Micro spring washer production parameter self-adaptive adjustment method based on reinforcement learning

Abstract

The invention discloses a reinforcement learning-based micro spring washer production parameter adaptive adjustment method, which comprises the following steps of S1, collecting and preprocessing multi-source data in a micro spring washer production process to form a multi-dimensional data set; s2, constructing a virtual simulation environment, and generating a preliminary adjustment strategy by using a reinforcement learning algorithm; s3, optimizing production parameters through a layered progressive self-adaptive reinforcement learning model; s4, monitoring and dynamically adjusting production parameters in real time by utilizing a sensing technology of an intelligent material; s5, monitoring and adjusting the topological structure of the system through a self-repairing topological network mechanism; s6, a self-evolution bifurcation control mechanism is applied, and key parameters are monitored and adjusted in real time; s7, predicting key parameters and product quality in real time by using a quantum cascade model prediction technology, and performing optimization adjustment; and S8, applying the optimized production parameters to actual production, and performing continuous verification and adjustment through a real-time feedback mechanism. According to the invention, self-adaptive adjustment and stable operation in the production process are realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning. Background Art

[0002] With the continuous development of intelligent manufacturing technology, the manufacturing industry has placed higher demands on precise control and real-time adjustment of various process parameters during production. Micro spring washers are high-precision components, and parameter control during their production directly affects the quality and performance of the final product. However, existing production parameter adjustment methods still have many shortcomings and challenges when dealing with complex and changing production environments.

[0003] First, traditional production parameter adjustment methods typically rely on pre-set rules or empirical formulas, which lack dynamic adaptability. In actual production, the production process parameters of micro spring washers are often affected by a variety of factors, such as temperature, pressure, equipment status, and raw material properties. These changes are difficult to fully control and adjust using fixed rules. Therefore, existing technologies often fail to achieve precise control of production parameters in complex production environments, leading to unstable product quality.

[0004] Second, existing production parameter adjustment methods typically rely on offline data analysis, collecting and analyzing historical data from the production process to generate adjustment strategies, which are then applied to the production process. This approach suffers from data processing delays, making it unable to respond to changes in the production environment in real time, resulting in a lag in production parameter adjustment. Furthermore, offline data analysis methods often lack effective data processing capabilities when dealing with real-time data streams, making it impossible to accurately extract useful information from large amounts of real-time data, resulting in insufficient precision in production parameter adjustment.

[0005] Furthermore, existing production parameter adjustment methods lack a unified processing and fusion mechanism for multi-source, heterogeneous data. The production process of micro spring washers involves a variety of data types, including equipment operating data, environmental data, and process parameter data. These data come from diverse sources and formats, making it difficult for existing technologies to effectively integrate these multiple sources. This results in an inability to fully utilize all information when generating adjustment strategies, limiting the effectiveness and accuracy of the adjustment results.

[0006] Furthermore, traditional production parameter adjustment methods often lack the ability to dynamically monitor and adjust the overall system structure. The topology of the micro-spring gasket production system can change during production, due to factors such as equipment failures and disconnections. Existing technologies often struggle to detect and fix these issues in a timely manner, impacting the stability and reliability of the entire production system.

[0007] Finally, existing technologies often use linearized or simplified models for analysis and adjustment when dealing with complex nonlinear systems. This approach has significant limitations when dealing with complex systems in practice. In the production of micro-spring washers, many process parameters exhibit nonlinear coupling relationships. Existing linearized adjustment methods are unable to accurately describe and control these complex relationships, resulting in poor adjustment results and potentially even causing system instability.

[0008] Therefore, how to provide a method for adaptively adjusting the production parameters of micro spring washers based on reinforcement learning is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0009] One purpose of the present invention is to propose a method for adaptively adjusting the production parameters of micro spring washers based on reinforcement learning. The present invention makes full use of reinforcement learning algorithms, intelligent material sensing technology and quantum cascade model prediction technology, and describes in detail the steps of realizing real-time parameter optimization and adaptive adjustment in a complex production environment. It has the advantages of strong dynamic adaptability, high production quality stability and excellent real-time response capability.

[0010] A method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning according to an embodiment of the present invention includes the following steps:

[0011] S1. Collect multi-source heterogeneous data in the production process of micro spring washers, form a multidimensional dataset, and preprocess it;

[0012] S2. Based on the preprocessed multidimensional dataset, a virtual simulation environment is constructed. The production process parameters, equipment operating status, environmental variables, and product quality indicators are used as the state space, and the production parameter adjustment is used as the action space. The reinforcement learning algorithm is used to optimize the production parameters and generate a preliminary adjustment strategy.

[0013] S3. Initialize a hierarchical progressive adaptive reinforcement learning model. Optimize basic production parameters through the primary-level model. The secondary-level model integrates and optimizes the primary-level results. The senior-level model generates a multi-objective optimization strategy that comprehensively considers production efficiency, quality stability, and energy consumption. The effectiveness of the strategy is verified through a simulation environment.

[0014] S4. During the actual production process, the multifunctional integrated sensing technology of smart materials is used to monitor key process parameters and environmental changes in real time, and production parameters are dynamically optimized based on feedback from a hierarchical progressive reinforcement learning model.

[0015] S5. Through the self-repairing topology network mechanism, the topology structure of the production system is monitored in real time, abnormal nodes and connections are identified, and the network structure is dynamically adjusted to further optimize production parameters;

[0016] S6. Apply the self-evolutionary bifurcation control mechanism to monitor the changes of key parameters in the production system in real time, identify parameters close to bifurcation points, and adjust the evolution path of parameters to ensure stable operation of the system;

[0017] S7. Use quantum cascade model prediction technology to predict key parameters and product quality in the production process in real time, adjust production parameters, and optimize product quality;

[0018] S8. Apply the optimized production parameters to actual production, continuously verify and adjust the model through a multi-source real-time feedback mechanism, and optimize the continuous adaptive adjustability and operational stability of the production process.

[0019] Optionally, S3 includes the following steps:

[0020] S31. Initialize a hierarchical progressive adaptive reinforcement learning model, combine it with a multi-dimensional topological braiding network mechanism, and model and optimize the multi-dimensional parameters involved in the production process;

[0021] S32. Input the preprocessed multidimensional dataset into the primary hierarchical model, construct a basic parameter braiding structure using a multidimensional topological braiding network mechanism, and perform preliminary optimization of the nodes and edges in the braiding network using a reinforcement learning algorithm to generate an optimized primary parameter combination;

[0022] S33, input the optimization parameter combination generated by the primary level model into the intermediate level model, and the intermediate level model uses the dynamic adjustment function of the multi-dimensional topological weaving network to adjust the node weights w in the topological structure. i and edge weight w ij To optimize:

[0023]

[0024] in, represents the edge weight between node i and node j at the tth iteration, represents the difference between the states of node i and node j at the tth iteration, represents the square of the Euclidean distance between the feature vectors corresponding to node i and node j, α and β are hyperparameters that control the adjustment amplitude and decay speed, and γ is the neighborhood influence factor. and Represents the association matrix value between node i and node j and neighboring node k, Represents the association between node i and node j and their common neighboring nodes and their weight influence;

[0025] S34. Input the optimization parameter combination generated by the intermediate-level model into the advanced-level model. The advanced-level model, in combination with the global production goal, utilizes the global optimization capability of the multi-dimensional topological weaving network to comprehensively adjust the entire network structure. The topological relationship between nodes is adjusted through the reinforcement learning algorithm. Based on the global optimization strategy of the weaving operation, a multi-objective optimization strategy is generated that comprehensively considers production efficiency, quality stability, and energy consumption.

[0026] S35. The multi-objective optimization strategy generated by the advanced hierarchical model is verified through a simulation environment. The simulation environment uses the dynamic feedback function of the multi-dimensional topology weaving network mechanism to evaluate and adjust the optimization strategy in real time, and finally outputs the optimization parameter combination used in actual production. The results are fed back to the multi-dimensional topology weaving network to further improve its topological structure.

[0027] Optionally, the S4 includes the following steps:

[0028] S41. In actual production processes, the multifunctional integrated sensing technology of smart materials is used to monitor key process parameters, equipment operating status, and environmental changes in the production system in real time. Embedded sensors collect physical signals such as temperature, pressure, strain, and vibration to generate high-frequency, multi-dimensional real-time data streams.

[0029] S42, inputting the collected real-time data stream into the spatiotemporal hypersurface mapping algorithm module, mapping the multidimensional data onto a dynamic spatiotemporal hypersurface, and generating a hypersurface feature vector for a reinforcement learning model by capturing the spatiotemporal dependencies in the data;

[0030] S43. Based on the multi-level dynamic graph energy optimization method, the dynamic adjustment of production parameters is optimized at multiple levels. At the primary level, the energy function of the system is defined as:

[0031]

[0032] Among them, E (t) represents the total energy of the system at the tth iteration, are the optimized node and edge weights, is the square of the Euclidean distance between the hypersurface feature vectors of node i and node j, is the state difference, α is the energy adjustment factor, β is the weight factor of the difference between the optimization target and the current state, Represents the difference between the current eigenvector and the optimized target eigenvector;

[0033] S44. At a high level, the system energy function E (t) Perform multi-level optimization and dynamically adjust node and edge weights:

[0034]

[0035] Among them, λ is the step size parameter, Represents the energy function E (t) About edge weights The partial derivative of is the external interference factor between nodes, indicating the adjustment effect of environmental changes or external interference on edge weights. γ is the adjustment coefficient, which is used to control the impact of external interference factors on system weights.

[0036] S45. Dynamically adjust production parameters based on the feedback signal generated by the reinforcement learning model, and verify the adjustment results through the response mechanism of the smart material to form a closed-loop control.

[0037] Optionally, the S5 includes the following steps:

[0038] S51. Use the self-repairing topology network mechanism to monitor key parameters and topology status in the production system in real time, use embedded sensors to collect status information and connection status of each node in the production process, and generate real-time network topology data;

[0039] S52: Input the collected network topology data into the topology entropy optimization module, and calculate the improved entropy value H of the network. (t) To evaluate the chaos and order of the network:

[0040]

[0041] Among them, H (t) represents the improved entropy value of the network at the tth iteration, is the state probability of node i at the tth iteration, are the optimized node and edge weights, is the square of the Euclidean distance between the hypersurface feature vectors of node i and node j, is the state difference, is the degree of influence of the optimization result on the current iteration, α and β are the adjustment factors to control each influence;

[0042] S53, analysis entropy value H (t) The system detects changes in the network, identifies the most unstable or abnormal areas, and determines the nodes and connections that need to be repaired first. After identifying the high entropy areas, it further calculates the state stability and connection strength of each node in these areas to accurately locate the parts that need to be repaired.

[0043] S54. Topology repair is performed on nodes and connections in the high entropy area. By dynamically adjusting the node state probability and connection weight, the network structure is optimized and the overall entropy of the system is reduced. The optimization formula is:

[0044]

[0045] in, represents the connection weight after repair at the t+1th iteration, λ is the step size parameter, which controls the convergence speed of weight adjustment. To improve the entropy value H (t) About edge weights The partial derivative of is the external interference factor, is the historical optimization result repair factor, and γ is the adjustment coefficient;

[0046] S55. During the repair process, the entropy value changes of the repair area are continuously monitored, and the network structure is stabilized and ordered by dynamically adjusting the node state probability and connection weight;

[0047] S56. After the repair is completed, the topology entropy optimization module feeds back the repaired network parameters to the control module of the self-repairing topology network mechanism, and optimizes the global entropy value of the repaired topology structure to further optimize the robustness and adaptability of the network.

[0048] Optionally, the S6 includes the following steps:

[0049] S61. In the actual production process, the self-evolving bifurcation control mechanism is applied to monitor the changes of key parameters in the production system in real time. Embedded sensors are used to collect key process parameters such as temperature, pressure, and vibration, and generate high-frequency real-time data streams.

[0050] S62, input the collected real-time data stream into the bifurcation point identification module, combine the multi-scale bifurcation tracking technology, synchronously monitor the system state changes at different time and space scales, and calculate the state change rate of each node and the mutual influence coefficient β between adjacent nodes ij , identify the key nodes in the system close to the bifurcation point:

[0051]

[0052] in, represents the state variable of node i at time t, represents the state variable of node j at time t, β ij is the mutual influence coefficient between node i and node j, γ i is the self-regulation coefficient of node i, Represents the nonlinear self-evolution function of node i, and the input control variable Related;

[0053] S63: For the key nodes identified as being close to bifurcation points, path analysis and control are gradually performed based on the multi-scale bifurcation tracking technology. By adjusting the evolution path of the key nodes, the system is stabilized.

[0054]

[0055] in, represents the control input of node i in the next time step, λ is the step coefficient, and controls the adjustment amplitude. is the potential energy function of node i, κ is the adjustment coefficient, which is used to balance the influence between adjacent nodes;

[0056] S64. Control input generated by the self-evolving bifurcation control mechanism Adjust key parameters in the production system in real time, input the adjusted parameters into the system's feedback module to verify the adjustment effect and further optimize the control strategy;

[0057] S65. Feedback the adjusted system state to the reinforcement learning model. By comparing the actual system behavior with the expected goals, the adjustment strategy of production parameters is continuously optimized to ensure the long-term stability and efficient operation of the production system.

[0058] Optionally, the S7 includes the following steps:

[0059] S71. In the actual production process, the quantum cascade model prediction technology is used in combination with the vortex quantum field effect to make real-time predictions of key process parameters and product quality in the production system;

[0060] S72. Generate a prediction model for optimizing production parameters through the superposition of quantum states and dynamic adjustment of the vortex field;

[0061] S73. In the prediction process, construct an innovative quantum prediction function based on multidimensional interactions

[0062] Among them, Q i (t) represents the predicted value of the i-th key parameter at time t, α k is the weight coefficient in the quantum cascade model, Based on the current state variables and control inputs The quantum state function of is the influence of vortex quantum field effect on the i-th key parameter, is the connection weight of the key node after repair, Based on the status of adjacent nodes and optimized edge weights The quantum state function of , λ and γ are adjustment coefficients, which control the influence of vortex field effect and interaction between adjacent nodes respectively;

[0063] S74. Based on the above prediction results, dynamically adjust the control input of the production system

[0064]

[0065] in, represents the control input to the i-th node in the next time step, μ is the step size coefficient, which controls the adjustment amplitude, ν is the adjustment coefficient, which controls the influence of the vortex quantum field effect, and κ is the influence coefficient of the adjacent node state on the control input of the current node;

[0066] S75. Apply the updated control input to the production system and verify the accuracy of the prediction results through a feedback mechanism. If the prediction error exceeds a predetermined threshold, dynamically adjust the parameters of the quantum cascade model based on the feedback data until the system reaches stability.

[0067] The beneficial effects of the present invention are:

[0068] (1) The present invention achieves real-time parameter optimization and adaptive adjustment during the production of micro-spring washers by combining a reinforcement learning algorithm, a hierarchical progressive adaptive reinforcement learning model, intelligent material sensing technology, and quantum cascade model prediction technology. Particularly in complex production environments, the present invention overcomes the limitations of traditional production control methods, enabling dynamic monitoring and adjustment of production parameters, significantly improving the stability and adaptability of production quality.

[0069] (2) By introducing a self-repairing topological network mechanism and a self-evolving bifurcation control mechanism, the present invention optimizes the topological structure of the production system and the stability of key parameters, enhances the system's real-time response to abnormal situations, and effectively avoids potential failures and quality deviations in the production process. Combined with a multi-dimensional dynamic feedback mechanism, the present invention significantly improves the overall robustness and operational efficiency of the production system.

[0070] (3) The present invention provides an efficient, intelligent and adaptive micro spring gasket production parameter adjustment solution by comprehensively utilizing a variety of cutting-edge technologies, enabling the system to automatically optimize the production process in a constantly changing production environment, ensuring the consistency of production quality and efficient operation, thereby achieving comprehensive control and optimization of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0072] Figure 1 This is a flow chart of a method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning proposed by the present invention;

[0073] Figure 2 This is a schematic diagram of the structure of the hierarchical progressive adaptive reinforcement learning model proposed in the present invention;

[0074] Figure 3 This is a flowchart for the prediction and optimization of key parameters of the quantum cascade model prediction technology combined with the vortex quantum field effect in the present invention. DETAILED DESCRIPTION

[0075] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0076] refer to Figure 1-3 A method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning comprises the following steps:

[0077] S1. Collect multi-source heterogeneous data in the production process of micro spring washers, form a multidimensional dataset, and preprocess it;

[0078] S2. Based on the preprocessed multidimensional dataset, a virtual simulation environment is constructed. The production process parameters, equipment operating status, environmental variables, and product quality indicators are used as the state space, and the production parameter adjustment is used as the action space. The reinforcement learning algorithm is used to optimize the production parameters and generate a preliminary adjustment strategy.

[0079] S3. Initialize a hierarchical progressive adaptive reinforcement learning model. Optimize basic production parameters through the primary-level model. The secondary-level model integrates and optimizes the primary-level results. The senior-level model generates a multi-objective optimization strategy that comprehensively considers production efficiency, quality stability, and energy consumption. The effectiveness of the strategy is verified through a simulation environment.

[0080] S4. During the actual production process, the multifunctional integrated sensing technology of smart materials is used to monitor key process parameters and environmental changes in real time, and production parameters are dynamically optimized based on feedback from a hierarchical progressive reinforcement learning model.

[0081] S5. Through the self-repairing topology network mechanism, the topology structure of the production system is monitored in real time, abnormal nodes and connections are identified, and the network structure is dynamically adjusted to further optimize production parameters;

[0082] S6. Apply the self-evolutionary bifurcation control mechanism to monitor the changes of key parameters in the production system in real time, identify parameters close to bifurcation points, and adjust the evolution path of parameters to ensure stable operation of the system;

[0083] S7. Use quantum cascade model prediction technology to predict key parameters and product quality in the production process in real time, adjust production parameters, and optimize product quality;

[0084] S8. Apply the optimized production parameters to actual production, continuously verify and adjust the model through a multi-source real-time feedback mechanism, and optimize the continuous adaptive adjustability and operational stability of the production process.

[0085] In this embodiment, S3 includes the following steps:

[0086] S31. Initialize a hierarchical progressive adaptive reinforcement learning model, combine it with a multi-dimensional topological braiding network mechanism, and model and optimize the multi-dimensional parameters involved in the production process;

[0087] S32. Input the preprocessed multidimensional dataset into the primary hierarchical model, construct a basic parameter braiding structure using a multidimensional topological braiding network mechanism, and perform preliminary optimization of the nodes and edges in the braiding network using a reinforcement learning algorithm to generate an optimized primary parameter combination;

[0088] S33, input the optimization parameter combination generated by the primary level model into the intermediate level model, and the intermediate level model uses the dynamic adjustment function of the multi-dimensional topological weaving network to adjust the node weights w in the topological structure. i and edge weight w ij To optimize:

[0089]

[0090] in, represents the edge weight between node i and node j at the tth iteration, represents the difference between the states of node i and node j at the tth iteration, represents the square of the Euclidean distance between the feature vectors corresponding to node i and node j, α and β are hyperparameters that control the adjustment amplitude and decay speed, and γ is the neighborhood influence factor. and Represents the association matrix value between node i and node j and neighboring node k, Represents the association between node i and node j and their common neighboring nodes and their weight influence;

[0091] S34. Input the optimization parameter combination generated by the intermediate-level model into the advanced-level model. The advanced-level model, in combination with the global production goal, utilizes the global optimization capability of the multi-dimensional topological weaving network to comprehensively adjust the entire network structure. The topological relationship between nodes is adjusted through the reinforcement learning algorithm. Based on the global optimization strategy of the weaving operation, a multi-objective optimization strategy is generated that comprehensively considers production efficiency, quality stability, and energy consumption.

[0092] S35. The multi-objective optimization strategy generated by the advanced hierarchical model is verified through a simulation environment. The simulation environment uses the dynamic feedback function of the multi-dimensional topology weaving network mechanism to evaluate and adjust the optimization strategy in real time, and finally outputs the optimization parameter combination used in actual production. The results are fed back to the multi-dimensional topology weaving network to further improve its topological structure.

[0093] In this embodiment, S4 includes the following steps:

[0094] S41. In actual production processes, the multifunctional integrated sensing technology of smart materials is used to monitor key process parameters, equipment operating status, and environmental changes in the production system in real time. Embedded sensors collect physical signals such as temperature, pressure, strain, and vibration to generate high-frequency, multi-dimensional real-time data streams.

[0095] S42, inputting the collected real-time data stream into the spatiotemporal hypersurface mapping algorithm module, mapping the multidimensional data onto a dynamic spatiotemporal hypersurface, and generating a hypersurface feature vector for a reinforcement learning model by capturing the spatiotemporal dependencies in the data;

[0096] S43. Based on the multi-level dynamic graph energy optimization method, the dynamic adjustment of production parameters is optimized at multiple levels. At the primary level, the energy function of the system is defined as:

[0097]

[0098] Among them, E (t) represents the total energy of the system at the tth iteration, are the optimized node and edge weights, is the square of the Euclidean distance between the hypersurface feature vectors of node i and node j, is the state difference, α is the energy adjustment factor, β is the weight factor of the difference between the optimization target and the current state, Represents the difference between the current eigenvector and the optimized target eigenvector;

[0099] S44. At a high level, the system energy function E (t) Perform multi-level optimization and dynamically adjust node and edge weights:

[0100]

[0101] Among them, λ is the step size parameter, Represents the energy function E (t) About edge weights The partial derivative of is the external interference factor between nodes, indicating the adjustment effect of environmental changes or external interference on edge weights. γ is the adjustment coefficient, which is used to control the impact of external interference factors on system weights.

[0102] S45. Dynamically adjust production parameters based on the feedback signal generated by the reinforcement learning model, and verify the adjustment results through the response mechanism of the smart material to form a closed-loop control.

[0103] In this embodiment, S5 includes the following steps:

[0104] S51. Use the self-repairing topology network mechanism to monitor key parameters and topology status in the production system in real time, use embedded sensors to collect status information and connection status of each node in the production process, and generate real-time network topology data;

[0105] S52: Input the collected network topology data into the topology entropy optimization module, and calculate the improved entropy value H of the network. (t) To evaluate the chaos and order of the network:

[0106]

[0107] Among them, H (t) represents the improved entropy value of the network at the tth iteration, is the state probability of node i at the tth iteration, are the optimized node and edge weights, is the square of the Euclidean distance between the hypersurface feature vectors of node i and node j, is the state difference, is the degree of influence of the optimization result on the current iteration, α and β are the adjustment factors to control each influence;

[0108] S53, analysis entropy value H (t) The system detects changes in the network, identifies the most unstable or abnormal areas, and determines the nodes and connections that need to be repaired first. After identifying the high entropy areas, it further calculates the state stability and connection strength of each node in these areas to accurately locate the parts that need to be repaired.

[0109] S54. Topology repair is performed on nodes and connections in the high entropy area. By dynamically adjusting the node state probability and connection weight, the network structure is optimized and the overall entropy of the system is reduced. The optimization formula is:

[0110]

[0111] in, represents the connection weight after repair at the t+1th iteration, λ is the step size parameter, which controls the convergence speed of weight adjustment. To improve the entropy value H(t) About edge weights The partial derivative of is the external interference factor, is the historical optimization result repair factor, and γ is the adjustment coefficient;

[0112] S55. During the repair process, the entropy value changes of the repair area are continuously monitored, and the network structure is stabilized and ordered by dynamically adjusting the node state probability and connection weight;

[0113] S56. After the repair is completed, the topology entropy optimization module feeds back the repaired network parameters to the control module of the self-repairing topology network mechanism, and optimizes the global entropy value of the repaired topology structure to further optimize the robustness and adaptability of the network.

[0114] In this embodiment, S6 includes the following steps:

[0115] S61. In the actual production process, the self-evolving bifurcation control mechanism is applied to monitor the changes of key parameters in the production system in real time. Embedded sensors are used to collect key process parameters such as temperature, pressure, and vibration, and generate high-frequency real-time data streams.

[0116] S62, input the collected real-time data stream into the bifurcation point identification module, combine the multi-scale bifurcation tracking technology, synchronously monitor the system state changes at different time and space scales, and calculate the state change rate of each node and the mutual influence coefficient β between adjacent nodes ij , identify the key nodes in the system close to the bifurcation point:

[0117]

[0118] in, represents the state variable of node i at time t, represents the state variable of node j at time t, β ij is the mutual influence coefficient between node i and node j, γ i is the self-regulation coefficient of node i, Represents the nonlinear self-evolution function of node i, and the input control variable Related;

[0119] S63: For the key nodes identified as being close to bifurcation points, path analysis and control are gradually performed based on the multi-scale bifurcation tracking technology. By adjusting the evolution path of the key nodes, the system is stabilized.

[0120]

[0121] in, represents the control input of node i in the next time step, λ is the step coefficient, and controls the adjustment amplitude. is the potential energy function of node i, κ is the adjustment coefficient, which is used to balance the influence between adjacent nodes;

[0122] S64. Control input generated by the self-evolving bifurcation control mechanism Adjust key parameters in the production system in real time, input the adjusted parameters into the system's feedback module to verify the adjustment effect and further optimize the control strategy;

[0123] S65. Feedback the adjusted system state to the reinforcement learning model. By comparing the actual system behavior with the expected goals, the adjustment strategy of production parameters is continuously optimized to ensure the long-term stability and efficient operation of the production system.

[0124] In this embodiment, S7 includes the following steps:

[0125] S71. In the actual production process, the quantum cascade model prediction technology is used in combination with the vortex quantum field effect to make real-time predictions of key process parameters and product quality in the production system;

[0126] S72. Generate a prediction model for optimizing production parameters through the superposition of quantum states and dynamic adjustment of the vortex field;

[0127] S73. In the prediction process, construct an innovative quantum prediction function based on multidimensional interactions

[0128] Among them, Q i (t) represents the predicted value of the i-th key parameter at time t, α k is the weight coefficient in the quantum cascade model, Based on the current state variables and control inputs The quantum state function of is the influence of vortex quantum field effect on the i-th key parameter, is the connection weight of the key node after repair, Based on the status of adjacent nodes and optimized edge weights The quantum state function of , λ and γ are adjustment coefficients, which control the influence of vortex field effect and interaction between adjacent nodes respectively;

[0129] S74. Based on the above prediction results, dynamically adjust the control input of the production system

[0130]

[0131] in, represents the control input to the i-th node in the next time step, μ is the step size coefficient, which controls the adjustment amplitude, ν is the adjustment coefficient, which controls the influence of the vortex quantum field effect, and κ is the influence coefficient of the adjacent node state on the control input of the current node;

[0132] S75. Apply the updated control input to the production system and verify the accuracy of the prediction results through a feedback mechanism. If the prediction error exceeds a predetermined threshold, dynamically adjust the parameters of the quantum cascade model based on the feedback data until the system reaches stability.

[0133] Example 1:

[0134] In order to verify the feasibility of the present invention in implementation, the present invention was applied to a precision manufacturing plant located in Suzhou City, Jiangsu Province, which focuses on the production of micro spring washers. The micro spring washers of this factory are mainly used for high-precision mechanical equipment and medical devices, which have extremely high requirements for product accuracy and consistency. However, traditional production processes have obvious shortcomings in responding to these challenges, mainly manifested in frequent fluctuations in production parameters, unstable product quality, frequent manual intervention, and low production efficiency. For this reason, the factory decided to introduce a micro spring washer production parameter adaptive adjustment method based on reinforcement learning of the present invention to optimize its production process.

[0135] In this example, we first used advanced sensor technology to collect multi-source heterogeneous data in real time during the production process of micro spring washers. This data includes equipment operating status, production process parameters, environmental variables, and product quality indicators. The collected data is transmitted to the central control system via the Industrial Internet and preprocessed to form a multidimensional dataset. Based on this dataset, we constructed a virtual simulation environment, using production process parameters, equipment operating status, environmental variables, and product quality indicators as the state space and production parameter adjustments as the action space. We then used a reinforcement learning algorithm to generate a preliminary adjustment strategy. Using this virtual simulation environment, we evaluated and selected the optimal preliminary adjustment strategy to improve the stability of the production process.

[0136] Next, we further optimized production parameters using a hierarchical, progressive, and adaptive reinforcement learning model. This model optimizes basic production parameters at the primary level, integrates and optimizes the primary-level results at the secondary level, and generates a multi-objective optimization strategy at the advanced level that comprehensively considers production efficiency, quality stability, and energy consumption. The model's effectiveness was verified in a simulation environment.

[0137] During actual production, the factory leverages the multifunctional integrated sensing technology of smart materials to monitor key process parameters and environmental changes in real time. Through the smart material's embedded sensors, the system can keenly sense subtle changes in temperature, pressure, and vibration, and feed these changes into a hierarchical, progressive reinforcement learning model. Based on this real-time feedback, the model dynamically adjusts production parameters to ensure continuity and stability of the production process.

[0138] Furthermore, through a self-healing topology network mechanism, the factory can monitor the production system's topology in real time, identify abnormal nodes and connections, and dynamically adjust the network structure to further optimize production parameters and ensure overall system stability. This mechanism demonstrates remarkable adaptive capabilities during the production process, significantly reducing downtime and defective product rates caused by abnormal fluctuations in production.

[0139] The factory also introduced a self-evolving bifurcation control mechanism to monitor changes in key parameters in the production system in real time, identify parameters approaching bifurcation points, and adjust the evolution path of these parameters to ensure the system remains stable. Through multi-scale bifurcation tracking technology, the factory can simultaneously monitor system state changes at different temporal and spatial scales, making timely adjustments and preventing the production system from entering an unstable or abnormal state.

[0140] In the process of predicting and optimizing key parameters, the factory applied quantum cascade model prediction technology, combined with vortex quantum field effects, to provide real-time predictions of key production parameters and product quality. By superimposing quantum states and dynamically adjusting the vortex field, the factory generated a predictive model for optimizing production parameters. In practice, the factory can dynamically adjust the control inputs of the production system based on the prediction results, thereby further optimizing product quality and production efficiency.

[0141] Table 1: Comparison of main indicators before and after improvement of micro spring gasket production process

[0142] index Before improvement After improvement Defective rate 4.5％ 1.2％ Fluctuations in product quality 1.5％ 0.5％ Customer satisfaction 75％ 92％ Production efficiency standard 20% increase Energy consumption standard 15% reduction

[0143] After more than three months of implementation, the factory has achieved remarkable results. The defective rate has dropped from 4.5% to 1.2%, product quality fluctuations have been controlled within 0.5%, and customer satisfaction has increased from 75% to 92%. Furthermore, production efficiency has increased by 20%, and energy consumption has decreased by 15%. This has not only greatly enhanced the factory's market competitiveness but also significantly improved its economic benefits.

[0144] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning, characterized in that: The steps include: S1. Collect multi-source heterogeneous data in the production process of micro spring washers, form a multidimensional dataset, and preprocess it; S2. Based on the pre-processed multidimensional data set, a virtual simulation environment is constructed, and the production parameters are optimized using the reinforcement learning algorithm to generate a preliminary adjustment strategy; S3. Initialize the hierarchical progressive adaptive reinforcement learning model and verify the effectiveness of the strategy through a simulation environment; S4. During the actual production process, the multifunctional integrated sensing technology of smart materials is used to monitor key process parameters and environmental changes in real time, and production parameters are dynamically optimized based on feedback from a hierarchical progressive reinforcement learning model. S5. Through the self-repairing topology network mechanism, the topology structure of the production system is monitored in real time, abnormal nodes and connections are identified, and the network structure is dynamically adjusted to further optimize production parameters; S6. Apply the self-evolutionary bifurcation control mechanism to monitor the changes of key parameters in the production system in real time, identify parameters close to bifurcation points, and adjust the evolution path of parameters to ensure stable operation of the system; S7. Use quantum cascade model prediction technology to predict key parameters and product quality in the production process in real time, adjust production parameters, and optimize product quality; S8. Apply the optimized production parameters to actual production, continuously verify and adjust the model through a multi-source real-time feedback mechanism, and optimize the continuous adaptive adjustability and operational stability of the production process.

2. The method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning according to claim 1, characterized in that: The S3 specifically includes: S31. Initialize a hierarchical progressive adaptive reinforcement learning model, combine it with a multi-dimensional topological braiding network mechanism, and model and optimize the multi-dimensional parameters involved in the production process; S32. Input the preprocessed multidimensional dataset into the primary hierarchical model, construct a basic parameter braiding structure using a multidimensional topological braiding network mechanism, and perform preliminary optimization of the nodes and edges in the braiding network using a reinforcement learning algorithm to generate an optimized primary parameter combination; S33, input the optimization parameter combination generated by the primary level model into the intermediate level model, and the intermediate level model uses the dynamic adjustment function of the multi-dimensional topological weaving network to adjust the node weights w in the topological structure. i and edge weight w ij To optimize: in, represents the edge weight between node i and node j at the tth iteration, represents the difference between the states of node i and node j at the tth iteration, represents the square of the Euclidean distance between the feature vectors corresponding to node i and node j, α and β are hyperparameters that control the adjustment amplitude and decay speed, and γ is the neighborhood influence factor. and Represents the association matrix value between node i and node j and neighboring node k, Represents the association between node i and node j and their common neighboring nodes and their weight influence; S34. Input the optimization parameter combination generated by the intermediate-level model into the advanced-level model. The advanced-level model, in combination with the global production goal, utilizes the global optimization capability of the multi-dimensional topological weaving network to comprehensively adjust the entire network structure. The topological relationship between nodes is adjusted through the reinforcement learning algorithm. Based on the global optimization strategy of the weaving operation, a multi-objective optimization strategy is generated that comprehensively considers production efficiency, quality stability, and energy consumption. S35. The multi-objective optimization strategy generated by the advanced hierarchical model is verified through a simulation environment. The simulation environment uses the dynamic feedback function of the multi-dimensional topology weaving network mechanism to evaluate and adjust the optimization strategy in real time, and finally outputs the optimization parameter combination used in actual production. The results are fed back to the multi-dimensional topology weaving network to further improve its topological structure.

3. The method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning according to claim 1, characterized in that: The S4 specifically includes: S41. In actual production processes, the multifunctional integrated sensing technology of smart materials is used to monitor key process parameters, equipment operating status, and environmental changes in the production system in real time. Physical signals are collected through embedded sensors to generate high-frequency, multi-dimensional real-time data streams. S42, inputting the collected real-time data stream into the spatiotemporal hypersurface mapping algorithm module, mapping the multidimensional data onto a dynamic spatiotemporal hypersurface, and generating a hypersurface feature vector for a reinforcement learning model by capturing the spatiotemporal dependencies in the data; S43. Based on the multi-level dynamic graph energy optimization method, the dynamic adjustment of production parameters is optimized at multiple levels. At the primary level, the energy function of the system is defined as: Among them, E (t) represents the total energy of the system at the tth iteration, are the optimized node and edge weights, is the square of the Euclidean distance between the hypersurface feature vectors of node i and node j, is the state difference, α is the energy adjustment factor, β is the weight factor of the difference between the optimization target and the current state, Represents the difference between the current eigenvector and the optimized target eigenvector; S44. At a high level, the system energy function E (t) Perform multi-level optimization and dynamically adjust node and edge weights: Among them, λ is the step size parameter, Represents the energy function E (t) About edge weights The partial derivative of is the external interference factor between nodes, γ is the adjustment coefficient; S45. Dynamically adjust production parameters based on the feedback signal generated by the reinforcement learning model, and verify the adjustment results through the response mechanism of the smart material to form a closed-loop control.

4. The method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning according to claim 1, characterized in that: The S5 specifically includes: S51. Use the self-repairing topology network mechanism to monitor key parameters and topology status in the production system in real time, use embedded sensors to collect status information and connection status of each node in the production process, and generate real-time network topology data; S52: Input the collected network topology data into the topology entropy optimization module, and calculate the improved entropy value H of the network. (t) To evaluate the chaos and order of the network: Among them, H (t) represents the improved entropy value of the network at the tth iteration, is the state probability of node i at the tth iteration, are the optimized node and edge weights, is the square of the Euclidean distance between the hypersurface feature vectors of node i and node j, is the state difference, is the degree of influence of the optimization result on the current iteration, α and β are the adjustment factors to control each influence; S53, analysis entropy value H (t) The system detects changes in the network, identifies the most unstable or abnormal areas, and determines the nodes and connections that need to be repaired first. After identifying the high entropy areas, it further calculates the state stability and connection strength of each node in these areas to accurately locate the parts that need to be repaired. S54. Topology repair is performed on nodes and connections in the high entropy area. By dynamically adjusting the node state probability and connection weight, the network structure is optimized and the overall entropy of the system is reduced. The optimization formula is: in, represents the connection weight after repair at the t+1th iteration, λ is the step size parameter, which controls the convergence speed of weight adjustment. To improve the entropy value H (t) About edge weights The partial derivative of is the external interference factor, is the historical optimization result repair factor, and γ is the adjustment coefficient; S55. During the repair process, the entropy value changes of the repair area are continuously monitored, and the network structure is stabilized and ordered by dynamically adjusting the node state probability and connection weight; S56. After the repair is completed, the topology entropy optimization module feeds back the repaired network parameters to the control module of the self-repairing topology network mechanism, and optimizes the global entropy value of the repaired topology structure to further optimize the robustness and adaptability of the network.

5. The method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning according to claim 1, characterized in that: The S6 specifically includes: S61. In the actual production process, the self-evolving bifurcation control mechanism is applied to monitor the changes of key parameters in the production system in real time, and the key process parameters are collected by embedded sensors to generate high-frequency real-time data streams. S62, input the collected real-time data stream into the bifurcation point identification module, combine the multi-scale bifurcation tracking technology, synchronously monitor the system state changes at different time and space scales, and calculate the state change rate of each node and the mutual influence coefficient β between adjacent nodes ij , identify the key nodes in the system close to the bifurcation point: in, represents the state variable of node i at time t, represents the state variable of node j at time t, β ij is the mutual influence coefficient between node i and node j, γ i is the self-regulation coefficient of node i, Represents the nonlinear self-evolution function of node i, and the input control variable Related; S63: For the key nodes identified as being close to bifurcation points, path analysis and control are gradually performed based on the multi-scale bifurcation tracking technology. By adjusting the evolution path of the key nodes, the system is stabilized. in, represents the control input of node i in the next time step, λ is the step coefficient, and controls the adjustment amplitude. is the potential energy function of node i, κ is the adjustment coefficient, which is used to balance the influence between adjacent nodes; S64. Control input generated by the self-evolving bifurcation control mechanism Adjust key parameters in the production system in real time, input the adjusted parameters into the system's feedback module to verify the adjustment effect and further optimize the control strategy; S65. Feedback the adjusted system state to the reinforcement learning model. By comparing the actual system behavior with the expected goals, the adjustment strategy of production parameters is continuously optimized to ensure the long-term stability and efficient operation of the production system.

6. The method for adaptively adjusting production parameters of micro spring washers based on reinforcement learning according to claim 1, characterized in that: The S7 specifically includes: S71. In the actual production process, the quantum cascade model prediction technology is used in combination with the vortex quantum field effect to make real-time predictions of key process parameters and product quality in the production system; S72. Generate a prediction model for optimizing production parameters through the superposition of quantum states and dynamic adjustment of the vortex field; S73. In the prediction process, construct an innovative quantum prediction function based on multidimensional interactions Among them, Q i (t) represents the predicted value of the i-th key parameter at time t, α k is the weight coefficient in the quantum cascade model, Based on the current state variables and control inputs The quantum state function of is the influence of vortex quantum field effect on the i-th key parameter, is the connection weight of the key node after repair, Based on the status of adjacent nodes and optimized edge weights The quantum state function of , λ and γ are adjustment coefficients, which control the influence of vortex field effect and interaction between adjacent nodes respectively; S74. Based on the above prediction results, dynamically adjust the control input of the production system in, represents the control input to the i-th node in the next time step, μ is the step size coefficient, which controls the adjustment amplitude, ν is the adjustment coefficient, which controls the influence of the vortex quantum field effect, and κ is the influence coefficient of the adjacent node state on the control input of the current node; S75. Apply the updated control input to the production system and verify the accuracy of the prediction results through a feedback mechanism. If the prediction error exceeds a predetermined threshold, dynamically adjust the parameters of the quantum cascade model based on the feedback data until the system reaches stability.

Images