Cross-factory collaborative production quality optimization method based on federated learning

Abstract

The application provides a cross-factory collaborative production quality optimization method based on federated learning, and relates to the technical field of data processing, wherein the method comprises the following steps: acquiring a plurality of identified production node groups corresponding to a plurality of target factories, identifying the intersection identified production node from the plurality of identified production node groups, collecting a circuit control operation parameter set of the intersection identified production node, performing federated learning on the circuit control operation parameter set according to a collaborative control optimization target, acquiring a global optimization control parameter, generating a plurality of local optimization control parameters based on the plurality of target factories according to the global optimization control parameter, and optimizing the production quality of the intersection identified production node. The technical problem that the prior art cannot accurately identify and optimize the bottom-layer common execution unit affecting the production quality is solved. The technical effect of accurately targeting the bottom-layer common execution unit affecting the production quality, collaboratively utilizing the cross-factory operation data, and realizing the global optimization of the production control parameter is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a cross-factory collaborative production quality optimization method based on federated learning. BACKGROUND

[0002] In the existing manufacturing industry, in order to improve the production quality of products, it is usually relied on macro adjustment means at the process level, such as optimizing the robot trajectory, adjusting the processing torque, or centrally controlling the driving end parameters and the equipment running state. However, the fundamental factors affecting the product quality and production efficiency are usually at the bottom layer of the executor, including the control accuracy, response speed and energy conversion efficiency of the executor. The existing analysis and optimization methods pay insufficient attention to these bottom layer factors, lack optimization improvement from the executor performance, and do not fully consider the real-time control ability of the data execution end and the coordination between nodes. In addition, the data distribution characteristics across factories make data privacy protection a key factor restricting comprehensive optimization, and the existing methods are difficult to achieve collaborative use of distributed data while strictly protecting the sensitive data of each factory. Therefore, simply relying on macro adjustment at the process level is difficult to significantly improve the overall production quality, and it is also difficult to guarantee the precision, consistency and reliability of the production process. In summary, under the premise of strictly protecting the data privacy of each factory, the existing technology has the technical problems that it cannot effectively overcome the device heterogeneity, it is difficult to accurately identify and optimize the bottom layer public execution unit affecting the production quality, it lacks the collaborative utilization ability of the distributed running data across factories, it cannot realize the global optimization of the production control parameters, and it affects the product production quality. SUMMARY

[0003] The purpose of the present application is to provide a cross-factory collaborative production quality optimization method based on federated learning, which solves the technical problems that under the premise of strictly protecting the data privacy of each factory, the existing technology cannot effectively overcome the device heterogeneity, it is difficult to accurately identify and optimize the bottom layer public execution unit affecting the production quality, it lacks the collaborative utilization ability of the distributed running data across factories, it cannot realize the global optimization of the production control parameters, and it affects the product production quality.

[0004] In view of the above problems, the application provides a cross-factory collaborative production quality optimization method based on federated learning, which comprises the following steps: acquiring a plurality of identified production node groups corresponding to a plurality of target factories, wherein the identified production node is a production node driven by an H-bridge driving circuit, and the identified production node group is a combination of identified production nodes of each target factory; identifying the plurality of identified production node groups to obtain an intersection identified production node, and collecting a circuit control operation parameter set of the intersection identified production node, wherein the circuit control operation parameter set comprises H-bridge static configuration parameters and H-bridge dynamic operation parameters; defining a collaborative control optimization target, performing federated learning on the circuit control operation parameter set according to the collaborative control optimization target, acquiring a global optimization control parameter, and generating a plurality of local optimization control parameters based on the plurality of target factories according to the global optimization control parameter; and performing production quality optimization on the intersection identified production node by using the plurality of local optimization control parameters.

[0005] Optionally, the plurality of identified production node groups are identified to obtain H-bridge electrical property information, H-bridge driving control objects and H-bridge driving functions; feature recognition is performed according to the H-bridge electrical property information, H-bridge driving control objects and H-bridge driving functions to determine electrical specification feature vectors, mechanical load feature vectors and function target feature vectors; the intersection identified production node is clustered according to the electrical specification feature vectors, mechanical load feature vectors and function target feature vectors to obtain a plurality of groups of intersection identified production nodes; and a plurality of federated learning groups are constructed according to the plurality of groups of intersection identified production nodes.

[0006] Optionally, a plurality of circuit control operation parameter sets corresponding to the plurality of federated learning groups are collected; federated learning is performed on the plurality of circuit control operation parameter sets to obtain a plurality of global optimization control parameters, and the plurality of global optimization control parameters are used to perform production quality optimization on the local optimization control parameters of the plurality of groups of intersection identified production nodes, respectively.

[0007] Optionally, the H-bridge static configuration parameters comprise switch device types, rated voltages, current capacities and dead time settings, and the H-bridge dynamic operation parameters comprise PWM frequencies, duty cycle curves, current loop control gains, temperature rises and electric energy conversion efficiencies.

[0008] Optionally, a global optimization model is initialized; the initialized global optimization model is distributed to the plurality of target factories, the plurality of target factories train the initialized global optimization model according to corresponding circuit control operation parameter sets and collaborative control optimization target score labels to obtain a plurality of local optimization models; model update parameters corresponding to the plurality of local optimization models are uploaded, the initialized global optimization model is iteratively trained by using the corresponding model update parameters for multiple rounds to obtain a converged global optimization model and corresponding global optimization control parameters.

[0009] Optionally, the model parameters of the initialized global optimization model are updated by using corresponding model update parameters to obtain a round global optimization model; the round global optimization model is sent to the plurality of target factories to obtain a plurality of round local optimization models, the model parameters of the round global optimization model are updated according to corresponding round model update parameters of the plurality of round local optimization models to obtain a second round global optimization model, and the process is repeated until a global optimization model meeting a collaborative control optimization target score threshold is obtained as a converged global optimization model.

[0010] Optionally, the collaborative control optimization target is a multi-objective optimization function, and the multi-objective optimization function is a linear weighted fitting function of a plurality of optimization targets; wherein the plurality of optimization targets include an energy efficiency target, a control performance target, a stability target and a thermal management target.

[0011] Optionally, the optimization circuit control operation parameters of the intersection identified production nodes and a quality optimization increase index are continuously monitored; the optimization circuit control operation parameters greater than a preset quality optimization increase index threshold are identified according to the size of the quality optimization increase index to obtain identified optimization circuit control operation parameters; and the federated learning optimization is performed again according to the identified optimization circuit control operation parameters to obtain updated global optimization control parameters.

[0012] Optionally, initial collaborative production result data of the intersection identified production nodes are recorded; corresponding optimization collaborative production result data of the intersection identified production nodes based on the optimization circuit control operation parameters are collected; the initial collaborative production result data and the optimization collaborative production result data are compared to obtain a quality optimization increase index.

[0013] The application also provides a computer readable storage medium, which stores a computer program, and the computer program realizes the steps of the federated learning based cross-factory collaborative production quality optimization method when executed.

[0014] One or more technical solutions provided in the application have at least the following technical effects or advantages:

[0015] The method for optimizing production quality across factories based on federated learning provided by the embodiments of the present application comprises the following steps: obtaining a plurality of identified production node groups corresponding to a plurality of target factories, wherein the identified production node is a production node driven by an H-bridge driving circuit, and the identified production node group is a combination of identified production nodes of each target factory; identifying the plurality of identified production node groups to obtain an intersection identified production node, and collecting a circuit control operation parameter set of the intersection identified production node, wherein the circuit control operation parameter set comprises H-bridge static configuration parameters and H-bridge dynamic operation parameters; defining a collaborative control optimization target, performing federated learning on the circuit control operation parameter set according to the collaborative control optimization target, obtaining a global optimization control parameter, generating a plurality of local optimization control parameters corresponding to the plurality of target factories based on the global optimization control parameter; and using the plurality of local optimization control parameters to optimize the production quality of the intersection identified production node. The method achieves global optimization of production control parameters by overcoming equipment heterogeneity and accurately targeting the bottom-layer common execution unit that affects production quality, while strictly protecting the data privacy of each factory and collaboratively using distributed operation data across factories.

[0016] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the following detailed description of the embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following detailed description of the embodiments of the present application. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can obtain other drawings from the provided drawings without creating any creative labor.

[0018] Figure 1 The flowchart of the method for optimizing production quality across factories based on federated learning provided by the present application.

[0019] Figure 2 The flowchart of obtaining a global optimization control parameter in the method for optimizing production quality across factories based on federated learning provided by the present application. DETAILED DESCRIPTION

[0020] This application provides a cross-factory collaborative production quality optimization method based on federated learning. It addresses the technical problems of existing technologies, which, while strictly protecting the data privacy of each factory, cannot effectively overcome equipment heterogeneity, struggle to accurately identify and optimize underlying common execution units affecting production quality, and lack the ability to collaboratively utilize distributed operational data across factories. This results in the inability to achieve global optimization of production control parameters, thus impacting product quality. The proposed method achieves the technical effect of overcoming equipment heterogeneity, accurately targeting underlying common execution units affecting production quality, and collaboratively utilizing distributed operational data across factories to achieve global optimization of production control parameters, all while strictly protecting the data privacy of each factory.

[0021] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0022] Example 1, as Figure 1 As shown, this application provides a method for optimizing cross-factory collaborative production quality based on federated learning. This method includes:

[0023] Obtain multiple identification production node groups corresponding to multiple target factories. The identification production node is a production node driven by an H-bridge driver circuit, and the identification production node group is a combination of identification production nodes for each target factory.

[0024] Specifically, first, a plurality of target factories are determined, which are factories participating in collaborative production quality optimization and belong to the same factory type, i.e., the plurality of target factories produce the same or similar products. Then, by analyzing the electrical design drawings, equipment technical documents, and production process records of each factory, production nodes in the factory driven by the bottom-layer common execution unit H-bridge driving circuit are identified as identified production nodes, and the relevant information of the identified production nodes is recorded, including the functional description of the nodes, the circuit topology structure, and the operating parameters, etc. The identified production nodes play a control role in the production process of the factory products. The H-bridge driving circuit is a common power electronic circuit composed of four switching elements such as power transistors, MOSFETs, or IGBTs. By controlling the conduction and cutoff of the switching elements on the diagonal, the direction of the current is bidirectionally controlled, thereby driving the motor to reverse, adjusting the speed, or achieving power inversion. The H-bridge driving circuit adopts a half-bridge or full-bridge form, and the specific driving methods include switching PWM control, closed-loop current control, speed feedback control, or position feedback control, etc. The H-bridge driving circuit is directly connected to the production node equipment through input control signals, such as stepper motors, servo motors, linear actuators, or conveyor motor, and by adjusting the PWM duty cycle or current loop control parameters, precise motion control, speed adjustment, or torque adjustment are achieved. At the same time, the H-bridge driving circuit can be combined with sensor feedback such as position sensors, encoders, or Hall sensors to form a closed-loop control, ensuring high precision and high stability of the node operation. For example, the plurality of target factories are a plurality of electronic product manufacturing factories distributed in different regions, and each device in the plurality of electronic product manufacturing factories is analyzed to determine the production nodes driven by the H-bridge driving circuit. For example, in the placement machine of the electronic product manufacturing factory, the H-bridge driving circuit is used to control the precise movement of the placement head to ensure that electronic components can be accurately placed at the predetermined position on the circuit board, and in the automatic optical detection equipment, the H-bridge driving circuit is used to control the movement of the detection platform to detect the circuit board in all directions. Moreover, in the transmission belt system of the circuit board, the H-bridge driving circuit is used to control the start, stop, and speed adjustment of the conveyor belt to ensure the smooth progress of the production process. Then, all the identified production nodes in each factory are aggregated to form an identified production node group for each factory, which contains the identified production nodes and the functional description of the corresponding nodes, the circuit topology structure, and the operating parameters, etc., and can fully reflect the production characteristics and control requirements of the factory.

[0025] The plurality of identified production node groups are identified to obtain intersection identified production nodes, and the circuit control operating parameter set of the intersection identified production nodes is collected, including H-bridge static configuration parameters and H-bridge dynamic operating parameters.

[0026] Further, the H-bridge static configuration parameters include switch device type, rated voltage, current capacity and dead time setting, and the H-bridge dynamic operation parameters include PWM frequency, duty cycle curve, current loop control gain, temperature rise and power conversion efficiency.

[0027] Specifically, the identification production node groups of all target factories are obtained, and based on the similarity of the identification production nodes, the identification production nodes with common characteristics in multiple target factories are identified through electrical parameters, control strategies and functional targets. For example, the similarity of the characteristics of the identification production nodes in different target factories is calculated using similarity measurement methods such as Euclidean distance, cosine similarity, etc. By setting a similarity threshold, the identification production nodes with similar electrical configuration, control characteristics and functional targets are extracted as intersection identification production nodes, which refer to production nodes with the same or similar H-bridge drive circuit parameters in multiple factories. For example, in an electronic product manufacturing plant, multiple production factories assemble circuit boards, each factory uses different H-bridge drive circuits to drive its automation equipment, and the identification production nodes of each factory may differ in electrical configuration and function, but their goal is to produce the same type of circuit board. Through feature matching of the identification production nodes, the intersection identification production nodes are obtained. Then the circuit control operation parameters of each factory intersection identification production node are collected to form a circuit control operation parameter set. The circuit control operation parameter set includes two parts: H-bridge static configuration parameters and H-bridge dynamic operation parameters. H-bridge static configuration parameters refer to parameters set in advance before the node device runs, including switch device type, rated voltage, current capacity and dead time setting. The switch device type, such as MOSFET and IGBT, determines the switching speed and power carrying capacity of the circuit. The rated voltage refers to the maximum voltage value that the H-bridge drive circuit can withstand, which is related to the power voltage in the circuit. The current capacity refers to the maximum current value that the H-bridge drive circuit can withstand, which determines the ability of the circuit under high load. The two switching devices in the H-bridge drive circuit need to leave a certain time interval when switching, called dead time, which avoids short circuit and ensures the safe operation of the circuit. H-bridge static configuration parameters determine the basic electrical characteristics and operating range of the H-bridge drive circuit, which are usually stored in the controller of the H-bridge drive circuit and can be directly read through the H-bridge drive circuit configuration file, control module or identification label on the circuit board. H-bridge dynamic operation parameters refer to parameters that change over time during the operation of the node device, including PWM frequency, duty cycle curve, current loop control gain, temperature rise and power conversion efficiency. Through sensors and data acquisition modules installed on the production node, the running state of the device is monitored in real time, and H-bridge dynamic operation parameters are collected. For example, the current sensor monitors the current change in real time to provide current loop control gain and current flow data. The temperature sensor, such as thermocouple or RTD sensor, monitors the temperature rise in the circuit. The voltage sensor monitors voltage fluctuations to calculate PWM frequency and duty cycle. At the same time, the power meter measures the power conversion efficiency by collecting input and output power data in real time.The H-bridge driving circuit uses pulse width modulation (PWM) technology to control the opening and closing of the switching device. The PWM frequency determines the switching speed of the signal, affecting the smoothness and efficiency of the power output. The duty cycle refers to the ratio of the high-level duration to the cycle time in the PWM signal. By adjusting the duty cycle, the output voltage or current can be accurately controlled. The current loop is used to control the stability of the current. The current loop control gain determines the response speed and stability of the current loop control. The temperature rise refers to the temperature change during the operation of the circuit. The power conversion efficiency refers to the ratio of input power to output power. An efficient H-bridge driving circuit can minimize energy loss and improve the overall efficiency of the device. By accurately identifying the intersection of the production node and collecting its circuit control operation parameter set, data support is provided for subsequent federated learning optimization. The collected static configuration parameters and dynamic operation parameters can fully reflect the electrical characteristics and operating state of the production node, ensuring that the data of each production node accurately reflects its true operating state during cross-factory collaboration, providing a reliable basis for optimizing the global control strategy, thereby improving the production efficiency of each factory and optimizing the control performance while ensuring production stability.

[0028] The method further comprises defining a collaborative control optimization target, performing federated learning on the circuit control operation parameter set according to the collaborative control optimization target, obtaining a global optimization control parameter, and generating a plurality of local optimization control parameters corresponding to the plurality of target factories based on the global optimization control parameter.

[0029] Further, the collaborative control optimization target is a multi-objective optimization function, and the multi-objective optimization function is a linear weighted fitting function of a plurality of optimization objectives. The plurality of optimization objectives include energy efficiency, control performance, stability, and thermal management.

[0030] As shown in FIG. 8, in one possible implementation, the method for performing federated learning on the circuit control operation parameter set according to the collaborative control optimization target to obtain a global optimization control parameter comprises the following steps. Figure 2 The method comprises the following steps: initializing a global optimization model; distributing the initialized global optimization model to the plurality of target factories; training the initialized global optimization model according to the corresponding circuit control operation parameter set and the collaborative control optimization target score label to obtain a plurality of local optimization models; uploading the model update parameters corresponding to the plurality of local optimization models; performing multi-round iterative training on the initialized global optimization model using the corresponding model update parameters to obtain a converged global optimization model and the corresponding global optimization control parameter.

[0031] Specifically, according to the actual needs of the factory, the production process of the factory, the equipment capacity, the product specifications and the quality standards are analyzed, and the production targets of the factory are combined to select indicators reflecting the production quality and efficiency, such as energy efficiency targets, control performance targets, stability targets and thermal management targets, etc. The multi-index definition collaborative control optimization target is defined, which is a comprehensive performance index for evaluating and guiding the operation of the H-bridge driving node in the cross-factory collaborative production process. The multiple optimization targets include energy efficiency targets, control performance targets, stability targets and thermal management targets. The energy efficiency target is to maximize the conversion efficiency of the H-bridge driving circuit from electrical energy to mechanical energy, optimize the energy consumption of the circuit during operation, reduce unnecessary energy loss, and thus improve the overall energy efficiency. The control performance target is to minimize the tracking error of the motor, such as position overshoot, steady-state error, etc., to optimize the control accuracy, response time and stability of the circuit, and to ensure efficient and accurate control of the production process. The stability target is to minimize the current ripple and motor operating vibration amplitude, optimize the dynamic response of the circuit, and ensure stable operation of the motor under different working conditions to avoid production quality fluctuations caused by instability. The thermal management target is to minimize the temperature rise of power devices to avoid overheating of equipment. According to the actual production needs and historical production data, the influence of each target on the production quality or equipment life is evaluated, and the weights w are allocated according to the importance or sensitivity of the targets i , the sum of all weights is 1. For each optimization target, a corresponding evaluation function is defined: f1(x) for energy efficiency target, f2(x) for control performance target, f3(x) for stability target, and f4(x) for thermal management target, where x is the H-bridge control parameter vector. f i (x) is processed by normalization to ensure that indicators of different dimensions are comparable when combined. Multiple optimization targets are combined into a multi-objective optimization function F(x)=w1f1(x)+w2f2(x)+w3f3(x)+w4f4(x) through a linear weighted fitting function, where w1, w2, w3, w4 are weight coefficients representing the importance of each target in the final optimization.

[0032] After the cooperative control optimization objective is defined, federated learning is performed on the circuit control operation parameter set according to the cooperative control optimization objective: first, a machine learning model such as a neural network or a gradient boosting tree is used as a basic framework to initialize a global optimization model. For example, for a neural network, the input layer is used to accept static configuration parameters and dynamic operation parameters from different target factories, the multiple parameter data are standardized as inputs, the hidden layer uses several fully connected layers, the complex nonlinear relationship between the input features is captured through the ReLU or Leaky ReLU activation function, and the output layer outputs the corresponding optimized control parameters. In terms of weight initialization, the weights and bias parameters of each layer of the neural network are initialized randomly, such as Xavier or He initialization. If a gradient boosting tree is used, it is initialized as an empty tree or a small tree with a depth of 1, and the number of trees is gradually increased through subsequent local training. The initialization model does not contain any factory original data or node output information during the initialization process, and only serves as a starting point for local model training, ensuring the security of the data privacy of each factory during federated learning and meeting the demand for cross-factory cooperative optimization.

[0033] The built initialization global optimization model is distributed to the plurality of target factories through a secure communication channel, such as an encrypted channel, a VPN, an intranet, etc. Each target factory in the plurality of target factories receives the global optimization model and uses it as the starting point for local training of the local training model based on the circuit control operation parameter set and the collaborative control optimization target score label of the factory itself. During the training process, the weights of the global optimization model are updated using a backpropagation algorithm and an optimizer such as Adam. During each training, the gradient of the loss function is calculated to adjust the parameters of the model, so that the local optimization model can better meet the local production optimization target. Through training, a plurality of local optimization models are obtained, and a plurality of local optimization model corresponding model update parameters are obtained. The model update parameters are internal parameters used to aggregate and update the global model, including model weights, optimizer states, loss functions, gradient information, etc., and reflect the improvements of each factory to the initialization global optimization model during local training. After the plurality of target factories complete the local optimization model training, each factory uploads the model update parameters corresponding to the local optimization model to the global optimization system. The global optimization system uses the plurality of local update parameters to perform multiple rounds of iterative training on the initial global optimization model. Each iteration further optimizes the global model by updating the parameters of the global optimization model. After sufficient iterations, a converged global optimization model and corresponding global optimization control parameters are obtained. The global optimization control parameters include the static configuration parameters and dynamic operation parameters of the intersection identification production node H-bridge. Due to differences in the local production environment of each target factory, such as electrical characteristics, motor load, production speed, or equipment aging, factory adaptation factors are generated based on the actual characteristics of each factory, including electrical specification adjustment coefficients such as voltage and current, load compensation coefficients such as mechanical load or inertia, and function target coefficients such as precision and speed preference. Weighted mapping, constraint optimization mapping, and other methods are used to map and adjust the global optimization control parameters with the adaptation factors of each factory to generate a plurality of local optimization control parameters corresponding to the plurality of target factories. Weighted mapping refers to smoothing and integrating the global parameters and the actual parameters of the factory through factory adaptation weights. Constraint optimization mapping projects the local parameters within the upper and lower limits during the mapping process to ensure the safety of the equipment and the compliance of the production specifications. The plurality of local optimization control parameters refers to the updated H-bridge drive circuit control parameters. The generated local optimization control parameters are applied to the intersection identification production node of the corresponding factory. The intersection identification production node adjusts the behavior of the H-bridge drive circuit based on the local optimization control parameters to achieve optimal control. The collaborative control optimization target is used as the goal of federated learning for all factories, and then the production of the factories is controlled again according to the updated H-bridge drive circuit control parameters to ensure that each factory can operate according to the optimal scheme in actual production, thereby improving production quality.

[0034] The federated learning is performed on the circuit control operation parameter set according to the cooperative control optimization target, the equipment heterogeneity is overcome under the premise of strictly protecting the privacy of each factory data, the flexibility of each factory is ensured while the cooperation between factories is ensured, and the production quality of each factory is optimized in a global framework, the global optimization of the production control parameter is realized, and the production quality of the factory product is improved.

[0035] Further, the initialization global optimization model is iteratively trained by using the corresponding model update parameters, and the method comprises the following steps: updating the model parameters of the initialization global optimization model by using the corresponding model update parameters to obtain a round global optimization model; the round global optimization model is sent to the multiple target factories to obtain multiple one-round local optimization models, the model parameters of the one-round global optimization model are updated according to the one-round model update parameters corresponding to the multiple one-round local optimization models to obtain a two-round global optimization model, and the process is repeated to obtain a global optimization model satisfying a cooperative control optimization target score threshold as a converged global optimization model.

[0036] Specifically, after the global system receives the model update parameters from multiple factories, the parameters of the initialized global optimization model are updated by the multiple update parameters, wherein the parameters of the global optimization model are updated based on the multiple update data by averaging or other fusion methods, to generate a round of global optimization model. Then, the round of global optimization model is sent to the multiple target factories, and each target factory performs local training on the round of global model based on the sent round of global optimization model, combined with the local circuit control operation parameter set and the collaborative control optimization target score label, to generate multiple one-round local optimization models, and obtain the model update parameters corresponding to the multiple one-round local optimization models. The target factories upload the one-round model update parameters corresponding to the multiple one-round local optimization models obtained in the local training to the global system, and the update parameters include the weight, bias, gradient and other information of the local model after local training. After the global system receives the one-round model update parameters corresponding to the one-round local optimization models of all target factories, the current one-round global optimization model is further updated using the one-round model update parameters. This process is repeated for multiple rounds, and the global optimization model is adjusted by the update parameters of the local optimization model in each round. The goal of each round of optimization is to improve the overall performance of the global optimization model in multiple factories by aggregating the local optimization results of each factory. With the iteration, the performance of the global optimization model will continuously improve, gradually reaching the collaborative control optimization target. When the output performance of the global optimization model no longer changes significantly after multiple rounds of iteration, and the evaluated collaborative optimization target has reached the collaborative control optimization target score threshold, it is considered that the global optimization model has converged, the iteration process is ended, and the converged global optimization model and the corresponding global optimization control parameters are obtained. The global optimization model is continuously optimized in multiple rounds of federated learning, and through the alternating training of the initialized global model and the local model, from the initialized global model to the iterative rounds, a converged global optimization model is finally obtained, which meets the multi-objective optimization requirements of the target factories and effectively improves the production quality and efficiency of the entire cross-factory.

[0037] The intersection identified production node is optimized for production quality using the multiple local optimization control parameters.

[0038] Specifically, after each target factory obtains multiple local optimization control parameters, i.e., after obtaining the updated H-bridge driving circuit control parameters, the local optimization control parameters of each factory are matched with the intersection identification production nodes, and the intersection identification production nodes are optimized and adjusted by using multiple local optimization control parameters. The H-bridge driving circuit adjusts the working state of the motor or the actuator according to the optimization parameters, so as to realize accurate action control, speed regulation and power distribution. By optimizing the H-bridge driving circuit control parameters such as PWM frequency, duty ratio, current control gain, temperature rise and power conversion efficiency, the working efficiency and product quality of the production nodes are improved, and energy waste and power consumption are reduced, thereby improving the overall production efficiency and equipment operation stability.

[0039] Based on the global optimization control parameters, the local optimization control parameters are generated, and the intersection identification production nodes are optimized by using the local optimization control parameters, so as to realize the cross-factory collaborative optimization target. Moreover, the bottom-layer common execution unit H-bridge driving circuit which affects the production quality is accurately adjusted by using the local parameters, so as to ensure that each production node of the factory can operate in an optimal state, thereby improving the overall production efficiency, product quality and stability. At the same time, the mechanism of federated learning guarantees the data privacy and effectiveness of cross-factory collaborative optimization, so that the overall production quality is significantly improved.

[0040] Further, after the multiple identification production node groups are identified to obtain the intersection identification production nodes, the method further includes: identifying the multiple identification production node groups to obtain H-bridge electrical property information, H-bridge driving control objects and H-bridge driving functions; performing feature recognition according to the H-bridge electrical property information, the H-bridge driving control objects and the H-bridge driving functions to determine electrical specification feature vectors, mechanical load feature vectors and function target feature vectors; clustering the intersection identification production nodes according to the electrical specification feature vectors, the mechanical load feature vectors and the function target feature vectors to obtain multiple groups of intersection identification production nodes; and constructing multiple federated learning groups according to the multiple groups of intersection identification production nodes.

[0041] Specifically, after the intersection of the plurality of identified production node groups is obtained, the H-bridge electrical property information, H-bridge drive control object, and H-bridge drive function of each identified production node can be identified and extracted through voltage, current, sensor, and H-bridge drive circuit configuration file, control module, or identification tag on the circuit board. The H-bridge electrical property information includes the electrical specifications of the H-bridge, such as rated voltage, current capacity, switch device type, dead time, etc., which determine the basic characteristics and operating range of the H-bridge drive circuit. The H-bridge drive control object refers to the type of load controlled by the H-bridge drive circuit, including but not limited to motors, electromagnets, etc. Each control object has different requirements for the drive mode and control strategy of the H-bridge. The H-bridge drive function refers to the specific function of the H-bridge drive circuit, such as speed regulation, reverse current control, or pulse width modulation signal adjustment, etc. Different drive functions will affect the control strategy and circuit performance.

[0042] Then, using principal component analysis, standardization, and normalization techniques, the H-bridge electrical property information, H-bridge drive control object, and H-bridge drive function are identified and characterized to determine the electrical specification feature vector, mechanical load feature vector, and functional target feature vector. The electrical specification feature vector includes the quantified information of the H-bridge electrical properties, such as rated voltage, current capacity, etc., obtained by standardizing the H-bridge electrical property information and converting it into a feature vector. The mechanical load feature vector includes the characteristics of the control object of the H-bridge drive, such as motor power demand, load inertia, speed, etc., used to determine the working conditions and control requirements of the load. The functional target feature vector contains the quantified information of the H-bridge drive function, such as PWM adjustment frequency, current control gain, etc., representing the target characteristics that the H-bridge drive circuit needs to achieve during operation. According to the electrical specification feature vector, mechanical load feature vector, and functional target feature vector, the rated voltage, current capacity, switch device type, etc. in the electrical specification feature vector are standardized, combined with the load power, speed, and inertia in the mechanical load feature vector, and the PWM adjustment frequency, current loop control gain, etc. in the functional target feature vector, and the equivalent drive index of each H-bridge drive circuit is calculated by weighted synthesis. The calculated equivalent drive index can reflect the comprehensive driving capability of the H-bridge drive circuit under specific load and functional conditions.

[0043] According to the equivalent capacity index, the multiple H-bridge driving circuits are processed in a hierarchical manner. For example, by using a clustering method such as K-means clustering, DBSCAN, etc., the intersection identification production nodes are clustered according to the equivalent driving index, and the intersection identification production nodes are divided into multiple categories. The H-bridge circuits with similar performance are divided into the same driving level, thereby forming multi-level driving circuit information, i.e., multiple groups of intersection identification production nodes, which can reflect the driving capacity difference and coupling relationship between different nodes, and ensure that the H-bridge driving levels of the nodes in each group are comparable and balanced. According to the clustering results, the obtained multiple groups of intersection identification production nodes are used to construct multiple federated learning groups. Each federated learning group in the multiple federated learning groups has similar electrical characteristics and control requirements, and the H-bridge driving levels are comparable and balanced. The intersection identification production nodes in each group share optimization tasks, and the local optimization and cross-group federated learning are used to realize the targeted collaborative optimization of the nodes of different level driving circuits, thereby fully utilizing the global optimization information while ensuring the adaptability and production quality optimization effect of the nodes of different level H-bridge driving.

[0044] Further, according to the multiple groups of intersection identification production nodes, multiple federated learning groups are constructed. The method comprises: collecting multiple circuit control running parameter sets corresponding to the multiple federated learning groups; performing federated learning on the multiple circuit control running parameter sets to obtain multiple global optimization control parameters; and using the multiple global optimization control parameters to respectively perform production quality optimization on the local optimization control parameters of the multiple groups of intersection identification production nodes.

[0045] Specifically, for each federated learning group, a set of circuit control operation parameters is collected, including but not limited to static configuration parameters and dynamic operation parameters of the H-bridge circuit, the H-bridge static configuration parameters including switch device type, rated voltage, current capacity and dead time setting, and the H-bridge dynamic operation parameters including PWM frequency, duty cycle curve, current loop control gain, temperature rise and power conversion efficiency. The set of circuit control operation parameters includes the circuit control characteristics and performance of each production node, and the federated learning is performed according to the set of circuit control operation parameters to ensure that each node can be accurately optimized. In the federated learning process, a global optimization model is first initialized, and then the initialized global optimization model is sent to the plurality of target factories, each factory performs local training according to the local set of circuit control operation parameters and the collaborative optimization target to obtain a plurality of local optimization models and corresponding model update parameters, and the model update parameters are uploaded to the global system. The global system performs multi-round iterative training on the initialized global optimization model according to the uploaded local update parameters to obtain a converged global optimization model and corresponding global optimization control parameters. After obtaining a plurality of global optimization control parameters, the local optimization control parameters of each intersection-identified production node in each federated learning group are optimized for production quality according to the plurality of global optimization control parameters, to ensure that the optimal control effect is achieved in the local range. Through federated learning, cross-factory collaborative optimization is achieved, so that each target factory can perform local optimization based on the global optimization target to improve production quality and efficiency.

[0046] Further, after the intersection-identified production nodes are optimized for production quality using the plurality of local optimization control parameters, the method further includes: continuously monitoring the optimized circuit control operation parameters of the intersection-identified production nodes and the quality optimization increase indicators; determining the optimized circuit control operation parameters greater than a preset quality optimization increase indicator threshold according to the size of the quality optimization increase indicators to obtain identified optimized circuit control operation parameters; and re-performing federated learning optimization according to the identified optimized circuit control operation parameters to obtain updated global optimization control parameters.

[0047] Further, continuously monitoring the optimized circuit control operation parameters of the intersection-identified production nodes and the quality optimization increase indicators includes: recording initial collaborative production result data of the intersection-identified production nodes; collecting corresponding optimized collaborative production result data of the intersection-identified production nodes based on the optimized circuit control operation parameters; comparing the initial collaborative production result data and the optimized collaborative production result data to obtain the quality optimization increase indicators.

[0048] Specifically, before the optimization process starts, the initial production result data of each intersection identification production node is recorded, the initial production result data refers to the performance of the production node without optimization, including production efficiency, quality indicators and equipment state, etc., the production efficiency is, for example, the number of products produced per unit of time, the quality indicators are, for example, product defect rate, rework rate, etc., and the equipment state includes temperature, vibration, load, etc., for reflecting the health state of the H-bridge feedback circuit control equipment. After the H-bridge feedback circuit executes the optimized control strategy optimization circuit control operation parameter, the optimized collaborative production result data of the intersection identification production node under the optimized control parameter is collected, and the optimized collaborative production result data reflects the changes of the production node under the optimized parameter, including production efficiency, quality indicators and equipment state, etc. The initial collaborative production result data and the optimized collaborative production result data are compared and analyzed by difference, the improvement range of each index such as production efficiency and product quality is calculated, and the quality optimization increase index is obtained. According to the size of the quality optimization increase index, the optimized circuit control operation parameter is compared with the preset quality optimization increase index threshold, and the quality optimization increase index threshold is set according to the actual factory demand. When the optimized index exceeds the quality optimization increase index threshold, it indicates that the optimization circuit control operation parameter has a significant effect on the improvement of production quality, and the optimization circuit control operation parameter greater than the preset quality optimization increase index threshold is identified as a high-value parameter. The identified optimization circuit control operation parameter is used to re-optimize the federal learning, that is, the parameter is updated in the global model. By summarizing the local learning results of each factory based on the identified optimization parameter, the updated global optimization control parameter is obtained, so as to further improve the cross-factory collaborative optimization effect and ensure the continuous improvement of the production process and improve the production quality optimization.

[0049] In summary, the cross-factory collaborative production quality optimization method based on federal learning provided in the present application has the following beneficial effects:

[0050] By identifying and intersecting the identified production node groups of multiple target factories, combining H-bridge electrical properties, driven objects and functions for feature extraction and clustering, without sharing original data, using circuit control operation parameter set to develop federated learning, obtaining converged global optimization model and corresponding global control parameters, and then issuing multiple local optimization parameters to realize production quality optimization of intersection nodes. Compared with traditional single factory or centralized optimization method, the sharing of cross-factory data value is realized through the use of federated learning mechanism, taking into account data privacy and collaborative optimization, improving the relevance and accuracy of global optimization parameters. Moreover, by constructing multi-objective optimization functions such as energy efficiency, control performance, stability and thermal management, the performance imbalance caused by single index optimization is avoided, thereby comprehensively improving the quality and reliability of the production process. In addition, through continuous monitoring and quality optimization, the index feedback mechanism is formed, forming a dynamic closed-loop iteration, so that the global optimization model can be updated continuously with the changes of production environment and equipment state, and the long-term effective optimization effect is maintained. In summary, the cross-factory collaborative production quality optimization method based on federated learning overcomes the heterogeneity of equipment, accurately targets the underlying public execution unit H-bridge drive circuit that affects production quality, and collaboratively uses distributed operation data across factories to achieve global optimization of production control parameters, improving production consistency and yield.

[0051] In the second embodiment, based on the same inventive concept as the cross-factory collaborative production quality optimization method based on federated learning in the preceding embodiments, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the steps of the cross-factory collaborative production quality optimization method based on federated learning in any one of the first embodiment.

[0052] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein. Obviously, for those skilled in the art, without departing from the principles of the present application, the present application can be improved and modified, and these improvements and modifications also fall within the scope of protection of the present application.

Claims

1. A cross-factory collaborative production quality optimization method based on federated learning, characterized in that, The method includes: Obtain multiple identification production node groups corresponding to multiple target factories. Here, the identification production node is a production node driven by an H-bridge driver circuit, and the identification production node group is a combination of identification production nodes for each target factory. The multiple identifier production node groups are identified to obtain the intersection identifier production node, and the circuit control operation parameter set of the intersection identifier production node is collected. The circuit control operation parameter set includes H-bridge static configuration parameters and H-bridge dynamic operation parameters. Define a collaborative control optimization objective, perform federated learning on the circuit control operation parameter set according to the collaborative control optimization objective, obtain global optimization control parameters, and generate multiple local optimization control parameters based on the multiple target factories according to the global optimization control parameters; The production quality of the intersection identifier production node is optimized using the multiple local optimization control parameters. After identifying the intersection of the multiple identifier production node groups, the method further includes: The multiple identification production node groups are identified to obtain H-bridge electrical attribute information, H-bridge drive control objects, and H-bridge drive functions; Based on the H-bridge electrical attribute information, H-bridge drive control object and H-bridge drive function, feature identification is performed to determine the electrical specification feature vector, mechanical load feature vector and functional target feature vector. Based on the electrical specification feature vector, mechanical load feature vector, and functional target feature vector, the intersection identifier production nodes are clustered to obtain multiple sets of intersection identifier production nodes. Multiple federated learning groups are constructed based on the production nodes of the multiple sets of intersection identifiers; The static configuration parameters of the H-bridge include the switching device type, rated voltage, current capacity, and dead time setting. The dynamic operating parameters of the H-bridge include PWM frequency, duty cycle curve, current loop control gain, temperature rise, and power conversion efficiency.

2. The cross-factory collaborative production quality optimization method based on federated learning as described in claim 1, characterized in that, Multiple federated learning groups are constructed based on the intersection identifiers of the production nodes, and the method includes: Collect multiple sets of circuit control operation parameters corresponding to the multiple federated learning groups; Federated learning is performed on the multiple sets of circuit control operation parameters to obtain multiple global optimization control parameters. The multiple global optimization control parameters are then used to optimize the production quality of the local optimization control parameters of the multiple sets of intersection identifier production nodes.

3. The cross-factory collaborative production quality optimization method based on federated learning as described in claim 1, characterized in that, Federated learning is performed on the circuit control operating parameter set according to the aforementioned collaborative control optimization objective to obtain global optimization control parameters. The method includes: Initialize the global optimization model; The initial global optimization model is distributed to the multiple target factories. The multiple target factories train the initial global optimization model according to the corresponding circuit control operation parameter set and the collaborative control optimization target score label to obtain multiple local optimization models. Upload the model update parameters corresponding to the multiple local optimization models, and use the corresponding model update parameters to perform multiple rounds of iterative training on the initialized global optimization model to obtain a converged global optimization model and the corresponding global optimization control parameters.

4. The cross-factory collaborative production quality optimization method based on federated learning as described in claim 3, characterized in that, The method involves iteratively training the initial global optimization model using the corresponding model update parameters, including: The model parameters of the initial global optimization model are updated using the corresponding model update parameters to obtain a round of global optimization model; The first-round global optimization model is distributed to the multiple target factories to obtain multiple first-round local optimization models. The model parameters of the first-round global optimization model are updated according to the first-round model update parameters corresponding to the multiple first-round local optimization models to obtain the second-round global optimization model. This process is repeated until a global optimization model that satisfies the collaborative control optimization objective scoring threshold is obtained as the converged global optimization model.

5. The cross-factory collaborative production quality optimization method based on federated learning as described in claim 3, characterized in that, The cooperative control optimization objective is a multi-objective optimization function, which is a linear weighted fitting function of multiple optimization objectives; The optimization objectives include energy efficiency objectives, control performance objectives, stability objectives, and thermal management objectives.

6. The cross-factory collaborative production quality optimization method based on federated learning as described in claim 1, characterized in that, After optimizing the production quality of the intersection identifier production node using the multiple local optimization control parameters, the method further includes: Continuously monitor the optimized circuit control operating parameters and quality optimization indicators of the intersection identifier production node; Based on the magnitude of the quality optimization increase index, the optimized circuit control operation parameters that are greater than the preset quality optimization increase index threshold are identified and value-marked to obtain the marked optimized circuit control operation parameters. Based on the identified optimized circuit control operating parameters, federated learning optimization is performed again to obtain updated global optimized control parameters.

7. The cross-factory collaborative production quality optimization method based on federated learning as described in claim 6, characterized in that, The method for continuously monitoring the optimized circuit control operating parameters and quality optimization improvement indicators of the intersection identifier production node includes: Record the initial collaborative production result data of the intersection identifier production nodes; Collect the optimized collaborative production result data corresponding to the intersection identifier production nodes under the optimized circuit control operating parameters; By comparing the initial collaborative production result data and the optimized collaborative production result data, a quality optimization increase index is obtained.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the cross-plant collaborative production quality optimization method based on federated learning as described in any one of claims 1 to 7.

Images