 4 Yes PPO BE Schematic diagram of algorithm training principle. DETAILED DESCRIPTION

[0044] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0045] In the present invention, "several" means one or more, "multiple" means more than two, "greater than", "less than", "exceed" and the like are understood to exclude the number itself; "above", "below", "within" and the like are understood to include the number itself. In the description of the present invention, if there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0046] Example 1

[0047] Reference Figure 1 As shown, this embodiment provides a method for adaptively deciding robot grinding and polishing process parameters, including:

[0048] S1. After the trajectory of each grinding and polishing process ends, determine whether it enters the excellent experience pool based on the reward value of the trajectory; wherein the trajectory is composed of state, action and reward value, and the state, action and reward value respectively represent the machining allowance of the workpiece in the grinding and polishing process (i.e., the difference between the current surface state of the workpiece and the target surface state), the change of grinding parameters (grinding pressure, speed, angle, etc.) and the grinding quality of the workpiece (dimensional accuracy, surface finish, etc.).

[0049] Wherein, judging whether the trajectory enters the excellent experience pool according to the reward value of the trajectory includes:

[0050] When the reward value of the trajectory is greater than the average reward value of all trajectories in the current excellent experience pool, the trajectory with the smallest reward value in the current excellent experience pool is discarded, and the reward value of the trajectory is added to the current excellent experience pool.

[0051] Specifically, the trajectory with higher return value is (s t ,a t ,r t ,s t+1 ,a t+1 ,r t+1 ……s T ,a T ,r T ) in the form of excellent experience pool P BE Inside, the state is s T 、Action is a T and the return value is r T ; After each trajectory ends, determine whether the reward value of the trajectory can be put into the excellent experience pool P BE The details are as follows:

[0052]

[0053] In the formula, Indicates excellent experience pool P BE The mean return of the m trajectories in G n represents the reward of the nth trajectory. When the reward of the m+1th trajectory is greater than the mean reward of the m trajectories, the excellent experience pool P BE The trajectory with the smallest reward value is discarded, and the m+1th trajectory is put into the excellent experience pool, so that the optimization network O(φ) is closest to the best sample trajectory when optimizing action selection.

[0054] In this embodiment, the optimization network O(φ) is based on MAML-PPO BE Algorithm, MAML-PPO BE The algorithm training principle is as follows Figure 3 As shown, the model defines a task set T i ~p(T), loss parameter network f θa 、f θc And initialize the loss parameters and At the beginning, a batch of tasks T1-T are sampled from the task set m , extract k trajectories for each sampled task, and use its own excellent experience to optimize the proximal policy optimization algorithm (PPO) for action selection BE ) for training and the network loss and Return to the meta-learner. Find the optimal parameter θ through gradient descent a ′ and θ c ′, so that the total loss on m tasks Minimum. When extracting the next batch of tasks T m+1 -T n When , a one-dimensional update is performed, that is, from the parameter θ a ′ and θ c ′Start to minimize the network loss on the new task using the gradient descent method, start training from a better position, and optimize the convergence speed and training results.

[0055] In the learning and training process of the traditional PPO algorithm, the agent randomly selects the next action to update, which reduces the pertinence and efficiency of learning. In order to guide the agent to learn from better experience trajectories and improve the utilization rate of excellent samples, the present invention uses a proximal strategy optimization algorithm (PPO) based on its own excellent experience to optimize action selection. BE )PPO BE The algorithm continues to use the actor-critic framework in the conventional PPO algorithm. The critic network evaluates the current strategy and provides state and action value functions, which the actor network uses to update the objective function. During the training process, the actor network selects and executes action a according to the previous strategy and the guidance of the optimized action selection network O(φ). t, generating sample experience P = (st, at, r t ,s t+1 ), and store it in the experience pool. After each trajectory ends, determine whether the trajectory is eligible to be placed in the excellent experience pool P BE and through the excellent experience pool P BE Synchronously update the optimization action selection network O(φ), PPO BE The training principle of the algorithm is as follows Figure 4 shown.

[0056] On this basis, the PPO algorithm adds a bias term and an entropy regularization term to the state value function estimate to reward exploration. The optimized function is shown in the following formula:

[0057]

[0058] Where C1 and C2 are two constant hyperparameters, which are the deviation terms of the state value function and represent the entropy of the strategy; L clip (θ) is the objective function of the PPO algorithm; (V πθ (s t )-V target ) 2 represents the deviation term of the state function; H(s t ,π θ ) represents the entropy of the policy.

[0059] In the learning and training process of the traditional PPO algorithm, the agent randomly selects the next updated action, which reduces the pertinence and efficiency of learning. In order to guide the network to learn from better experience trajectories, improve sample utilization, and optimize cumulative benefits, a proximal strategy optimization algorithm (PPO) based on its own excellent experience optimization action selection is proposed. BE ).

[0060] The objective function L of the optimization network clip (θ) is:

[0061]

[0062] In the formula, is the importance sampling weight, representing the ratio of the new strategy to the old strategy; π θ (a t s t ) means that at a given s t New strategy π in state θ Produce action a t The probability of θold (a t s t ) means that at a given s t The old policy π in the state θ Produce action at The probability, p t The larger the (θ) is, the greater the update gradient of the new strategy relative to the old strategy; clip(p t (θ),1-ε,1+ε) is a clipping function used to control the policy update gradient, which limits the policy update gradient p t (θ) is in the interval [1-ε,1+ε], where ε is a hyperparameter; is the valuation of the advantage function; it represents the value of the current state and action With status s t The current state value function under The difference is used to measure the quality of the current state and action relative to the average level.

[0063] S2. Train and optimize the network O(φ) for the trajectories in the excellent experience pool.

[0064] The loss function L of the optimization network is O(φ) As shown below:

[0065]

[0066] In the formula, O(φ) is the optimized network, a t is the action; N is the number of trajectories.

[0067] S3, optimizing the actions of the trajectories in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool.

[0068] Update the importance sampling weights of the optimized network to:

[0069]

[0070] In the formula, β is the optimization factor constrained between (0,1); Represents the distribution probability of the current network strategy.

[0071] S4. Optimizing and updating the optimized network according to the optimized excellent experience pool.

[0072] In PPO BE In the algorithm, the objective function used to update the actor network can be expressed as follows:

[0073]

[0074] PPO BE The algorithm continues the idea of ​​the PPO algorithm and introduces a clipping function in the objective function to limit the importance sampling weight r(φ,β) to the interval [1-ε,1+ε]. The clipping function is shown as follows:

[0075]

[0076] In PPO BE In the algorithm, when the critic network is updated, the value function V of the current state is πθ (s t ) and the state value function V at the previous moment πθ (s t-1 ) is limited to [-ε,ε], thus obtaining the restricted state value function As shown below:

[0077]

[0078] The state value function after restriction The calculated error and the current state value function V πθ (s t ) The error calculated is compared to the minimum value, which can effectively reduce the loss of the critic network. BE The final complete objective function of the algorithm is shown as follows:

[0079]

[0080] Where C1 and C2 are two constant hyperparameters, which are the deviation terms of the state value function and represent the entropy of the strategy; is the state value function after restriction; V target is the target state function; is the current state value function.

[0081] The framework structure of the adaptive decision-making system for grinding and polishing process parameters based on the meta-reinforcement learning algorithm is as follows: Figure 2 As shown, it specifically includes a grinding and polishing process parameter adaptive decision system, a state acquisition system, and a grinding and polishing process parameter execution system.

[0082] State acquisition system: used to collect parameters such as the initial roughness value of the repaired blade, the remaining amount to be processed, the processing technical requirements, the material characteristics, and the part characteristics, and used as the input of the adaptive decision-making system for grinding and polishing process parameters;

[0083] Process parameter adaptive decision system: Based on the robot grinding and polishing process parameter adaptive decision method, the corresponding action output is calculated according to the state input of the state acquisition system;

[0084] Grinding and polishing process parameter execution system: Based on the results calculated by the grinding and polishing process parameter adaptive decision-making system, the grinding and polishing contact force, grinding and polishing speed, grinding and polishing time and other parameters are controlled through various actuators to select the type of consumables and implement the grinding and polishing operation.

[0085] This method is based on a model-agnostic meta-learning algorithm (MAML) and a proximal policy optimization algorithm (PPO) that optimizes action selection based on its own excellent experience. BE ) combined with a meta-reinforcement learning algorithm. This method can not only ensure the accuracy of material removal in the specific blade repair task, but also quickly adapt to different blade repair and grinding and polishing tasks through a small number of training samples, and adaptively adjust the grinding and polishing process parameters to achieve accurate material removal.

[0086] Example 2

[0087] Based on the same inventive concept, this embodiment provides a robot grinding and polishing process parameter adaptive decision-making system, and its principle of solving the problem is similar to the robot grinding and polishing process parameter adaptive decision-making method, and the repeated parts will not be repeated.

[0088] This embodiment provides a robot grinding and polishing process parameter adaptive decision system, including:

[0089] A judgment module is used to judge whether each grinding and polishing process trajectory enters the excellent experience pool according to the reward value of the trajectory after the trajectory ends; wherein the trajectory is composed of a state, an action and a reward value, and the state, the action and the reward value respectively represent the machining allowance of the workpiece in the grinding and polishing process, the grinding parameter change and the grinding quality of the workpiece;

[0090] A training module, used for training and optimizing a network for trajectories in the excellent experience pool;

[0091] An action optimization module, used for optimizing the action of the trajectory in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool;

[0092] An updating module is used to optimize and update the optimized network according to the optimized excellent experience pool.

[0093] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0097] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A robot grinding and polishing process parameter adaptive decision method, characterized in that: include: After each grinding and polishing process track is completed, it is determined whether it enters the excellent experience pool according to the reward value of the track; wherein the track is composed of state, action and reward value, and the state, action and reward value respectively represent the machining allowance, grinding parameter change and grinding quality of the workpiece in the grinding and polishing process; Training and optimizing a network for trajectories in the excellent experience pool; Optimizing the actions of the trajectories in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool; The optimized network is optimized and updated according to the optimized excellent experience pool.

2. The method for adaptively determining the parameters of a robot grinding and polishing process according to claim 1, characterized in that: The step of judging whether the trajectory enters the excellent experience pool according to the reward value of the trajectory includes: When the reward value of the trajectory is greater than the average reward value of all trajectories in the current excellent experience pool, the trajectory with the smallest reward value in the current excellent experience pool is discarded, and the reward value of the trajectory is added to the current excellent experience pool.

3. The method for adaptively determining robot grinding and polishing process parameters according to claim 1, characterized in that: The objective function L of the optimization network clip (θ) is: In the formula, is the importance sampling weight, representing the ratio of the new strategy to the old strategy; π θ (a t |s t ) means that at a given s t New strategy π in state θ Produce action a t The probability of θold (a t |s t ) means that at a given s t The old policy π in the state θ Produce action a t The probability, p t The larger the (θ) is, the greater the update gradient of the new strategy relative to the old strategy; clip(p t (θ), 1-ε, 1+ε) is the clipping function used to control the policy update gradient, which limits the policy update gradient p t (θ) is in the interval [1-ε, 1+ε], where ε is a hyperparameter; is the valuation of the advantage function; it represents the value of the current state and action With status s t The average value of The difference is used to measure the quality of the current state and action relative to the average level.

4. The method for adaptively determining the parameters of a robot grinding and polishing process according to claim 1, characterized in that: The loss function L of the optimization network is O(φ) As shown below: In the formula, O(φ) is the optimized network, a t For action.

5. A robot grinding and polishing process parameter adaptive decision method according to claim 1, characterized in that: Optimizing the action of the trajectory in the excellent experience pool according to the optimization network includes: Update the importance sampling weights of the optimized network to: In the formula, β is the optimization factor constrained between (0,1); Represents the distribution probability of the current network strategy.

6. The method for adaptively determining robot grinding and polishing process parameters according to claim 3, characterized in that: Optimizing and updating the optimized network includes: The updated objective function is shown as follows: Where C1 and C2 are two constant hyperparameters, which are the deviation terms of the state value function and represent the entropy of the strategy; is the state value function after restriction; V target is the target state function; is the current state value function.

7. The method for adaptively determining robot grinding and polishing process parameters according to claim 1, characterized in that: The state value function after the restriction as follows: Where V πθ (s t-1 ) is the state value function at the previous moment; the difference is limited to [-ε,ε].

8. A robot grinding and polishing process parameter adaptive decision system, characterized in that: include: A judgment module is used to judge whether each grinding and polishing process trajectory enters the excellent experience pool according to the reward value of the trajectory after the trajectory ends; wherein the trajectory is composed of a state, an action and a reward value, and the state, the action and the reward value respectively represent the machining allowance of the workpiece in the grinding and polishing process, the grinding parameter change and the grinding quality of the workpiece; A training module, used for training and optimizing a network for trajectories in the excellent experience pool; An action optimization module, used for optimizing the action of the trajectory in the excellent experience pool according to the optimization network to obtain an optimized excellent experience pool; An updating module is used to optimize and update the optimized network according to the optimized excellent experience pool.

Images