Injection molding parameter self-learning collaborative regulation and control system

Abstract

The invention relates to the technical field of injection molding process control, and discloses an injection molding parameter self-learning collaborative regulation and control system which is characterized in that multi-source parameters such as temperature and pressure are acquired in real time through a data acquisition module, and the multi-source parameters are mapped to a low-dimensional manifold space through a dynamic manifold construction module to decouple a nonlinear coupling relationship; the chaos synchronization control module suppresses external disturbance based on master-slave system design, and the distributed optimization module realizes multi-target real-time decision by adopting a split Bregman algorithm; the execution module converts the optimization instruction into a control signal and triggers closed-loop feedback; and the federated learning architecture realizes cross-device strategy migration by extracting continuous homology features of a manifold space and combining a Hausdorff distance and homeomorphic mapping, and drives a global model to update. According to the invention, the technical problems of difficult dynamic regulation and control of nonlinear parameters, insufficient real-time performance of multi-constraint optimization, low cross-scene repeated parameter adjustment efficiency and lack of self-adaptability of long-term process drift in the injection molding process are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of injection molding process control, and in particular to an injection molding parameter self-learning collaborative control system. Background Art

[0002] In the field of injection molding process control, traditional methods usually use linear or static models to independently control key parameters such as temperature and pressure. However, the nonlinear coupling relationship between temperature and pressure during melt flow will change dynamically with material properties and environmental disturbances, making it difficult to maintain process stability with single parameter control, and external disturbances can easily cause parameter fluctuations, affecting product consistency.

[0003] Existing injection molding optimization technologies are mostly based on centralized solvers to achieve multi-objective decision-making, but are limited by computing resources and real-time requirements. It is difficult to coordinate multiple constraints such as energy consumption and quality within a limited time. Especially under complex working conditions, it is easy to fall into local optimality or violate process constraints, and cannot meet the real-time control needs of high-precision injection molding.

[0004] In response to the injection molding production needs of multiple devices and multiple scenarios, the current method usually requires repeated parameter adjustments for different devices or materials, and lacks the ability to transfer knowledge across scenarios. This limitation is particularly prominent when new materials are introduced or molds are replaced, resulting in long debugging cycles and high costs, making it difficult to adapt to flexible production needs.

[0005] In addition, existing systems mostly rely on open-loop control strategies and lack the ability to adapt to long-term process drift. As equipment wears out or the environment changes, parameter settings gradually mismatch actual operating conditions and require manual intervention and adjustment, making it difficult to achieve continuous optimization and energy consumption control, which restricts the improvement of intelligent production levels. Summary of the invention

[0006] In response to the shortcomings of the prior art, the present invention provides a self-learning and collaborative control system for injection molding parameters, which solves the technical problems that nonlinear parameter coupling is difficult to dynamically control during the injection molding process, multi-constraint optimization lacks real-time performance, repeated parameter adjustment in new scenarios is inefficient, and long-term process drift lacks closed-loop adaptability.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an injection molding parameter self-learning collaborative control system, comprising: Data acquisition module, used to obtain the melt temperature, injection pressure, injection speed, material properties and environmental variables of the injection molding machine in real time; A dynamic manifold construction module receives the output data of the data acquisition module, maps high-dimensional parameters to low-dimensional manifold space through a nonlinear dimensionality reduction method, and decouples the nonlinear coupling relationship between the parameters; A chaotic synchronization control module, based on the manifold space state output by the dynamic manifold construction module, constructs a master-slave chaotic system and dynamically adjusts the coupling strength to suppress the influence of external disturbances on parameter stability; A distributed optimization module solves the multi-objective optimization problem according to the synchronization error signal of the chaotic synchronization control module and generates an injection molding machine control parameter instruction; An execution module sends the instructions generated by the distributed optimization module to the injection molding machine actuator and monitors the process parameter deviation in real time; The federated learning module receives the manifold spatial features and optimization strategies of each device, and realizes cross-scenario knowledge transfer through cloud aggregation and edge updates.

[0008] Preferably, the material physical property parameters acquired by the data acquisition module include material viscosity and thermal conductivity, and the environmental variables include mold temperature and environmental humidity.

[0009] Preferably, the nonlinear dimensionality reduction method comprises: Calculate the sample similarity matrix through the kernel function; Perform eigendecomposition on the normalized Laplace matrix to extract the low-dimensional manifold space; Online manifold projection of new data is performed via an incremental expansion algorithm.

[0010] Preferably, the incremental expansion algorithm weights historical feature vectors by kernel function similarity to achieve dynamic updating of the manifold space.

[0011] Preferably, the master-slave chaotic system comprises: The main system simulates parameter dynamics based on the physical equations of the injection molding process; Constructing a synchronous error feedback mechanism from the controller in the manifold space; The adaptive coupling matrix dynamically adjusts the control strength according to the Lyapunov stability condition.

[0012] Preferably, the distributed optimization module uses a splitting iterative algorithm to decompose the multi-objective optimization problem, including the following steps: Convert the optimization problem into augmented Lagrangian form; The optimization variables, auxiliary variables and Lagrange multipliers are updated in parallel through the edge computing unit.

[0013] Preferably, the closed-loop feedback mechanism of the execution module includes: Real-time monitoring of the deviation between actual parameters and target values; When the deviation exceeds the threshold, the online update of the manifold space and chaos control module is triggered.

[0014] Preferably, the federated learning architecture realizes knowledge transfer through the following steps: Each device extracts the continuous homology features of the manifold space and builds a local strategy library; The cloud aggregates the policy library and extracts common features across devices; Zero-shot policy transfer to new scenarios via geometric similarity computation.

[0015] Preferably, the geometric similarity is achieved by calculating the Hausdorff distance of the manifold space, and if the distance is less than a preset threshold, the homeomorphic mapping is triggered to generate an initial strategy.

[0016] The method for self-learning and collaborative control of injection molding parameters includes the following steps: Obtain the melt temperature, injection pressure, injection speed, material properties and environmental variables of the injection molding machine in real time and perform standardization; Map the standardized high-dimensional parameters to the low-dimensional manifold space to decouple the nonlinear coupling relationship between the parameters; A master-slave chaotic system is constructed based on the manifold space state, and external disturbances are suppressed by dynamically adjusting the coupling strength. According to the chaotic synchronization error signal, the multi-objective optimization problem of energy consumption and quality is solved in parallel in the edge computing unit to generate control instructions; Execute control instructions and monitor actual parameter deviations, triggering dynamic updates of manifold space and chaos control models; Periodically aggregate manifold features and strategies across devices, update the global model and send it to the edge to achieve zero-sample scenario migration.

[0017] The present invention provides an injection molding parameter self-learning collaborative control system. It has the following beneficial effects: 1. The present invention can effectively decouple the nonlinear coupling relationship between melt temperature, pressure and other parameters through the high-dimensional parameter dimensionality reduction mapping of the dynamic manifold construction module and the master-slave system design of the chaos synchronization control module, and suppress the influence of external disturbances on process stability in real time, thus solving the problem of insufficient accuracy of traditional linear control methods under complex working conditions.

[0018] 2. The distributed optimization module of the present invention is based on the split Bregman algorithm, combined with the parallel acceleration capability of the edge computing unit, which can complete multi-objective collaborative optimization of energy consumption, quality, etc. within milliseconds, generate high-precision control instructions, and meet the dual requirements of real-time and multiple constraints in the injection molding process.

[0019] 3. The present invention uses a federated learning architecture to extract the continuous homology features of the manifold space, combines cloud aggregation with Hausdorff distance-driven homeomorphic mapping, and can achieve zero-sample strategy migration in new scenarios, significantly reduce equipment debugging time, and solve the bottleneck problem of repeated parameter adjustment of traditional methods in new materials and new mold scenarios.

[0020] 4. The present invention forms a full-process closed loop from parameter collection, dynamic modeling, optimization and regulation to knowledge migration through the deviation feedback mechanism of the execution module and the periodic model update of federated learning, so that the system has continuous adaptive capabilities and maintains process stability and optimal energy consumption in long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a system architecture diagram of the present invention; DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Example: Please refer to the attached Figure 1 , an embodiment of the present invention provides an injection molding parameter self-learning collaborative control system, comprising: Data acquisition module, used to obtain the melt temperature, injection pressure, injection speed, material properties and environmental variables of the injection molding machine in real time; The data acquisition module is used to obtain multi-source heterogeneous data in the injection molding process in real time and convert it into a standardized format for subsequent modules to process. The construction of this module needs to cover key technical links such as sensor selection, data preprocessing and outlier filtering. The specific implementation methods are as follows: Sensor configuration and data acquisition: The data acquisition module includes temperature sensors, pressure sensors, speed encoders, material property detectors and environmental sensors. The temperature sensors are preferably placed at key nodes of the melt flow path (such as barrels, nozzles and mold cavities) to measure the melt temperature. And mold temperature The pressure sensor is installed in the injection cylinder and the mold cavity to monitor the injection pressure in real time. The speed encoder is integrated into the injection screw drive to obtain the injection speed. The material property tester measures the material viscosity through an online viscometer and thermal conductivity meter. and thermal conductivity Environmental sensors are preferably deployed outside the injection molding machine to collect environmental humidity. and ambient temperature.

[0024] Data standardization: The original data cannot be directly used for modeling due to dimensional differences and needs to be standardized. Parameters , its standardized value The calculation formula is: ; in For parameters The historical data average, The mean and standard deviation are preferably obtained based on historical data statistics during normal operation of the equipment, ensuring that the standardized data obeys a zero mean and unit variance distribution. This process can eliminate the dimensional differences between parameters and provide input data of uniform dimension for the dynamic manifold construction module.

[0025] Sliding window mechanism and outlier filtering: In order to suppress instantaneous noise interference, the data acquisition module introduces a sliding window mechanism. Define the window length The number of data points collected continuously is preferably set according to the injection cycle length. For the data in the window, the moving average of each parameter is first calculated. and standard deviation , and then the Laida criterion ( Criteria) Identify outliers: If a data point satisfy , it is determined as an outlier and removed. After removal, the missing data is preferably filled by linear interpolation. This mechanism can effectively improve data quality and prevent abnormal data from misleading input to subsequent modules.

[0026] Data caching and transmission protocol: The standardized data is transmitted to the dynamic manifold construction module via a high-speed bus. To ensure real-time performance, the data buffer preferably adopts a ring queue structure with a queue capacity of Determined according to the amount of data within the maximum injection molding cycle. The data transmission protocol is preferably based on a real-time Ethernet protocol (such as EtherCAT) to ensure millisecond latency. The data packet format definition includes timestamp, parameter type and value fields. The timestamp accuracy is preferably microseconds to achieve strict synchronization of multi-sensor data.

[0027] Data flow association between modules: The output of the data acquisition module includes the melt temperature , injection pressure , injection speed , material viscosity , thermal conductivity , mold temperature and ambient humidity This dataset is used as the input of the dynamic manifold construction module, providing it with basic data in the high-dimensional parameter space. At the same time, the closed-loop feedback signal of the execution module (such as the actual parameter deviation) triggers the dynamic parameter update of the data acquisition module (such as the mean With standard deviation online recalculation of the process), ensuring that standardized processing adapts to changes in process status.

[0028] The data acquisition module can achieve comprehensive perception and unified formatting of injection molding process parameters through multi-sensor collaborative measurement and standardized processing; the sliding window mechanism and outlier filtering method can effectively suppress noise interference and improve data reliability; standardized data output provides compatible input for subsequent dynamic manifold construction modules and supports decoupling analysis of parameter coupling relationships; real-time transmission protocol and caching mechanism ensure low latency and synchronization of data flow to meet the real-time requirements of closed-loop control.

[0029] The dynamic manifold construction module receives the output data of the data acquisition module, maps the high-dimensional parameters to the low-dimensional manifold space through the nonlinear dimensionality reduction method, and decouples the nonlinear coupling relationship between the parameters; The dynamic manifold construction module is used to map the high-dimensional parameters of the injection molding process to the low-dimensional manifold space and decouple the nonlinear coupling relationship between the parameters. This module is built based on a nonlinear dimensionality reduction algorithm, involving key technical links such as kernel function calculation, Laplace matrix decomposition, and incremental manifold update. The specific implementation methods are as follows: Input data preprocessing: The input of the dynamic manifold construction module is the standardized parameter set output by the data acquisition module ,in is the number of samples, is the parameter dimension. The input data must satisfy the conditions of zero mean and unit variance to ensure the stability of the manifold learning algorithm.

[0030] Kernel matrix construction and similarity measurement: To capture the nonlinear structure of high-dimensional parameters, the sample similarity matrix is ​​first constructed. It is preferred to use the Gaussian kernel function to calculate the kernel matrix , whose elements are defined as: ; in For the and The standardized parameter vector of samples, is the kernel function bandwidth parameter. The bandwidth parameter is preferably determined adaptively by the maximum and minimum distance method: calculate the mean of the distances between all samples With standard deviation ,set up ( is the adjustment factor, and the preferred value range is ). This design can adaptively adjust the local and global structural weights to avoid sensitivity deviations caused by manual settings.

[0031] Laplacian matrix construction and eigendecomposition: based on kernel matrix Construct the normalized Laplacian matrix , the calculation process is: Calculate degree matrix , which is a diagonal matrix with elements ; The normalized Laplacian matrix is ​​defined as: ; Pair Matrix Perform eigendecomposition and solve the characteristic equation , before selecting The eigenvector corresponding to the smallest non-zero eigenvalue , forming a low-dimensional manifold space . Manifold Dimension Preferably determined by cumulative variance contribution rate: Setting threshold (The preferred value is ), select the smallest Make This process can preserve the main structural information of the parameter space while achieving dimensionality compression.

[0032] Incremental manifold update (Nyström extension): When new samples When it arrives, it needs to be projected into the existing manifold space without recalculating the full eigendecomposition. The specific steps are: 1. Calculate the kernel similarity vector between new samples and historical samples ; : The kernel similarity vector between the new sample and the historical sample; 2. Calculate the coordinates of the new sample in the manifold space using the Nyström expansion formula : ; in For the eigenvalues, is the corresponding eigenvector, For the new sample The algorithm can realize the dynamic expansion of manifold space and adapt to the time-varying characteristics of injection molding process parameters.

[0033] Dynamic maintenance of manifold space: To avoid manifold dimensionality The model rigidity caused by the fixed dimension is introduced to introduce the dimension adaptive adjustment mechanism. injection molding cycle) to calculate the information loss rate of the current manifold space .like (The threshold is preferably ), then re-perform the eigendecomposition and update value, Adjust the threshold for the dimension, is the information loss rate. This mechanism can ensure the adaptability of the manifold space to the dynamic changes of the process.

[0034] Outputs are associated with downstream modules: The output of the dynamic manifold construction module is a low-dimensional manifold space coordinate matrix (including historical and newly added samples), this data is used as the input of the chaos synchronization control module to provide it with redundant and decoupled parameter states. and the eigenvector Called by the federated learning architecture for cross-device knowledge transfer.

[0035] The dynamic manifold construction module can effectively extract the nonlinear low-dimensional embedding structure of high-dimensional parameters through kernel function and Laplace matrix decomposition, thereby decoupling the complex coupling relationship between parameters; the incremental update algorithm can adapt to the dynamic changes of the injection molding process and avoid the resource overhead of full calculation; the dimensional adaptive adjustment mechanism ensures the robustness of the manifold space to changes in process states; the output data provides the chaotic synchronization control module with de-redundant, low-dimensional and decoupled parameter states, supporting subsequent disturbance suppression and optimization control.

[0036] The chaos synchronization control module builds a master-slave chaotic system based on the manifold space state output by the dynamic manifold construction module and dynamically adjusts the coupling strength to suppress the influence of external disturbances on parameter stability; The chaotic synchronization control module is used to suppress dynamic disturbances in the injection molding process and achieve rapid stabilization of parameter states through the master-slave chaotic system. This module builds a synchronous error feedback mechanism based on the manifold space state, involving key technical links such as master system modeling, slave controller design, and adaptive coupling strength adjustment. The specific implementation methods are as follows: Main chaotic system modeling: The main system simulates the nonlinear dynamic behavior of the injection molding process, and its state variable is the melt temperature. and injection pressure Based on the non-isothermal viscous fluid dynamics equation, the dynamic equation of the main system is defined as: ; in, It is the material property related parameters, obtained by matching the material property database; is the injection speed; is the external disturbance term (such as sudden change in material viscosity, mold temperature fluctuation). This equation can reflect the nonlinear coupling relationship between temperature and pressure and the influence of external disturbances.

[0037] Synchronize from controller design to manifold space: low-dimensional manifold space output by dynamic manifold building module The slave system is constructed in, and its dynamic equation is defined as: ; in is the main system state vector, is the slave system state vector, is a nonlinear function in manifold space, preferably constructed by a local linear approximation method: ; : Local linear coefficient matrix and bias term in manifold space; in is the local linear coefficient matrix of the manifold space, is a bias term obtained by fitting neighborhood samples in the manifold space.

[0038] Adaptive coupling matrix adjustment: coupling matrix Dynamically adjust according to synchronization error.

[0039] : Adaptive coupling matrix Defining Synchronization Error , its dynamic equation is: ; After substituting the master-slave system equation, the error dynamic equation can be simplified to: ; in is the disturbance term. To suppress the error divergence, it is necessary to determine the Lyapunov stability condition Adjustment rules. Lyapunov stability guarantee: Select the Lyapunov function , and its time derivative is: ; To satisfy , coupling matrix The adjustment rule is preferably: ; in is the gain coefficient, is the minimum coupling strength threshold, preferably set To ensure stability.

[0040] Disturbance rejection and output association: Synchronization error of master-slave system As a disturbance estimation signal, it is input into the distributed optimization module to correct the constraints in the multi-objective optimization model. Called by the federated learning architecture for state matching during cross-device policy migration.

[0041] The chaotic synchronization control module can achieve rapid suppression of external disturbances through master-slave system design and adaptive coupling adjustment; the local linear approximation method based on manifold space can accurately describe the nonlinear dynamic characteristics; the Lyapunov stability condition ensures the exponential convergence of the synchronization error; the output signal provides a disturbance compensation basis for the downstream optimization module, while supporting the state migration requirements of the federated learning architecture.

[0042] The distributed optimization module solves the multi-objective optimization problem and generates the control parameter instructions of the injection molding machine according to the synchronization error signal of the chaotic synchronization control module; The distributed optimization module is used to solve the multi-objective optimization problem of the injection molding process in real time and generate control parameter instructions. This module realizes the distributed solution of the constrained optimization problem based on the split Bregman algorithm, involving key technical links such as objective function modeling, augmented Lagrangian form transformation and parallel iterative calculation. The specific implementation methods are as follows: Multi-objective optimization modeling: The input of the optimization problem is the synchronization error signal output by the chaos synchronization control module. and dynamic manifold space state The objective function is defined as the weighted sum of energy consumption and quality deviation: ; in is the parameter to be optimized, is the energy consumption function, is the quality index vector. Weight coefficient Preferably determined by the analytic hierarchy process (AHP) to balance the priorities of energy consumption and quality objectives.

[0043] Constraint definition: The optimization problem must satisfy the process feasibility constraints: 1. Melt temperature range: , boundary value Determined according to the thermal stability of the material; 2. Melt flow continuity equation: , ensuring that the melt flows without voids or stagnation during injection; 3. Synchronous error compensation constraints: ,in is the nominal parameter, is the error compensation coefficient, and the preferred value is .

[0044] Augmented Lagrangian form transformation: To handle the constraints, the original problem is transformed into augmented Lagrangian form:

[0045] in is an auxiliary variable, is the Lagrange multiplier, is the penalty coefficient, preferably the initial value is determined by experiment And dynamically adjust.

[0046] Split Bregman algorithm iteration: The optimization problem is decomposed into the following sub-problems and solved iteratively: Update optimization variables : ; This subproblem is solved by the conjugate gradient method, using the sparsity of the Hessian matrix to accelerate the calculation.

[0047] Update auxiliary variables : ; The contraction operator is defined as , used to implement constrained soft thresholding.

[0048] Update Lagrange multipliers : ; The iteration termination condition is preferably the relative residual ( Preferably ).

[0049] Edge computing acceleration: To meet the real-time requirements, each sub-problem is executed in parallel on the edge computing unit. Specifically: variable The updates are distributed to the FPGA chip, using hardware parallelism to accelerate matrix operations; variable and The update of is performed on the GPU, and the contraction operator and the vector update are parallelized by CUDA kernels.

[0050] Data exchange is implemented through a shared memory pool to ensure the global consistency of variable states during iteration.

[0051] Outputs are associated with downstream modules: The output of the optimization module is the control parameter instruction , directly sent to the execution module to drive the injection molding machine action. At the same time, the constraint violation amount in the optimization process Feedback to the chaos synchronization control module to adjust the synchronization error compensation coefficient , forming a closed-loop optimization mechanism.

[0052] The execution module sends the instructions generated by the distributed optimization module to the actuator of the injection molding machine and monitors the deviation of process parameters in real time; The execution module is used to convert the control parameter instructions generated by the distributed optimization module into the control signal of the injection molding machine actuator, and monitor the actual parameter deviation in real time to trigger closed-loop feedback. This module involves key technical links such as instruction conversion, actuator drive, deviation calculation and feedback triggering. The specific implementation methods are as follows: Control instruction parsing and conversion: The optimization instructions received by the execution module are standardized parameter vectors Since the physical range of the actuator does not match the standardized dimension, an inverse standardization conversion is required: ; in For the The physical value of a parameter (such as the actual temperature in °C), is the mean and standard deviation of the corresponding parameters in the data acquisition module. This transformation can map the dimensionless command to the operable range of the actuator.

[0053] Actuator drive and control: The converted physical instructions are sent to the injection molding machine actuator via the field bus, including: Heater control: PID algorithm is used to adjust the heater power to make the melt temperature Tracking setpoints The PID output is: ; in , proportionality coefficient , integral coefficient , differential coefficient Preferably the setting is carried out by the Ziegler-Nichols method.

[0054] Hydraulic pump control: According to the pressure setting value Adjust the hydraulic valve opening to control the injection pressure . Valve opening The relationship with pressure is modeled as: ; in is the pressure gain coefficient, is an offset term and is determined through hydraulic system calibration experiments.

[0055] Screw speed control: The screw is driven by a servo motor to control the injection speed Matching setting value . Motor speed command Calculated as: ; in is the screw diameter, preferably read from the equipment parameter library according to the injection molding machine model.

[0056] Actual parameter monitoring and deviation calculation: During the execution process, the data acquisition module obtains the melt temperature in real time , injection pressure And injection speed . Define the comprehensive deviation index for: ; in is the actual parameter measurement value, is the standard deviation of the standardized parameter. This indicator can quantify the multi-dimensional deviation between the set value and the actual value.

[0057] Feedback trigger mechanism: When the deviation index Exceeding the preset threshold , triggering the closed-loop feedback process: Send an update request to the dynamic manifold construction module to recalculate the manifold space coordinates ; Send disturbance compensation instructions to the chaos synchronization control module to adjust the coupling matrix The gain factor ; Update the constraint weights of the distributed optimization module , prioritize reducing the target item with the largest current deviation.

[0058] Threshold It is preferably set to the 95% quantile of the historical deviation distribution and dynamically adjusted through statistical learning to adapt to process changes.

[0059] Data flow association between modules: Deviation signal of the execution module As the input of closed-loop feedback, it drives the online update of the dynamic manifold building module and the chaos synchronization control module. At the same time, the actual control quantity of the actuator (such as heating power , valve position opening ) are collected by the federated learning architecture and used to build a device-level energy consumption model.

[0060] The execution module can achieve accurate mapping of optimization instructions to physical signals through inverse normalization transformation and actuator control model; PID algorithm and hydraulic model ensure stable tracking of temperature, pressure and speed; comprehensive deviation index quantifies the degree of deviation of multiple parameters and provides decision-making basis for closed-loop feedback; feedback trigger mechanism drives dynamic update of upstream modules to form an adaptive control closed loop; execution data supports device-level modeling of federated learning architecture to realize experience transfer across devices.

[0061] The federated learning module receives the manifold spatial features and optimization strategies of each device, and realizes cross-scenario knowledge transfer through cloud aggregation and edge updates.

[0062] The federated learning module is used to achieve cross-device / scenario knowledge sharing and migration, and build a global optimization strategy by aggregating the local experience of edge devices. This architecture involves key technical links such as local feature extraction, cloud aggregation, and zero-sample migration. The specific implementation methods are as follows: Local policy library construction: Each edge device obtains the manifold space coordinate matrix from the dynamic manifold construction module , and extract its continuous homology features The specific steps are: 1. Yes Construct the Vietoris-Rips complex and set the maximum neighborhood radius , preferably determined based on 20%-30% of the manifold space diameter.

[0063] 2. Calculate the persistent homology group , record the birth radius of each topological feature (connected branches, holes) Death Radius .

[0064] 3. Generate a persistence graph , describing the topological stability of the manifold space. This feature can reflect the process differences and commonalities between devices.

[0065] Cloud aggregation and meta-feature extraction: The cloud server receives all devices' , calculate the common topological characteristics through the optimal transmission theory. The specific process is: 1. Build a joint persistence graph ; :equipment Persistence graph of 2. Yes Perform density clustering to identify high-frequency feature point sets , the judgment conditions are: ; in is the kernel width parameter, the preferred value is ; is the frequency threshold, preferably set to the total number of devices 50% of.

[0066] 3. Extract meta-feature vector , quantifying the average persistence and dispersion of the topological structure, : meta-feature vector.

[0067] Zero-shot policy migration: When a new device is connected, its manifold space is calculated Manifold Space with Historical Devices The Hausdorff distance: ; like (The threshold is preferably the 90% quantile of the historical distance distribution), then through the homeomorphic mapping Migration strategy, : Hausdorff distance.

[0068] 1. In and The point correspondence is established between them, preferably using Procrustes analysis to minimize the mapping error: ; 2. Corresponding optimization strategy pass Reverse mapping to new device space to generate initial strategy .

[0069] Edge model update: The cloud will aggregate the meta-feature vector and migration strategies Send to edge devices, triggering the following updates: 1. Dynamic manifold building blocks based on Adjusting the manifold space dimensions ,ensure that it is compatible with the global topology; 2. Distributed optimization module will As the initial solution, it accelerates the convergence of local optimization; 3. Chaos synchronization control module is based on Reinitialize the coupling matrix , enhance the stability of cross-device synchronization.

[0070] The federated learning module can capture the topological commonalities of the device manifold space through continuous homological feature extraction; the cloud aggregation algorithm identifies stable features across devices and supports global experience abstraction; the Hausdorff distance and homeomorphic mapping realize zero-sample strategy migration for new scenarios and reduce cold start time; the edge model update mechanism ensures that the local module is compatible with global knowledge and improves the generalization ability of injection molding parameter control.

[0071] Another embodiment of the present invention provides a method for self-learning and coordinated control of injection molding parameters, comprising the following steps: Obtain the melt temperature, injection pressure, injection speed, material properties and environmental variables of the injection molding machine in real time and perform standardization; Map the standardized high-dimensional parameters to the low-dimensional manifold space to decouple the nonlinear coupling relationship between the parameters; A master-slave chaotic system is constructed based on the manifold space state, and external disturbances are suppressed by dynamically adjusting the coupling strength. According to the chaotic synchronization error signal, the multi-objective optimization problem of energy consumption and quality is solved in parallel in the edge computing unit to generate control instructions; Execute control instructions and monitor actual parameter deviations, triggering dynamic updates of manifold space and chaos control models; Periodically aggregate manifold features and strategies across devices, update the global model and send it to the edge to achieve zero-sample scenario migration.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Injection molding parameter self-learning collaborative control system, characterized in that: include: Data acquisition module, used to obtain the melt temperature, injection pressure, injection speed, material properties and environmental variables of the injection molding machine in real time; A dynamic manifold construction module receives the output data of the data acquisition module, maps high-dimensional parameters to low-dimensional manifold space through a nonlinear dimensionality reduction method, and decouples the nonlinear coupling relationship between the parameters; A chaotic synchronization control module, based on the manifold space state output by the dynamic manifold construction module, constructs a master-slave chaotic system and dynamically adjusts the coupling strength to suppress the influence of external disturbances on parameter stability; A distributed optimization module solves a multi-objective optimization problem according to a synchronization error signal of the chaotic synchronization control module and generates a control parameter instruction for the injection molding machine; An execution module sends the instructions generated by the distributed optimization module to the injection molding machine actuator and monitors the process parameter deviation in real time; The federated learning module receives the manifold spatial features and optimization strategies of each device, and realizes cross-scenario knowledge transfer through cloud aggregation and edge updates.

2. The injection molding parameter self-learning collaborative control system according to claim 1, characterized in that: The material property parameters acquired by the data acquisition module include material viscosity and thermal conductivity, and the environmental variables include mold temperature and environmental humidity.

3. The injection molding parameter self-learning collaborative control system according to claim 1, characterized in that: The nonlinear dimensionality reduction method comprises: Calculate the sample similarity matrix through the kernel function; Perform eigendecomposition on the normalized Laplace matrix to extract the low-dimensional manifold space; Online manifold projection of new data is performed via an incremental expansion algorithm.

4. The injection molding parameter self-learning collaborative control system according to claim 3, characterized in that: The incremental expansion algorithm achieves dynamic updating of the manifold space by weighting historical feature vectors through kernel function similarity.

5. The injection molding parameter self-learning collaborative control system according to claim 1, characterized in that: The master-slave chaotic system comprises: The main system simulates parameter dynamics based on the physical equations of the injection molding process; Constructing a synchronous error feedback mechanism from the controller in the manifold space; The adaptive coupling matrix dynamically adjusts the control strength according to the Lyapunov stability condition.

6. The injection molding parameter self-learning collaborative control system according to claim 1, characterized in that: The distributed optimization module uses a splitting iterative algorithm to decompose the multi-objective optimization problem, including the following steps: Convert the optimization problem into augmented Lagrangian function form; The optimization variables, auxiliary variables and Lagrange multipliers are updated in parallel through the edge computing unit.

7. The injection molding parameter self-learning collaborative control system according to claim 1, characterized in that: The closed-loop feedback mechanism of the execution module includes: Real-time monitoring of the deviation between actual parameters and target values; When the deviation exceeds the threshold, the online update of the manifold space and chaos control module is triggered.

8. The injection molding parameter self-learning collaborative control system according to claim 1, characterized in that: The federated learning architecture achieves knowledge transfer through the following steps: Each device extracts the continuous homology features of the manifold space and builds a local strategy library; The cloud aggregates the policy library and extracts common features across devices; Zero-shot policy transfer to new scenarios via geometric similarity computation.

9. The injection molding parameter self-learning collaborative control system according to claim 8, characterized in that: The geometric similarity is achieved by calculating the Hausdorff distance of the manifold space. If the distance is less than a preset threshold, the homeomorphic mapping is triggered to generate an initial strategy.

10. A method for self-learning and collaborative control of injection molding parameters, according to the self-learning and collaborative control system for injection molding parameters according to any one of claims 1 to 9, characterized in that: The following steps are involved: Obtain the melt temperature, injection pressure, injection speed, material properties and environmental variables of the injection molding machine in real time and perform standardization; Map the standardized high-dimensional parameters to the low-dimensional manifold space to decouple the nonlinear coupling relationship between the parameters; A master-slave chaotic system is constructed based on the manifold space state, and external disturbances are suppressed by dynamically adjusting the coupling strength. According to the chaotic synchronization error signal, the multi-objective optimization problem of energy consumption and quality is solved in parallel in the edge computing unit to generate control instructions; Execute control instructions and monitor actual parameter deviations, triggering dynamic updates of manifold space and chaos control models; Periodically aggregate manifold features and strategies across devices, update the global model and send it to the edge to achieve zero-sample scenario migration.

Images