Collaborative optimization method and system for intelligent spinning process path

Abstract

The invention provides a collaborative optimization method and system for an intelligent spinning process path, and the method comprises the steps: selecting different spinning process parameters according to an orthogonal design, and obtaining a physical field state and a shape quality parameter through finite element simulation and a spinning experiment; establishing a neural network model, determining a mapping relation between a physical field state and a shape quality parameter when the part is formed, determining a reward function, and training the neural network model; building a deep reinforcement learning framework, embedding the deep reinforcement learning framework into the spinning finite element model, and building a spinning forming process parameter control model; and an intelligent agent in the spinning forming process parameter control model continuously interacts with the environment, spinning process parameters are continuously adjusted according to return values, accumulated rewards in the forming process are maximized, and the optimal spinning process path is output. The spinning process parameter optimization method can solve the problems that an existing spinning process parameter optimization method is low in precision, small in application range and limited in optimization parameter.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a collaborative optimization method and system for an intelligent spinning process path. Background Art

[0002] With the continuous improvement of the performance requirements of parts in the fields of aerospace, national defense, military industry, transportation, etc., parts manufacturing is developing towards high performance, low energy consumption, and long life. Among them, complex rotating thin-walled parts, such as aircraft engine housings, missile fairings, and automobile wheels, as key parts in the above fields, have a wide range of application needs. As a continuous local plastic forming process, spinning forming has been widely used in the manufacture of complex rotating thin-walled parts due to its advantages of high process flexibility, simple mold structure, low forming load, and high production efficiency. However, the shape quality of parts is affected by a variety of process parameters, such as spinning temperature, spinning wheel feed ratio, thinning rate, etc., and the relationship between them is highly nonlinear, which brings great challenges to the manufacture of high-quality parts. Therefore, how to quickly and accurately optimize the above process parameters is of great significance to improving product quality, reducing production costs, and shortening the R&D cycle.

[0003] In view of the problem of forming process parameter optimization, the Chinese invention patent application with publication number CN118133615A discloses a method for optimizing the process parameters of fiber metal super-hybrid circular tube high-power spinning. The relationship equation between the spinning process parameters and the interface shear strength in the fiber circular tube is established through the spinning orthogonal experiment, and then the corresponding spinning process parameters when the interface shear strength of the circular tube is optimal are determined by combining the finite element simulation. However, the parameter dimensions collected by this method are limited, and the corresponding relationship equation is not accurate, and only the approximate process parameter range can be given. The Chinese invention patent application with publication number CN103182457A discloses a method for optimizing the spinning process parameters of aerospace thin-walled parts based on data mining. It uses the spinning blank material parameters and the spinning process parameters as input variables, and the part quality index as the output variable. The XGboost algorithm is used to establish a spinning thin-walled part quality prediction model to achieve the optimization of the spinning spinning process parameters. However, the model establishment requires a large amount of predefined label data and the model is only applicable to this type of process, which has certain limitations, and the algorithm itself will take up a lot of memory space when processing large-scale data, and the calculation cost is high. A Chinese invention patent application with publication number CN118607288A discloses a dynamic modeling prediction and online process optimization method for the spinning of curved parts. The method discretizes the spinning finite element simulation process into multiple continuous stages, obtains the forming results of different stages under different spinning process parameters, and establishes a proxy model of process parameters and forming results in different forming stages through the response surface method. Then, the optimal process parameters of each stage of spinning are obtained according to the preset optimization goals, so as to achieve process optimization of the whole spinning process. However, the optimization accuracy of the process parameters of the above method depends on the degree of discreteness of spinning. A larger degree of discreteness will cause the optimization process to become significantly longer, and can only achieve optimization of part geometric accuracy, ignoring the influence of process parameters on the mechanical properties of the workpiece. Summary of the invention

[0004] The purpose of the present invention is to provide a solution to the problems of low precision, small application scope and limited optimization parameters in the existing spinning process parameter optimization methods in view of the shortcomings of the above-mentioned background technology.

[0005] In order to achieve the above object, the present invention provides a collaborative optimization method for an intelligent spinning process path, comprising the following steps: S1, select different spinning process parameters according to orthogonal design, and then obtain the physical field state of the part during forming and the shape quality parameters of the part after forming through finite element simulation and spinning experiment; S2, establish a neural network model, determine the mapping relationship between the physical field state and the shape quality parameters during part forming, and determine the reward function through the weighted coefficient method to train the neural network model; S3, build a deep reinforcement learning framework, embed the deep reinforcement learning framework into the spinning finite element model through secondary development, and establish a spinning forming process parameter control model; S4, the intelligent agent in the spinning process parameter control model continuously interacts with the environment and continuously adjusts the spinning process parameters according to the reward value to maximize the cumulative reward in the forming process and finally output the optimal spinning process path.

[0006] Further, the spinning process parameters include thinning rate, spindle speed, spinning wheel feed speed, and heating power, and the spinning process parameters are set to fixed values ​​during the forming process; The physical field state includes the average strain of the deformation unit, the average strain rate, and the average forming temperature. When calculating all the physical field states, the stage with the instantaneous strain rate is removed; The morphological quality parameters include part geometry parameters and mechanical properties; the geometry parameters include part roundness, mold fit and theoretical wall thickness difference; the mechanical properties include part hardness, room temperature tensile properties and impact toughness.

[0007] Furthermore, in S2, the physical field state of the part formed by the finite element simulation is used as input, and the shape quality parameters obtained by the spinning experiment are used as output. The error between the weighted sum of the shape quality parameters is obtained by calculation, and then the weight coefficient of the neural network model is updated by back propagation and trained until convergence.

[0008] Furthermore, K-fold cross validation is used when training the neural network model. K-fold cross validation randomly divides the data into K parts, selects K-1 parts as the training set each time, and the remaining 1 part is the test set. Cross-validation is performed K times, and the average of the K times of accuracy is taken as the final evaluation indicator of the neural network model.

[0009] Furthermore, the reward function in S2 Confirmed by the weighted coefficient method, the expression is: ;

[0010] in, is the weight coefficient, The size of is allocated according to the importance of each indicator, and , is the total number of indicators, is the sub-reward item corresponding to the i-th indicator, Output results of mechanical properties in the neural network model. The bigger the reward, the greater the Output results for the geometric shape parameters in the neural network model, The smaller the reward, the greater the degree.

[0011] Furthermore, the deep reinforcement learning framework in S3 is modeled by a Markov decision process and defined as a four-tuple: state s, action a, reward r, and state transition probability distribution function p; state s corresponds to the physical field state of the unit during spinning finite element simulation; action a is the adjustment amount of the spinning process parameters; reward r is the calculated value of the reward function defined in S2, which is used to evaluate the immediate impact of the current action on the shape quality; function p is the probability distribution of transferring to the next physical field state under a given state s and action a.

[0012] Further, the spin forming process parameter control model includes an environment and an agent; The environment includes a spinning finite element model and a reward function; the agent includes a policy network and a value network; The policy network selects actions based on the current state s and continuously optimizes the strategy through the policy gradient method to maximize the cumulative reward. is the sum of the reward values ​​from the start to the end of the simulation, expressed as: ; in, is the discount rate, which is used to determine the priority of short-term returns, ranging from 0 to 1. For the moment, Represents the end moment, represents the time step, Represents the time step When, according to the current state Select and execute the current action After that, the immediate reward value of the environment feedback; The value network is used to evaluate the quality of the current strategy and provide guidance for updating the strategy network.

[0013] Furthermore, the spinning process parameter control model in S4 controls the interaction process through a preset control strategy: when the environment receives the action a of the agent, the environment obtains a reward r and a state s′ at the next moment according to the state s at the current moment, and the agent optimizes the strategy according to the accumulated reward value and determines the next action. During the continuous interaction process, the parameters θ of the strategy network and ω of the value network inside the agent are continuously updated, so that the actions of the agent are more and more in line with expectations, and finally the optimal spinning process path is learned.

[0014] Furthermore, the parameter update process of the strategy network and the value network when the agents interact is as follows: At this moment, the agent executes Action at all times Afterwards, it was observed from the environment Moment Rewards and the new state ; Policy network makes decisions , Represents a strategy, but the agent does not immediately execute the decision The value network scores actions : ; ; in, for The score of the decision in the state, for The score of the decision in the state, is the action value function; Based on this, the network target value is calculated and prediction error values : ; ; Update the value network parameter ω according to the gradient descent method: ; in, is the updated value of the value network parameter, is the gradient descent learning rate, Representation parameters The gradient of Then update the policy network parameters θ according to the gradient ascent method: in, is the updated value of the policy network parameters, is the gradient ascent learning rate, Representation parameters The gradient of Represents the output of the policy network, representing the agent in state Directly select an action .

[0015] The present invention also provides a collaborative optimization system for an intelligent spinning process path, which adopts a collaborative optimization method for an intelligent spinning process path as described above, including a multi-source data acquisition and preprocessing module, a neural network model construction and training module, a deep reinforcement learning framework integration module, and a dynamic process parameter optimization control module; The multi-source data acquisition and preprocessing module is used to generate different combinations of spinning process parameters, perform finite element simulation, and simultaneously acquire physical experimental data; The neural network model building and training module is used to train the neural network model and determine the mapping relationship between the physical field state and the form quality parameters during part forming; The deep reinforcement learning framework integration module is used to deeply integrate deep reinforcement learning with the finite element simulation environment, build a closed-loop control framework for autonomous optimization of spinning process parameters, and establish an intelligent decision-making cycle of state perception → strategy generation → action execution → reward feedback; The dynamic process parameter optimization control module is used to achieve dynamic adjustment of spinning process parameters through real-time interaction, and finally output the optimal process path.

[0016] The above scheme of the present invention has the following beneficial effects: The collaborative optimization method and system of the intelligent spinning process path provided by the present invention can realize the dynamic adjustment of the process parameters in the spinning process according to factors such as the ideal deformation temperature, strain rate and strain of the parts, and is suitable for the process parameter optimization of the spinning parts with complex curved generatrix and non-axisymmetric, so as to find the best process path for customizing and optimizing the comprehensive quality of the parts, and realize the characteristics of high accuracy and high reliability. The deep reinforcement learning framework adopted by the present invention is unsupervised learning, which does not require a large amount of predefined label data. The spin forming process parameter control model can obtain a large amount of data by interacting with the environment, thus reducing the cost of pre-data acquisition. The present invention adopts a neural network to determine the reward function in the spin forming process parameter control model, which can quickly establish a nonlinear mapping between factors such as deformation temperature and strain rate in the forming process and the shape quality index, and a wide variety of parameters can be optimized; The deep learning framework adopted by the present invention combines the advantages of value function and policy gradient, and adopts policy network and value network to deal with high-dimensional problems with continuous state and action space in spinning simulation, avoiding the limitation that the value function method can only deal with discrete action space problems, and has higher efficiency than pure policy gradient; Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the method steps of the present invention; Figure 2 It is a schematic diagram of the assembly of the geometric model in the spinning finite element model in the embodiment of the present invention; Figure 3 It is a schematic diagram of a control model for the spinning process parameters in an embodiment of the present invention.

[0018] [Description of Reference Numerals] 1-Blank; 2-Rotary wheel. DETAILED DESCRIPTION

[0019] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0020] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0021] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure, and the illustrations only show the components related to the present disclosure rather than the number, shape and size of the components in actual implementation. The type, quantity and proportion of each component in actual implementation may be changed at will, and the component layout type may also be more complicated. In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0022] like Figure 1 As shown, an embodiment of the present invention provides a collaborative optimization method for an intelligent spinning process path, comprising the following steps: S1, select different spinning process parameters according to orthogonal design, and then obtain the physical field state of the part during forming and the shape quality parameters of the part after forming through finite element simulation and spinning experiments.

[0023] The spinning process parameters include thinning rate, spindle speed, feed speed of spinning wheel 2, heating power, etc. The spinning process parameters are set to fixed values ​​during the forming process to reduce the simulation calculation amount and experimental operation difficulty. In this embodiment, the spinning process parameters within a suitable range are selected through orthogonal design, and then the spinning finite element simulation and spinning experiment are carried out under the process conditions selected by the orthogonal table to obtain the average physical field state during part forming and the form quality parameters of the part after forming.

[0024] In the spinning finite element simulation, the physical field state includes the average strain of the deformation unit, the average strain rate, the average forming temperature, etc. When calculating all the physical field states, the stage where the instantaneous strain rate is 0 should be removed. Specifically, the history curve of the temperature, strain, strain rate and other information of the unit in the generatrix direction of the part can be extracted, and then the data points with instantaneous strain rate greater than 0 and other physical field state variable values ​​corresponding to the unit at that moment can be screened out to calculate their respective average values.

[0025] The spinning finite element model can be constructed using existing commercial software such as ABAQUS, ANSYS, etc. For example, the construction of the spinning finite element model through ABAQUS includes: confirmation of the material properties of parts; establishment of the assembly relationship of parts; setting of geometric boundary conditions, including mold motion trajectory, contact relationship of each part, application of load conditions and thermal boundary conditions, etc.; ensuring uniform unit size during meshing, at least 3 or more units in the wall thickness direction of blank 1, using reduced integration units as unit type, and using hourglass enhancement mode; selecting the Explicit method for model solution to avoid convergence problems.

[0026] After the spinning experiment, the parts are tested for their shape quality. The quality parameters include the geometric shape parameters and mechanical properties of the parts. The geometric shape parameters include the roundness of the parts, the degree of mold adhesion and the theoretical wall thickness difference; the mechanical properties include the hardness of the parts, the tensile properties at room temperature and the impact toughness.

[0027] S2, establish a neural network model, determine the mapping relationship between the physical field state and the shape quality parameters during part forming, and determine the reward function through the weighted coefficient method to train the neural network model.

[0028] In this embodiment, the structural parameters in the neural network model include initial weights and thresholds, the number of hidden layers, the number of neurons in each layer, and the selection of activation functions, which are determined by an optimization algorithm. The optimization algorithm may adopt a Bayesian algorithm, a genetic algorithm, a particle swarm optimization algorithm, etc. The training process of the neural network model is specifically as follows: the physical field state of the part formed by finite element simulation is used as input, and the form quality parameters obtained by the spinning experiment are used as output. The error between the weighted sum of the quality parameters is obtained by calculation, and the weight coefficient of the neural network model is then updated by back propagation, and the training is performed until convergence. Among them, the physical field state and the form quality parameters are normalized according to The method is normalized to between 0 and 1 before input and output.

[0029] It should be noted that in this embodiment, K-fold cross validation is used during neural network model training to improve the reliability and stability of model evaluation. K-fold cross validation maximizes data utilization and reduces evaluation bias by repeatedly dividing the data set. Generally, the less the amount of data for training, the larger the K value. Specifically, the above data is randomly divided into K parts, and K-1 parts are selected as training sets each time, and the remaining 1 part is the test set. Cross-validate K times, and take the average value of K times accuracy as the final evaluation index of the neural network model. When the amount of data is less than 1000 groups, 5-fold or 10-fold cross validation is generally selected.

[0030] In this embodiment, the reward function Confirmed by weighted coefficient method, the specific expression is: ; in, is the weight coefficient, The size of is allocated according to the importance of each indicator, and , is the total number of indicators, is the sub-reward item corresponding to the i-th indicator, This is the output result of the mechanical performance in the neural network model. The larger the value, the greater the reward. The output result of the geometric shape parameter in the neural network model. The smaller the value, the greater the reward. Generally speaking, the mechanical properties and geometric shape accuracy of parts are equally important. Can be set to 1 / n.

[0031] S3, build a deep reinforcement learning framework, embed the deep reinforcement learning framework into the spinning finite element model through secondary development, and establish a spinning process parameter control model.

[0032] For example, the deep reinforcement learning framework can be written using the VUMAT subroutine in the ABAQUS software. In this embodiment, the deep reinforcement learning framework is modeled by a Markov decision process and is defined as a four-tuple: state s, action a, reward r, and state transition probability distribution function p. Among them, state s corresponds to the physical field state of the unit during the spinning finite element simulation. In this embodiment, it is specifically defined as the physical field state of the deformed unit that is undergoing plastic deformation during the spinning process, so as to avoid reading other undeformed units in the model, thereby significantly increasing the amount of calculation; action a is the adjustment amount of the spinning process parameters; reward r is the calculated value of the reward function defined in S2, which is used to evaluate the immediate impact of the current action on the shape quality; function p is the probability distribution of transferring to the next physical field state under a given state s and action a, that is, transferring to the next state under a given state s and action a. The probability distribution of .

[0033] It should be noted that the spin forming process parameter control model established in this embodiment is composed of an environment and an agent. The environment is composed of a spinning finite element model and a reward function; the agent is composed of a strategy network (actor) and a value network (critic).

[0034] In this embodiment, the policy network can select an action according to the current state s, and continuously optimize the policy through a policy gradient method (such as deterministic policy gradient, DPG) to maximize the cumulative reward. The specific expression is: ; Among them, the objective function Policy network parameters The gradient of Indicates that the state s is distributed The expectations below, Representation strategy The probability of the next state s appearing is, is the action value function, which means that in the strategy The expected cumulative return value after taking action a in state s is is the gradient of the action value function with respect to action a, is the output of the policy network, Represents the output pair parameters of the policy network The gradient of . Therefore, the policy network can select an action based on the current state s. Cumulative reward is the sum of the reward values ​​from the start to the end of the simulation, expressed as follows: ; in, is the discount rate, which is used to determine the priority of short-term returns, ranging from 0 to 1. For the moment, Represents the end moment, represents the time step, Represents the time step When, according to the current state Select and execute the current action After that, the immediate reward value of the environment feedback. When it approaches 0, the model pays more attention to short-term rewards, otherwise it treats all step rewards equally. In spinning, the performance of each part is equally important, so Should be close to 1.

[0035] In this embodiment, the value network can evaluate the quality of the current strategy, provide guidance for updating the strategy network, and improve learning stability and efficiency.

[0036] S4, the intelligent agent in the spinning process parameter control model continuously interacts with the environment and continuously adjusts the spinning process parameters according to the reward value, thereby maximizing the cumulative reward in the forming process and finally outputting the optimal spinning process path.

[0037] In this embodiment, the spin forming process parameter control model controls the interaction process through a preset control strategy: when the environment receives the action a of the agent (such as heating power, feed speed of the spinning wheel 2, etc.), the environment obtains a reward r and a state s′ at the next moment according to the current state s (temperature and strain of the deformation unit, etc.), and the agent optimizes the strategy according to the accumulated reward value and determines the next action, that is, starts the next interaction. During the continuous interaction process, the parameters θ of the strategy network and ω of the value network inside the agent are continuously updated, so that the actions of the agent are more and more in line with expectations, and finally the optimal spinning process path is learned.

[0038] The parameter update process of the strategy network and value network during agent interaction is as follows: At this moment, the agent executes Action at all times Afterwards, it was observed from the environment Moment Rewards and the new state ; Policy network makes decisions , but the agent does not immediately execute the decision ; The value network scores the actions: ; ; in, for The score of the decision in the state, for The score of the decision in the state, is the action value function; Based on this, the network target value is calculated (TD target) and forecast error value (TD error): ; ; Then the value network parameter ω is updated according to the gradient descent method: ; in, is the updated value of the value network parameter, is the gradient descent learning rate, usually around 10 -5 ~10 -2Range, its value is related to task complexity, network structure, etc. Representation parameters The gradient of . Then update the policy network parameters θ according to the gradient ascent method: ; in, The updated values ​​of the policy network parameters, is the gradient ascent learning rate, and its range is similar, Representation parameters The gradient of is the output of the policy network, representing the agent in state Directly select an action .

[0039] It should be noted that in this embodiment, when updating the network parameters inside the agent, experience replay is used to store the interaction records between the agent and the environment to reuse the collected data, overcome the correlation and non-stationary distribution problems of network data, and improve the efficiency and stability of network data training. Each time the network is updated, multiple groups of records (mini-batches) are randomly selected and the corresponding TD targets and TD error values ​​are calculated, and then the average gradient is calculated to update the network parameters. It should be noted that the mini-batch size is usually between 32 and 512, and 32 or 64 can be selected for smaller networks.

[0040] It should be noted that in this embodiment, when the agent interacts with the environment, noise can be added to improve the agent's exploration ability, avoid the strategy from falling into the local optimum, and improve learning efficiency. Among them, noise can be added in the policy network output action, value network state input, and experience playback stage. For example, before the agent executes the action selected by the strategy, a Gaussian distribution form can be added, such as ,in is the standard deviation, c is the noise range, Used to Restricted to the range .

[0041] The effect of this method is further illustrated by a specific case. A magnesium alloy fairing (thin-walled conical part) is made from an AZ31 magnesium alloy disc with a diameter of 200 mm and a thickness of 4 mm as a blank 1. It is required to optimize the laser heating power, the feed speed of the spinning wheel 2 and the spindle speed during shear spinning to achieve the highest possible room temperature tensile properties and impact toughness of the part while meeting the geometric dimensional accuracy (roundness tolerance of 0.01 mm, mold fit within 0.1 mm). The specific steps include the following: The forming parameter range of magnesium alloy conical parts is determined according to the size and material of spinning blank 1, that is, the heating power is 1000W, 1500W, 2000W, 2500W, the feed speed of spinning wheel 2 is 2mm / s, 4mm / s, 6mm / s, and the spindle speed is 3rps, 6rps, 9rps. The L16 (4^5) orthogonal table is used to determine 16 groups of orthogonal experiments with different factor levels.

[0042] Based on Abaqus finite element software, a laser-assisted shear spinning model of magnesium alloy was established. Figure 2 As shown. The boundary conditions of the model are set according to the orthogonal table. The laser heat source and the motion trajectory of the spinning wheel 2 in the model can be embedded through the VDEFLUX and VDISP subroutines; when dividing the mesh, ensure that there are 3 or more units in the thickness direction, and the unit type is C3D8RT enhanced with hourglass; the solver selects Explicit. After all simulation runs are completed, the temperature, strain, and strain rate history curves of all units in the generatrix direction of the cone are extracted, and the average values ​​of each variable corresponding to the instantaneous strain rate is not 0 are calculated. Based on the spinning experimental platform, a spinning orthogonal experiment is carried out to measure the geometric accuracy of the spun cone, and record the roundness and mold adhesion of the parts; then the parts are tested for room temperature tensile properties and impact toughness.

[0043] A BP neural network model was established, with the input layer being the unit average temperature, average strain, and average strain rate, and the output layer being roundness, mold adhesion, room temperature tensile properties, and impact toughness; the training data was normalized and then trained using 10-fold cross validation; the mean absolute error was selected to evaluate the model; the number of hidden layers and neurons in the model was optimized using the Bayesian algorithm to ensure that the model had the best predictive ability. The weighted coefficient method was used to determine the reward function of the spin forming process parameter control model: ; Among them, when the roundness and mold fit tolerances are less than 0.01mm and 0.1mm respectively, a=1, otherwise it is 0; and They are tensile properties and impact toughness respectively.

[0044] Based on the Abaqus secondary development platform, the actor-critic deep learning framework is embedded into the finite element model using the VUMAT subroutine to establish a spin forming process parameter control model, such as Figure 3 As shown in the figure, it is divided into two parts: environment and agent. The environment part consists of a shear spinning finite element model and a reward function, and the agent consists of a value network and a policy network. It belongs to the deep deterministic policy gradient algorithm (DDPG). The model uses the temperature, strain and strain rate of the deformation unit during spinning as the system state. sIn each control step, the laser heating power, spindle speed and feed speed of the rotary wheel 2 are used as system actions. a Due to the local forming characteristics of spinning, only the units near the contact point between the spinning wheel 2 and the blank 1 are selected each time the system state is read to reduce the state space of the system.

[0045] During the simulation, the agent in the spinning process parameter control model interacts with the environment continuously, that is, the environment receives the action , get reward based on the environment state at the previous moment System status , this part of information is in the form of a tuple Stored in the replay experience; then sample a batch of 32 experiences (min-batch) from the replay experience, and then use the policy network decision in the agent to get the new action under each set of experience , to enhance the model's exploration capabilities, add noise to the actions , the action noise of the heating power can be set = 0.5, c = 25; the value network in the agent is responsible for the previous action and the actions resulting from the decision Evaluated and , and the TD target value is calculated from this and TD error ; Update the parameters in the value network according to the error value , and update the parameters in the decision network , the learning rate during parameter update is set to 0.001, and after the network update is completed, the agent makes a decision output action , start the next interaction.

[0046] During the spinning process, the agent calculates the cumulative reward and , Taking 0.9, by selecting the process path with the maximum cumulative reward, the variation information of the spinning process parameters during the forming process, that is, the optimal spinning process parameters, is obtained.

[0047] To sum up, the collaborative optimization method of the intelligent spinning process path provided in this embodiment can realize the dynamic adjustment of the process parameters in the spinning process according to factors such as the ideal deformation temperature, strain rate and strain of the parts. It is suitable for the process parameter optimization of spinning parts with complex curved generatrix and non-axisymmetric spin parts, and finds the best process path for customizing and optimizing the comprehensive quality of parts, achieving the characteristics of high accuracy and reliability.

[0048] Based on the same inventive concept, this embodiment also provides a collaborative optimization system for an intelligent spinning process path, including a multi-source data acquisition and preprocessing module, a neural network model construction and training module, a deep reinforcement learning framework integration module, and a dynamic process parameter optimization control module. Among them, the multi-source data acquisition and preprocessing module is used to generate different combinations of spinning process parameters, perform finite element simulation, and obtain physical experimental data at the same time; the neural network model construction and training module is used to train the neural network model to determine the mapping relationship between the physical field state and the shape quality parameters during part forming; the deep reinforcement learning framework integration module is used to deeply integrate deep reinforcement learning with the finite element simulation environment, construct a closed-loop control framework for autonomous optimization of spinning process parameters, and establish an intelligent decision-making cycle of state perception → strategy generation → action execution → reward feedback; the dynamic process parameter optimization control module is used to realize the dynamic adjustment of spinning process parameters through real-time interaction, and finally output the optimal process path.

[0049] The collaborative optimization system for the intelligent spinning process path provided in this embodiment has the same inventive concept and beneficial effects as the aforementioned method, which will not be repeated here.

[0050] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0051] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A collaborative optimization method for an intelligent spinning process path, characterized in that: The steps include: S1, select different spinning process parameters according to orthogonal design, and then obtain the physical field state of the part during forming and the shape quality parameters of the part after forming through finite element simulation and spinning experiment; S2, establish a neural network model, determine the mapping relationship between the physical field state and the shape quality parameters during part forming, and determine the reward function through the weighted coefficient method to train the neural network model; S3, build a deep reinforcement learning framework, embed the deep reinforcement learning framework into the spinning finite element model through secondary development, and establish a spinning forming process parameter control model; S4, the intelligent agent in the spinning process parameter control model continuously interacts with the environment and continuously adjusts the spinning process parameters according to the reward value to maximize the cumulative reward in the forming process and finally output the optimal spinning process path.

2. The collaborative optimization method of an intelligent spinning process path according to claim 1, characterized in that: The spinning process parameters include thinning rate, spindle speed, spinning wheel feed speed, and heating power. The spinning process parameters are set to fixed values ​​during the forming process; The physical field state includes the average strain of the deformation unit, the average strain rate, and the average forming temperature. When calculating all the physical field states, the stage with the instantaneous strain rate is removed; The morphological quality parameters include part geometry parameters and mechanical properties; the geometry parameters include part roundness, mold fit and theoretical wall thickness difference; the mechanical properties include part hardness, room temperature tensile properties and impact toughness.

3. The collaborative optimization method of an intelligent spinning process path according to claim 1, characterized in that: In S2, the physical field state of the part formed by the finite element simulation is used as input, and the shape quality parameters obtained by the spinning experiment are used as output. The error between the weighted sum of the shape quality parameters is obtained by calculation, and then the weight coefficient of the neural network model is updated by back propagation and trained until convergence.

4. The collaborative optimization method of an intelligent spinning process path according to claim 1, characterized in that: K-fold cross validation is used when training the neural network model. K-fold cross validation randomly divides the data into K parts, selects K-1 parts as the training set each time, and the remaining 1 part is the test set. Cross validation is performed K times, and the average of the K times of accuracy is taken as the final evaluation indicator of the neural network model.

5. The collaborative optimization method of an intelligent spinning process path according to claim 1, characterized in that: Reward function in S2 Confirmed by the weighted coefficient method, the expression is: ； in, is the weight coefficient, The size of is allocated according to the importance of each indicator, and , is the total number of indicators, is the sub-reward item corresponding to the i-th indicator, Output results of mechanical properties in the neural network model. The bigger the reward, the greater the Output results for the geometric shape parameters in the neural network model, The smaller the reward, the greater the degree.

6. The collaborative optimization method of an intelligent spinning process path according to claim 1, characterized in that: The deep reinforcement learning framework in S3 is modeled by a Markov decision process and is defined as a four-tuple: state s, action a, reward r, and state transition probability distribution function p; state s corresponds to the physical field state of the unit during spinning finite element simulation; action a is the adjustment amount of the spinning process parameters; reward r is the calculated value of the reward function defined in S2, which is used to evaluate the immediate impact of the current action on the shape quality; function p is the probability distribution of transferring to the next physical field state under a given state s and action a.

7. The collaborative optimization method of an intelligent spinning process path according to claim 6, characterized in that: The spin forming process parameter control model includes environment and agent; The environment includes a spinning finite element model and a reward function; the agent includes a policy network and a value network; The policy network selects actions based on the current state s and continuously optimizes the strategy through the policy gradient method to maximize the cumulative reward. is the sum of the reward values ​​from the start to the end of the simulation, expressed as: ； in, is the discount rate, which is used to determine the priority of short-term returns, ranging from 0 to 1. represents the time step, For the moment, Represents the end moment, Represents the time step When, according to the current state Select and execute the current action After that, the immediate reward value of the environment feedback; The value network is used to evaluate the quality of the current strategy and provide guidance for updating the strategy network.

8. The collaborative optimization method of an intelligent spinning process path according to claim 7, characterized in that: The spinning process parameter control model in S4 controls the interaction process through a preset control strategy: when the environment receives the action a of the agent, the environment obtains a reward r and a state s′ at the next moment according to the state s at the current moment. The agent optimizes the strategy according to the accumulated reward value and determines the next action. During the continuous interaction process, the parameters θ of the strategy network and ω of the value network inside the agent are continuously updated, so that the actions of the agent are more and more in line with expectations, and finally the optimal spinning process path is learned.

9. The collaborative optimization method of an intelligent spinning process path according to claim 8, characterized in that: The parameter update process of the strategy network and value network when the agents interact is as follows: At this moment, the agent executes Action at all times Afterwards, it was observed from the environment Moment Rewards and the new state ; Policy network makes decisions , Represents a strategy, but the agent does not immediately execute the decision The value network scores actions : ； ； in, for The score of the decision in the state, for The score of the decision in the state, is the action value function; Based on this, the network target value is calculated and prediction error values : ； ； Update the value network parameter ω according to the gradient descent method: ； in, is the updated value of the value network parameter, is the gradient descent learning rate, Representation parameters The gradient of Then update the policy network parameters θ according to the gradient ascent method: ； in, is the updated value of the policy network parameters, is the gradient ascent learning rate, Representation parameters The gradient of Represents the output of the policy network, representing the agent in state Directly select an action .

10. A collaborative optimization system for an intelligent spinning process path, using a collaborative optimization method for an intelligent spinning process path as claimed in any one of claims 1 to 9, characterized in that: It includes multi-source data acquisition and preprocessing module, neural network model building and training module, deep reinforcement learning framework integration module, and dynamic process parameter optimization control module; The multi-source data acquisition and preprocessing module is used to generate different combinations of spinning process parameters, perform finite element simulation, and simultaneously acquire physical experimental data; The neural network model building and training module is used to train the neural network model and determine the mapping relationship between the physical field state and the form quality parameters during part forming; The deep reinforcement learning framework integration module is used to deeply integrate deep reinforcement learning with the finite element simulation environment, build a closed-loop control framework for autonomous optimization of spinning process parameters, and establish an intelligent decision-making cycle of state perception → strategy generation → action execution → reward feedback; The dynamic process parameter optimization control module is used to achieve dynamic adjustment of the spinning process parameters through real-time interaction, and finally output the optimal process path.

Images